Transcript for:
AI Project Management and Use Cases

e so while we are waiting if you have any thoughts questions you know you can just ask if not we can wait for a few minutes before know everybody joins in we can start e so let's wait e how has the weather been in general in your places guys has it been Hot all over place or has it been rainy cool it's very hot muggy same thing in Bangalore even though we had a slight showers a couple of days you know like yesterday and day before yesterday it's still very hot in Bangalore yeah even pun it is too H it's like everywhere it is getting H exactly oh see Samuel has good news he says heavy rain here in East Africa very good Le somebody's getting rain hi professor good evening good morning good evening all yeah hey guys hi professor good morning so good at least somebody's getting rain I'm happy yeah it's it's very hot I heard in all over India it's extremely hot extremely hot even in Bangalore know it's been like you know crazy the whole month of April generally we get rains not a single drop of rain in Bangalore whole month of April yeah it's it's B we just waiting guys couple of more minutes you know and then we can start very good evening good morning all of you morning and evening let me check is somebody in the waiting room no what was the thing uh yes Seline says hi professor may have a bit of R to clarify the group assignment definitely I spoke about it last time Selen I will repeat it this time absolutely and in fact I'm planning to do that so what I can do is you know let the first hour be over I will do it after the first hour but yes I will definitely yeah two more minutes and then we can start [Music] okay let me start sharing my screen okay okay [Music] okay okay chat let me just quickly search if there is anybody waiting in the nobody's waiting so I think we are good we can start guys I think you know it's 6:34 perfect we can start so I think you know so first of all again good morning evening or afternoon depending upon you know uh where you are on uh so I will uh uh basically talk about to what I'm basically going to talk about today and this is very important that I talk about today is guys what I thought was you know two things I'll go through today one thing is spend a little bit more time on the project management piece of it of AI projects and give you some examples and we'll talk about it and everything so basically today I'm going to spend a little bit more time and you know this is something that we are going to do is basically spend a little bit more time because as leaders in know this is something that we've talked about it I just want to emphasize a few things and I think it'll be good so let's spend a little bit time talking about it and I will also then walk you through some real examples of you know thing of it but before I start any questions or comments from the last class or anything okay be and second thing is you know uh I I will talk about the group assignment once again I will do that at after the first break so that you if somebody's a little late they can definitely also be part of that know thing okay uh with that let us start with you know and one of the things I've taken today is you know not that you know I've taken I thought let's take an area not that why I pick this area because I've done a few projects I can pick a is basically in general not just bfsi but all projects AI projects or in general projects right have three go goals reduction of cost growth in revenue and increase in customer satisfaction each one of it let's talk about so when you obviously write one of the important things that you know when you do an Roi on any project including an AI project you need to know where is this AI going to or this project project is going to be important very important is that part of it irrespective of you know if you're a small business owner a large business owner doesn't make a difference if you're implementing an AI project and you know in general you need to know that whether it is satisfying one or more of these objectives in some cases it May subject you know it may be more than one the three areas where you know generally this is going to happen and this is very important first thing is reduction in cost so basically what it means to say is is um what does reduction in cost means to say is most probably you are impacting the operating pieces of the of your company or of the process or whichever the business problem basically when you in reduce the cost what happens your profit increases right your bottom line increases everything being the same if you reduce your C cost your bottom line increases your profit increases right so the first thing you have to say is where does it happen so first thing it can happen is reduce your cost what do I mean by that reduce your cost means to say you may now be doing something more efficiently what do I mean by that so for example you may now say you know what hey they were actually you know it was taking you know uh two hours to do this now the same thing is being done let's say in an hour it's a combination of human and but AI is helping them to do the fast process faster so that I can process information faster so for example it might have taken six days to approve a loan now because of AI human and AI content it only takes me four days to know offer basically approve a loan which is good because that means you say technically without doing anything in a given month I can know basically increase the number of loans that I can approve so reduction in cost is a very important thing or it could be you know the process what should I say time taken or it could be you know the number of steps involved in the process it could have been you know that it had basically it had required six human interventions Now it only requires three human inter interventions because the other three human interventions could be done by AI anything like that right so or it could be for example you know let's say the basic x-ray is read by a doctor by an AI machine so the doctor can look at it but I spend a little bit time approve yeah yeah this looks good right and then so he or she may be able to see a little bit more six you know 10% more patients because they didn't have to spend a lot of time you know doing the review of the documents or you know things like that or maybe you know checking whether you know all your documents are in place in terms when you turn on the loan right when you turning the loan or any process if you want to start making sure all the documents are in place or for that matter I gave you I show you an example where you know the one of the agents said you know hey whether to I have six documents whether to check whether they are in compliant with our company policies earlier it has to somebody had to go read check now maybe you know generate or a can be able to do it so reduction in cost is a very very important thing oh sorry somebody said what is oh yo I'm apologist BFF sorry I didn't even see the question apologies banking finance and financial services and insurance so b stands for banking FS stands for financial services I stands for insurance I didn't see your comment but that's what it is sorry yes so induction cost the second thing is it may be help you to increase uh Revenue how it may be help you to launch new services or it may help you to you know uh basically uh do certain things you know which can be used to do so it can be either help you to grow in revenue or obviously it can be done and the third one is uh which is also very useful because growth in Revenue means you're increasing the Top Line not the bottom line Top Line and then the next one is increase in customer satisfaction yes because it may be able to detect uh what should I say problems earlier or in Netflix if the recommendation engine is really working well and it starts recommending really good movies that really to your taste obviously you'll be very happy with Netflix right wow Netflix is recommending movies that I have I've never thought about so things like that so there could be a recommendation Engine That Could be helping well which increases your customer satisfaction or it can basically be able to look at the trends much earlier and say hey there is you know little bit of a problem in terms of you know the feedback I'm getting or you know things like that it will be able to help in that so there's a lot of ways where you know it will be able to do it so once you do a project very very important from the technology side you have to do it but also from the objective side uh you know three things that you need to look at reduction in cost growth in revenue or increase in customer satisfaction any one of this is basically most of your projects either will be one or two or all three yeah okay but you know one or two or not it could be definitely you know um could beat your you know this one thoughts comments um you can also have Regulatory Compliance exactly you could also have uh Regulatory Compliance Regulatory Compliance sence is so Sony excellent question you bring up I just want to understand in what way are you telling that will uh will okay Regulatory Compliance is a mandatory thing right that will not make revenue for the company but it could have financial implications you could lose your license exactly so in that sense I agree with you so basically Al if you look at it that way you see are you you know basically Regulatory Compliance could be you know uh making sure that you know you are following you know you're being you're following all the rules you're being a good citizen which means to say that you know yes absolutely anything else any other thoughts or comments similarly can you think about any projects you know for example let's take reduction in cost can you give me examples in I just did bfsi some project can you give me I don't know is there anybody here from uh I don't know supply chain background can somebody body give me some examples from supply chain I'm just randomly picking supply chain or manufacturing or any other area that you can think where you know AI can help in ruction of cost I I would I would talk about the give example of supply chain especially when it comes to so many middle quote unquote middle men AI can easily take away that which would significantly reduce cost so I'm in an area where we do import uh Global products to Canada so there are portions of it where you could see if this is somehow AI or automated in a way it significantly will reduce the cost absolutely good point in fact you know uh very good point um so any other thoughts growth in Revenue can somebody think about any new things that can help in growth in Revenue so reduction in cost is basically have the Rippling effect on growth in Revenue because we are reducing the waste which is going basically if you're reducing the cost anyway uh in you increasing the growth potential right the revenues in yes absolutely you're right additional additional something growth potential projects right basically if you initiate some growth potential project M that will augment the growth in Revenue right by along with the reduc reduction in the cost absolutely um cross selling you can recommend uh Sim cross selling so that's a very good uh area where you know basically I was about especially you know if you are again let's say you have insurance you know the very simple example is Insurance you know uh car for example you know you have checking and you say basic accounting insurance and other products right you can mgage you can do cross- selling absolutely um predictive maintenance Muhammad says absolutely Muhammad that is very important in ruction and cost very very good actually Quality Inspection predictive maintenance is a very big deal in production and cost right because you want to make sure that you replace things at the right see see one of the things very important in production inance and it is a very very very critical thing is today predictive maintenance happens in a lot lot of places by sort of you know by what I used say is by calendar what do I mean by that every three months they replace seven Parts why because they know that it's easier to replace a $20 filter rather than risk it predictive maintenance nowadays are going into it and saying you know don't increase it by every 3 months probably you could do it by six months or something so that obviously will help absolutely um so absolutely and it can also help to sort of you know create new avenues and I'll give you an example right recently I was reading this and I I'm trying to remember the name of the company if I can't remember it by the what should I say by in the break I will look into it it's an Indian company a startup and it's making a lot of money see in India one of the things is happening and this is in the retail space and let me give you this example so that you know you guys can what and again as Leaders you need to understand this you know and again it's obviously it's an e-commerce space obviously if you're not in see today what was happening was if you look at uh see in India we have shopping malls the big stores but majority of the businesses if you look at it right are small businesses especially in tier 2 and tier three towns 99% of the businesses are small businesses now what used to happen is let's say and especially if you go into tier three and tier four towns the entire what should I say City I'm using the word place I I can't use the word City because technically it's not a city it could be a small know Village or it could be a village plus you know area there could be one or two stores that's it now what happens is these stores are General stores right they do sell in know groceries plus they also sell other things so generally a general store when I use the word general what I mean to say is they sell multiple things in a small you know uh shop right so now when they have to replenish their stock they it is a major problem even in India today what happens if you go to the big stores right here let's say even even if you a small business in a city what happens let's say you are running out of certain things what do you do you automatically call the wholesaler and the wholesaler basically will what they do they will okay you give them an order and they just come and deliver it to you the reason why they do that is because in addition to your order they may have five or six orders in the same area or in the same ZIP code or in the same thing so it works out for them because you know they're delivering it to five or six places they customer they basically Club it and so it works out for them because the cost of delivery and everything but if you go to tier two and tier three towns it's impossible because they have to drive so long in order to just deliver it to one store or maximum they be able to deliver it to two stores and that becomes a major major issue and so what used to happen and even it happens today is basically um the shop owner had to until and unless right he has a family and his a wife and his kids but a lot of times what used to happen is he would be alone or something he had to shut down the store for three days or at least a day or two go to the city order it basically pick up the stuff hire use another track or something they pay and come back and it was a really really a big problem the reason I'm bringing that you said about know middleman so I just talking about that and I remember this the moment you use that word now what has happened in the cities right there are two companies one company is called uh uh Gio the second company is called I can't arban something I can't remember the name there's two companies now what they are doing is at least for the city tier one and a little bit of tier two they are cutting down the middleman what they are saying is guys listen you don't need to what should I say go to the middleman in order for you to get it basically even though they call them whole I will give it to you at the near cost price at what I get it from the manufacturer because I'm buying it in such bulk problem basically I will give it to you at a very little margin if to you so that your cost becomes a lot less and so obviously selling price is the MRP in India what happens is every most products has an MRP which is what you know maximum retail price that's the price they sell it at so you can sell it at that but your cost of acquisition becomes a lot cheaper then what you to go to a wholesaler because I am basically buying at a lot so now that what should I say phenomenon is happening but again it's happening in tier one and Tier Two Cities slightly in Tier Two Cities but tier one most of them are ordering this this so if you are for example I don't know how many of you know this for example but in us that even in India we have the same model but in us it's a lot bigger called Costco right or a Sams but mostly Costco Costco I think I don't know how many of you know their margins is 2% that's it they don't take a profit more than 2% but this but they sell everything bulk but they sell so much that you know they can make it up so these guys are aling I'm going to buy in Pulp and you know instead of 2% you know I'm going to make it very little for you so that was that's happening but still if you look at it 90% of the stores or even 85% of the stores are in tier three tier tier two tier three tier four towns in India they couldn't figure it out what to do so this company this startup and I can I forgot the name in the in the break I will basically what they did was how do I solve this problem right so they solve this problem through AI technology and some really innovativeness what they did was they said okay they went to a lot of these stores if you look at a lot of these stores there some especially tier two tier three there is a lot of them have space available in their stores so what these guys have done is is they have identified in every region enough stores where instead of me storing it in a central warehouse and then trying to ship it out it makes no sense now the distributor it into a few this what should I say they think into they've gone nearer basically Edge Computing kind of a thing they've gone nearer and nearer to the stores and they've stored it there and from there what they have done they're using the local what should I say transport that are easily available there on a large scale in order to deliver it and they are just you know it's one of the fastest growing should I say startups in India I can't remember it name elastic or I I I I will I will basically let you know the name guys and I was amazed that you know basically when you know you said I'm cutting out the middleman this is a great you know one of the really really smart startup ideas that are happening in India so it's very important that you understand that you know there are a lot of ways you can you know play and and AI is basically tells you where to stay what is the demand they're doing all kinds of Ni analytics question questions comments in fact if I have that video right I I will just play to you in the break if I find that and I was reading about it and then I went and saw that video and I was very impressed very nice idea okay can't remember the name anyway this I will come to you guys this is my two slides on my course and presentation I will do it after the first break so I'll give it to me so now in order for you to start an AI project in your company so I'm talking about two levels see some companies have already been doing a lot of AI so then means to say you know you're already in in the higher level of this 1 2 3 4 5 six but I just wanted to sort of give you that uh uh what should I say formula General formula I'm not saying this the only formula but a formula or a framework first thing you have to basically make sure in any company if you are the startup company that's a different thing but if you a medium siiz big siiz companies you see all of these things you know they need first of all the leadership should be trained in analytics what do I mean by that they need to know understand what AI can do for them very very important and you know it's a very very critical thing that they understand what AI can do for them lot of this when we try to do corporate training right what unfortunately happens this and this is something and you may guys tell me no it's not true or it is true the when we go to a corporate training they say oh just train the engineers and they will do their that's a very very yes we have to train the engineers I'm not denying we have to but if the leadership is not comfortable and understanding what AI can do for them training the engineers will not help them to be an AI successful company and this is something guys in your own companies when I use the word leadership here doesn't mean it always has to be the CEO leadership at a level who can actually make decisions it could be at the director level it could be at the GM level it could be at the VP level whatever that level is right that is a very important thing I don't mean what should I say cxo or CEO level absolutely it's good that the CEO also understands but some it most of the time it is better it comes stop down but a lot of these companies I feel they try to do it bottom up and again I'm open for suggestions comments you know thoughts I I completely agree with you Professor on this based on my own experience with like leadership team does not have that vision of what AI can do um right and like training or like the junior members are getting training on AI that's basically they enrolled in some courses right Google provided courses and all right so but the leadership do not have a vision how that theyi can transform their industry I'm I'm from Banking and finance industry itself like do bank so I there are cases I mean so it's depends upon the business units as well like which business unit you are in right and and accordingly like the use case will uh if that use case fit into that business unit probably that will transform the the revenue of the business unit but then it's always a challenge to make leadership understand what is what is the gain in pain of absolutely very good excellent point because it's very very important that they understand because they the reason why I wanted I'm stressing this is because even if there's a failure right they need to understand why it failed or what is it that it didn't achieve it potential that's okay right if not you know they may just dismiss it and say you know what this is not working and that is something so that is the reason why absolutely right uh you know uh Deepak you also had a very good point absolutely and that is something that you know so I always put it at the bottom so I said number one right most people when they go say what model are using what building we will get to that I'm not denying but first thing is you know get the understanding of you know so when you guys you know if you go into leadership position that's the most important thing you need to what AI can do for you obviously if it's a small business you the startup that's a different question but if you're working for any medium or big siiz companies very important that the leadership gets it the second thing is know identification of the problem s again I've spoken enough about it but I just want to stress it again you all again this is something the leadership and I mean the business heads the implementation you know what I said right managers or The Innovation managers or the imple they are the people who have identify the business and in order for them to do that they also have to be educated on the AIP if they are not educated on the AIP of it it's going to be very difficult for them to see and they have to be with you throughout the process so that is the next thing so these two are the most important parts before you start the AI journey in your company you a lot of your companies maybe this is already there so what I'm trying to get here is if it's already there fantastic if it's not there I would say if you want to start it it's not easy but this is the framework for a success I'm not saying this is the only framework but a framework once that you have that right the third thing is basically plan to do a pilot project because you see very important that you understand and in the pilot project what are the things that you need to understand first of all think small when you're doing your first or your second AI cases do not think you can you know boil the ocean or whatever the terminology is right be very very very you know sort of you know uh what should I say conservative in your first project and make sure that you know you when you do the pilot make sure that you are addressing not whether you know uh is the data that you're going will you able to get the right data will you be able to basically you know convince the people whom they're you know basically using it that it's not think about small one of these things and this is what you need to think about don't think how do I clean the data don't think those are all things that are important I'm not denying it but they can happen the things that you really need to worry about is you know do I have the right data do will I get you know uh what should I say uh uh the right what should I say access to the what should I say the people who will be using it so that I can get the feedback third thing is you know making sure that you know the even though that is going to be a challenge the accuracy of the models things like that you need to sort of know worry about it and say so pick projects where you know you're pretty sure that you know and work with the data scientists or somebody to make sure that you know the success is pretty there the budget part budget part someone has like budget is missing so I think that's the I think in my absolutely so when I say Workshop of the business heads on identification of the problem you're absolutely right that he has he has raised it and the bu part I I should I can add you're right I could have added another bullet point but I thought workshop on the identification of the problem in and the pilot project sort of the budget is sort of taken care of here but you're absolutely right but you're absolutely right budget is important and when you bring up a good point the reason I I say this is and maybe your some companies it will not be an issue I raise it is because people say that hey let's try it out and even if it becomes the success people don't have the budget oh okay fine I will think about it in my next budget yeah absolutely you're right and then this will be like you know September or July of this year and they say when is your financial cycle oh it starts in April 1 of next year what does that mean so I understand that budget is not an easy thing but you need to so that's the reason why if the leadership is behind it they will find the budget yeah and probably this identification of problem how how much we can sell it to the leadership that will attract the budget okay if they convinced okay the problem is really really make is a groundbreaking problem that we are solving by AI then it become I mean there they will be to sanction The Bu but same yeah absolutely but very difficult but at the same time be very careful when you pick your first project right so you need maybe have to work with your internal you know teams or somebody and find a budget with your manager or a senior manager or a director make one or two project have success maybe show that successes to the leadership team and educate them before you embark on a serious problem so these are things it just nothing to do with a it has to do something to do with obviously but also project management yeah it's project management and change management together absolutely project management and change management and you're absolutely right so these are the kinds of things that are you know very important guys because I see a lot of you know everybody blames it on technology and I'm sure you know AI is also I'm not saying it's perfect I will never say that but there's a lot of things you can do to make the transition smoother the next thing is once you have that and let's say the pilot is good and that project either you know you expand that project or you know come up like that you do a pilot for a small thing you see success and say hey let's go big the next most important thing is the in-house analytics infrastructure either do you have it inhouse or do you basically when I use the word inhouse it could be you know what I mean to say is either you do it inhouse in the sense you know with your storage or or you may directly say I will do it in the cloud because for us you know cloud is our in-house infrastructure again depends on the size of the project scale of the uh this one what should I say uh uh size of the project size of the data and all those things but that is also plays a very important SU infrastructure is also very important because you need to secure the right infrastructure but all of these things killing the delivery team and model development 456 depends upon very very important as you let's say for example you are driving this project from let's say you are the implementation lead or implementation manager or whatever we call that right implementation lead or the Implement in the previous classes you need to understand who eventually will be using it so you need to understand when I use the word infrastructure you need to work with the the the what should I say the data engineering team not that you will be responsible for Designing the infrastructure but the needs you need to make sure and within your brain you should have there are two needs for me the first need is the training need when I train something when I create something the second need is when I deploy something what are my needs there so you need to have both of this very very very sure the training needs and the deploying needs and sometimes they could be very different I've given enough examples right and just to repeat myself so that they're all on the same page so for example if you take a bfsi example right for example a loan prediction or something when I train I train it with a lot of data when I predict I may be predicting one what should I say application at a time so I may not need so much of resources at the prediction time whereas I am doing some recommendation engines or something like that then I may need both training and my what should I say uh deployment are you know very very high infrastructure so you when I use the word infrastructure I mean both sides of it the training and the deployment part of it so you need to understand and think about those things very very critical second you also need to think about is you know what is it that you know what are the requirements who will be using this project and what are the requirements of the people using this project in the sense basically how quickly they want it what is how what aspect they want it how do they want it what are the formats they want it what are my what it dashboarding that I need to do is there any graph I need to create what how soon do they want it is it something that internal use or is this public facing when I use public facing what I mean to say is the consumers or the general public will be using it so that is very very important when you talk about the and based on that you need to have the Skilling the delivery team when I use the word Skilling the delivery team again a few things which I've already said I wanted to repeat it because it is a very important is one not only the technical skill set but also you need champions for the project if by chance people are consuming that this project is going to have impact on their jobs impact on the whatever you need internal Champions to basically talk to the user group and say Hey listen this is going to be there or if you are basically consumer facing then you really really need the people who basically are going to be dealing with the consumers or dealing with them to understand what is that that you need to do so very very thing only after this finally does the you know model develop building and deployment all this data cleaning and all that stuff comes here I've just clubbed all of it into one p so as you can see right a lot of things has to happen even before you can even imagine that I need to do at the development model development and everything and this is a very important you know project design aspect of any project that you need to spend a lot of time Deepak you had a comment that says uh basically what you're saying is that uh CI I need a CIO generally for mid-size and large siiz companies absolutely so if the role of CIO is not there absolutely no no I agree with you cios are important and absolutely roles are not there but these are the kinds of things that guys I I told I I've basically talked about it earlier but I'm also talking about it now the reason I'm spending so much of time and talking about it is very critical to follow these steps questions comments okay so the next one I'm talking about is okay and I have the report where I got this data from it is in this is the report guys you see here right I have the report where I got all this data from so uh and let's talk about the top three and then we'll talk about the bottom three and we will you know uh we will talk about it because it's important the first one is it said this report which I have you know I have it here if you want to you can always you know take that basically says 88% of the organization stud reported gaps in the practices for AI projects what are these gaps gaps could be you know hey ien ification of the right problem gaps could be you know that you know um when the project is started lot of people don't think about the deployment final deployments anything right gaps could be so 88% is not a small number right which is a huge number and you can say Okay 88% of what and I think the number of organizations it studied was probably like a I don't remember the number but it was somewhere in the 800 100 or 900 companies not small but 900 companies is not bad right are these the companies that they already um undertaking some AI projects exactly they are taking undertaking some AI projects absolutely good I wonder how many companies are not doing anything see that is something very difficult to get right nobody wants to say I'm not doing anything in AI right you bring up a very good point but nobody in today's if you ask anybody even if they're not doing it nobody would say I'm not doing anything in here right they say absolutely we are studying it we are looking into it or we are planning to do some pilot projects you know what I mean to say absolutely yes we we going through the same thing now we we had a discussion with our exacts I ask a few questions what are we doing about AI um how are we going to integrate it and they we're working on it uh yeah exactly what standard answer right yeah we we started with the AI policy development okay what is it don't use chat GPT okay yeah I I totally agree with that so whenever so that's that's a very uh consistent feedback from leadership team especially for financial institutions deployment and implementation of AI it's one thing to do the research but when it comes to deployment and implementation there's there's always a big roadblock yeah and and sometimes to be fair right and I'm sure you know that and I'm sure you see I think lot of you I'm sure there's a few quite a few of you here in the financial sector one thing they're afraid of is in the financial sector and there are certain sectors they're afraid of and I'll tell you why also I'm sure you can they're so scared because if they something goes wrong the regulatory brings their big hammer on them they are so risk aive e-commerce companies are not so risk aversive because even if they don't see even if the recommendations goes wrong right okay I made the wrong recommendation people are not going to bring the hammer on you right whereas in the financial companies for reason you know first of all Financial companies they are very conservative and traditional in their approaches so they're very scared because the Regulatory and the trust that they have a financial company needs to have in order to attract Finance or money or whatever that money is is so high that even slight blemish into that they're so scared about it right just like you know for example even a small rumor that says you know this bank is in trouble people just rush to the bank to withdraw their money that rumor could be false but even a small rumor could really ruin a financial institution yeah credit F and obviously there are so many examp absolutely credit spe and I don't know how many of you know the story I work for them huh I work for them oh okay I've liveed through it yeah so yeah there are multiple instances here right so there's so much of you know but I agree with you that doesn't mean they should not see one of the things that at least see one of the things which I am also afraid of is you know one of the things is you know as you guys rightfully said they are not even ready to play with it in their own companies forget it you know whether they put it out or not let them at least start playing with it and they're not even lot of these company not even ready to do that so anyway so that is something you know very very very critical in terms of you know uh organization the second one is you know 21% of the total wastage in AI project can be recovered with effective project management purposes absolutely right because again lot of things for example and I'll use similar examples you know if you don't tell the uh developer or the developer that you know it has to go into production and the last time you tell him obviously the project will be delayed because you have to do a certain thing if you do not tell uh the project the AI whoever the project that output has to be within six seconds and at the end if you say I needed it in six seconds so there's a lot of these wastages in terms of that and also sometimes what happens is Project scope creep which is a very common thing or gold you know shifting small things happens I'm not denying it so all of this you know basically comes to this wastage the third part what is happening is a lot of these companies and I will talk about it in the later on so even though they use the standard you know what should I say agile or you know this they are now finding that you know and I'll tell you why if I come to the next slide they're sort of mixing a couple of these things in order to create their own methodology specifically for DS and data DS is data science AI is AI artificial intelligence project because standard practices sometimes do not take into account certain things and what are those certain things I'll talk about it next time so they have sort of developing their own practices or developing their own framework and the reason reason why all of this is happening is there are three sort of you know what should I say challenges emerged the first challenge was there's a limited effectiveness of traditional project management practices when applied directly to that is true right so what are they saying the saying is you know if you look at a regular project management practices right there are a certain things that you do not sort of you know it's no more you know I just manage the project in a standard way where you know I give it a bunch of uh uh outputs and I expect a bunch of inputs and I say I'm assuming that you know a general software project because if you look at a traditional software project the sort of the goal is fixed I need to do this this is the goal and you try to work towards that goal whereas in AIML project that's not the case right and I next slide a little bit I'll talk about it the need for experimentation is extremely high that is one of the major things in it and as I showed you right at the beginning there is no way you know what is going to work what is not going to work even whether the data you have is right it's not enough do I need to create new features do I don't need to create new features do I have to do something all those there lot of uncertainities third defining and measuring success is difficult as setting kpis and pegging them to a business value depends on availability of data Model Behavior and other Factor exactly you cannot say I need 90% accuracy you may not be able to achieve 90% accuracy you may only be able to achieve 30% accuracy but it may be as I told earlier right it may be better than the 10% accuracy you're having now so basically saying you know I want this accuracy only it's very difficult in an a project which all relative and all relative in terms of you know what you want to achieve is this like a one- term exercise or you want to Evol it even though sometimes it may be equivalent to what a human is doing or slightly less than what a human is doing if you see the potential and you say know what with more data with more thing I can improve it over the years with over the months or you know that kind of you know sort of flexibility you need to have and also you need that you know so it's very very difficult the way you define an AI project lot of these project management folks and I don't know are generally used to this strict boundaries so it's very important that they have been trained in sort of trained as in you know understand how AI works and that's very important even not just at the leadership level but also at gut level questions comments okay so as you said here right 76% of the organization used a custom methodology what do I mean by custom methodology they have taken few things from a few things first of all I think I don't know how many of you know this but I'm going to spend a little bit time explaining to you crisp DM agile and waterfall let me do first Gile and waterfall and then do you know DM stands for data management so basically it is of data management and things like that uh agile and waterfall um does every everybody know what they are or do you want me to spend some time explaining this be honest it's okay if you don't know don't I I'll spend I'm not an expert on it but at least I can give you an overview on that how many of you want me to okay let me spend a minute then on each one of it right so that you know I'll take it from right to left there is a reason why I'm doing it so from waterfall agile and then crisp DM now waterfall the reason why is called waterfall is because if you look at any waterfalls right look at any general waterfalls what happens water starts flowing and suddenly all of the water falls into the next level then what may happen it may water May flow flow flow then there may be a drop again all the water falls to the next drop then it may flow flow it drops to the next wall right so basically that is what why it is called if you look at if you look at earlier software development what used to happen they they ised to develop the entire project or at least significant amount of the project and then dump it to QA QA what they did basically did their QA whatever it was you know they gave them a feedback until the feedback was given the software developers just waited the time you know quiding that come saying it's in QA let them give us feedback once the feedback was given whatever then from the what should I say software Q&A it could have gone into what productionize team waited and know and they did their own test they did their own changes they had to productionize the product at that point if they find anything they said hey we have an issue do you think hey this is not working they said oh no no QA tested it again the QA will test it if not again it so it was a very very very I take something give it to the next person take the ball give it to the next person take the ball give it to the next person and when the one person was working everybody else was sort of piddling with them that is what is called waterfall model and that worked successfully I'm not saying you in some project it makes sense it worked very successfully so that is the reason why you have the word waterfall is because all of it you do you give it to the next you tump it to the next person where the waterfalls all of the waterfalls to the next level in this case the next functionality waterfall only works well if the requirements are 100% capture up front exactly and c not absolutely so one of the challenges and absolutely Sony thank you and about exactly these things work well whenever Ren the requirements so earlier what used to happen it's nearly impossible right very few times you get everything right at the first time right probably you know very few times so whenever if by chance you're doing it and let's say in Q suddenly you find out know hey there's a new requirement what used to happen again the QA testing used to I'm sorry the development used to happen make all the changes again dump again dump so or if it is not important what they used to say oh don't worry about it phase two or version 2.0 we'll put it in the next version the most common uh answer you used to get from project management project managers where because they were responsible for the end to endend right away they used to say oh not critical dump it into 2.0 right that was the standard answer you got right I'm sure a lot of you were invol were involved in the waterfall models works well only I think for migration project where we know where we what the target state looks like so but for a long time it did work people used to do that right I I don't think it necessarily worked well just that people did a lot of it and quite often times if you didn't get if you got to capture the requirements fine but then the business may have changed especially if the project was a long running project you know so know businesses are not static as well and then what you deliver is not what uh is useful absolutely so if by chance something changed or so anyway that was waterfall model now let come to the next model which is agile the word obviously right agile the word itself in English forget all of it agile means quick right the word agile in English means quick so basically they said okay uh why do I have [Music] to do this let me now split the projects into basically into multiple small pieces or sections or whatever you want to call and I will say I'm going to deliver this functionality in two weeks next I'm going to deliver this functionality in two weeks I'm going to deliver this functionality in two weeks I'm going to deliver this functionality in two weeks so basically instead of me delivering everything I'm going to deliver functionality by functionality now there's an advantage to it what is the advantage as everybody was saying right even if the what should I say uh something changes I am not going to wait wait six months before I am delivering it I can be agile I can be a little bit more you know uh quick in order to change it because I may have delivered only one functionality maybe I now I'll change the second I will work on it or even if you make want to make changes I can do it because every two weeks I'm de you know dropping a load or a code uh what should I say code right or whatever right I'm developing it the problem there is one of the problems where in aile a lot of people say somebody had to had a complete Vision because even though they're developing to the testing you know they have to make sure that you know they don't just test the the new load but it's again this is part of the bigger what should I pieces and you have to make sure that you know you they develop it correctly making sure that you know whatever they've developed has been you know backward compati there a lot of things that you need to start understanding and giving them feedback at the good news with waterfall model I had the entire code so there was no more new development now I can test back compatibility forward compatibility everything and check there's a problem here I'll have to make sure that I sort of you know making sure even because I'm getting pieces of it I'm not seeing the full picture the idea of a is you go with a minimal product and and then you basically iterate exactly so it is something that you need to be careful with the third one is basically was sort of done for what should I say a data science project it was called crisp DM now in crisp DM basically it involves you know collection of data understanding of the business and everything one of the things that it really doesn't take into account at a very high level crisp DM is the the pilot the understanding of you know the accuracy defining of the accuracy defining Al there certain things it doesn't really really Define it so what people are doing is basically are saying okay all of it has it some good things and some bad things why don't I combine all of this and make them into so basically they're taking pieces of each one of them and creating their own Frameworks and that's what happening so questions comments on that okay it's 7:30 do you want to take a 10 minute break now or do you want to wait for 10 more minutes for this slide to be over and then take a 10-minute break either way I'm fine so let's complete this professor okay good so I'll take another 10 or 15 minutes guys hopefully everybody is okay the next part is you know one of the things that is very important that you know especially if you look at traditional software products one thing that never existed there it may but very little is read they don't basically go back and you know do some uh retraining of your software thing right whereas in models it's a very very important aspect of it it's not that I deliver I'm done with it it has to have constant constant monitoring in terms of you know uh retraining and that review and monitoring is a very very very important model maintenance either it could be a retrain or a rebuild very very important that when you Define the strategy that you need to have that the third part which is also very what should I say important is when you do a software project right in general you don't do much experimentation whereas when you're doing a d project especially model building nine out of 10 times it's experimentations when you start you have no idea which model is going to work you have no very little you may have some idea you don't know what are the things that you need to tweak so there's a lot of experimentation lot of you know thinks that you know there is so much of experiment that you need to do which is something new in Ai and ml which is something that you know you need to account for when you're designing this timelines for these projects lot of them you know don't take that especially if you're not training a traditional product managers think okay you know production time 3 weeks two weeks I understand but there's a lot of experiments that you need obviously I talked about kpis very very very you know difficult and second thing also sometimes what happens is you know how are you going to Define this you know uh kpis right so for example let us say it is you know um [Music] mttr uh in what sense mttr did you say Arun so coming from the maintenance point of view uhhuh yeah okay um I don't know how many how many of you know what mttr stands for meantime to repair meantime to repair exactly so basically what he's talking about is when you look at you know um experiments or ntity right basically understanding right kpis it is very important that if something goes bad you can see how can I fix it second thing I was just about I was talking about is also for example let us say it has 100 it is medical thing it's reading 100 EX now it makes a mistake of two will you accept it I'm asking a real question this is a very important question that people have to let's have 100 xray it makes two mistakes will you accept the model compared to what compared to what obviously right but how many time it's very difficult to get compared to what because human data is not so easily available right how many times do you know that that the doctor has made a mistake lot of times the doctor might have made a mistake right but then what happens doctor prescrib something it doesn't work then doctor say oh maybe this maybe this you know what I'm trying to get here right compared to what I a very good question you bring up so but you are what are you comparing it to see these are very very important decisions as Leaders you need to take when you do not have a specified goal or something that you say this is what is today how many times do you know that the do sometimes the doctors make mistakes reading xrays or sometimes they give a medicine not because they intentionally want to BU mistakes but mistakes happens right so so that's another thing that you'll have to worry about and the last one is 54 million data million data workers and you know worldwide and spent 44% of the and unsuccessful data activities what do I mean by that that means to say you know lot of corruption in the data lot of you know you know uh ah data not available so there's a lot of other challenges that you'll have to also overcome so there are a lot of challenges in Di and AI project just wanted to let you know okay let's do this it's 7:36 can we take a 10 minute break come back eight and then first thing I will do is talk you about the group project I talked about it last time I will repeat it this time guys absolutely can we take a 10-minute break yeah Professor just a quick request can we get the link where you got this information absolutely here I told you right uh here you see this okay and it's I told you right I have it here thank you thank you absolutely because this this is very insightful analysis that 54 million data workers are facing this challenge yeah so I got it from here so I that's the reason I put here PMI reports thank you okay let's take a 10 minute break right come back and we will talk Professor can you share the link one more time please okay okay I'm sorry I was mute it is part of your slide what I'm trying to say here is let me share that I think somebody it is part of your slides do you see it here maybe he just left maybe you send this link on chat yeah um I put it this way okay maybe I can how do I it is not copying why is it not copying it should right uh okay you can move this copy yeah yeah yeah copy and where is chat here is chat paste yeah did you get it yeah got it so Professor when we were discussing this mttr right uh where the uh like acceptance of failures right like when we say okay can we accept a model failure if it is detecting cancer right so if acceptance of failure is so rare like is it correct to use terms like mttr because by definition mttr should be valid or it's actually valid in manufacturing where we all that okay there will be failures so what will be the meantime to repair right exactly here also absolutely here the question is you may have to accept failure and you may be able to repair it or you may not be able to repair it my question is is it okay to use the terms which are very generalized yeah well so the way you need to use mttr according to me you bring up is during the retraining process if the accuracy is going wrong which is what mty and if something happen here the way it happen is if the accuracy is going down what time does it take for me to retrain or rebuild the model ah okay that is your mttr in AI generally okay okay okay it is the same as to repair you also here repairing it right indirectly you're repairing it correct correct okay so that is your mptr here okay guys I will just stop my share and I will be back in e e e e e e e e e e e e e e can we start yes Professor yeah okay so just to show you guys this is the link I hope everybody got the link if you don't it's in the chat so this is the actually the where I got all the data from so nice report read it and just to give you a brief understanding um nascom oh Jessica you want it to be shared again absolutely I can share it again uh it's in the chat did you get it Jessica yes sir thank you sure absolutely no problem it's here in the chat and you can just click on it and it'll open this report okay yeah yeah it's a nice report don't read it now please listen to the class but you can read it later okay any questions or comments guys okay now what they have done I'm going to the next slide is basically they have created this sort of you know house right how am I going to achieve better know so they saying what are the pillars and so basically they're setting there are certain foundations out of which then you know you put your pillars and you know basically build a house right so that's how General starting so the things that is you know what is it that you know you need to do so if you look at it one of the things on the left hand side they have basically divided that into two kinds of things first what they have said is individuals and even though when they've talked about uh uh what should I say individuals they're talking about two types of capabilities one is technical capabilities one is techno business capabilities now one is the technical capabilities what they are saying is obviously there are two ways you can do it one you can obviously do it you know by reskilling your existing employees when they say reskilling what they're saying is you can rescale them for different things so for example if you are you know you have software Engineers who are good lot of software Engineers you can sort of slowly resk them to understand Ai and everything and they can become data scientist if you have people who are not software Engineers but they're technical oriented right you can res skill them to basically do data data acquisition you can do them to do data cleaning lot of other jobs so you can res skill them to do that so reskilling is something that you need to understand at the technical level do I because you can't just say I'm going to hire a brand new every time I can't go higher and hire new yeah some cases you might have to but in a lot of cases you may say is there a way for me to rescale my existing because they understand the domain or they they have worked enough in the company they understand the data so there may be a lot of advantages to do reskilling so that is basically you know the bottom line just saying upskilling path for entry level practitioners to data science from other fields and that is something that you know you need to understand how am I going to do that a either am I going to do you know internally if I have enough capabilities or I may bring in some external agencies to do that the second level of which they're talking about is training on basic flow of data science projects to for all practitioners very very important as I told you right uh we are doing a couple of corporate trainings right for example some companies where we are taking them from you know believe it or not from the person who books the travel all the way up to the CEOs the reason being the person who books the travel is not going to be a data scientist but he or she the travel you know if it is an in-house travel they're doing and they have quite a bit of people who are taking care of travel and other needs they may have one or two use cases they may say hey if they understand they say hey can can AI do this can generative I do this absolutely can generate way I do it today for example right there so many things you know I have to do it manually baby instead of me doing it very complex going into three things maybe can I ask a question that it'll be able to compile all the data and give it to me in a nice format that I understand by me just querying the system by not making it you know some very difficult SQL you know query which nobody understands right can I ask the same question in plain English so there is that kind of you know lot of demand is happening guys lot of lot of demand is now happening slowly from AI to gen the first thing that we are seeing the shift is people are saying in order to get information in my company I have to become a rocket scientist I want to convert that can I make that into plain English or plain any language right I'm just using English as see whether it is plain English or Arab or Chinese or Cantonese or Hindi is a different question but NLP natural language can I ask it in a language which I can understand rather than giving a very complicated question mark star M M2 as ID F you know half the time I have no idea that is the question a lot of people are asking and that is so lot of J projects that people are coming asking us to do is not do anything new basically saying develop an AI project interface where put the data so that I can get basic information which is very difficult for me to get today because I have do four five things with General asking sound standard questions lot of companies are starting that way just even like for example I don't know whether you guys uh know this or not even when AI started a lot of time even today if you go to a couple of companies some company says forget AI can you now put all the data into one place I can first let's have a game plan to standardize the data into Central system and create basic dashboarding I need to know what's happening first before I can say I need to do AI or something basic dashboarding I need dashboarding is nothing but you know creating Tablo dashboards that says you know this is my output this is now constantly can I get that can I make some basic sense out of the data that I have forget predicting something gen is also lot of people are asking such projects very common and very very you know most people are saying you know I hear the genic can do all those things that all good can we just do basic these things so that is what the second level here the second row or the second list training of basic flows or di for all practitioners second specialized trainings for dedicated roles to manage complex di project at an organization level this is where the implementation Engineers the you know Innovation engineer I talked about remember Innovation leads come into picture identify a few have specialized trainings for them then there are certain you know organizational which you know basically if you look at it right organizational things which is basically if you look at here I said right in-house infrastructure analytics infrastructure Skilling of the delivery team that is what they're saying basically you know organization data strategy invest in relevant tools data legs basically inhouse infrastructure basically create that infrastructure then adapt reusability of low code platforms obviously I as I told you right there nobody wants people to code it in python or something because it's good but a lot of people are saying don't do that help me out in you know uh if I have to do a project and if I have to modify it give me some tools like you know rapid Miner or something where I don't need to learn coding I can just do it so those are sort of the building blocks and then we'll build the pillars questions are comments on the building blocks guys oh one more thing I completely forgot and you guys also didn't remember me after this Slide the group project I said I will talk about it right you guys also didn't remind me when I forgot about it after this slide I will definitely talk about it sorry about that I completely completely you know the thought went out of my mind now I remembered it any questions or comments on the the whatever the few you have next let's talk about the pillars first one business understanding obviously right you need people who understand to come up with the right business problems absolutely you need the business understanding data preparation obviously do I have the right data do I have the right set of things then we go into modeling this is the the blue the what should I say the unstructured what should I say practices they're saying what do I mean by that this uh there is no formula to this right picking the right problem you know making sure I have the right data you have to experiment it you have to basically do a lot of experimentations and things like that next implementation is basically bringing it into production is not an experiment hey Jack if you can just please uh this one yourself mute yourself H hey Jack can you please mute yourself Jack if you don't mind can you just mute yourself thank you thanks Jack appreciate it um the next one is the implementation so obviously there is no experiment there right you make sure that the fls are there closing what do I mean by closing closing means to say making sure that you know it is uh what should I say a it is doing what it is and then you'll also have to make sure that you know um once you implement it you know it is working as it's at least for the beginning and then you again go back to the you know the retraining and if there is needed overarching practices applicable quality assurance obviously and risk management deriving effective outcomes for these are the basically the so they sort of you know created this house questions comments there's a lot more about this if you want to read about in the report I'm just giving you the highlights piece of it guys okay now I will go to the next slide I can but let me go back sorry guys let me go back and H now one of the things that we made a decision that I last to send it out I think ven last time said that and I'm okay with it that the order at which you will present will be reverse of the last time guys that's what we agreed with everybody is that okay with everybody right I don't want to change that it's okay right whatever last time you had the order we are just reversing that order if you presented it last you'll be presenting it first okay next basically what I want to do is the way the group project it I'm sure all of you know your groups the way the project has to happen is it basically you can take a business case from any domain any field it can be from any domain any field any education Finance any know bfsi I'm sorry uh supply chain I don't care guys any field micle what are the things I'm expecting to see in the presentation again the number of slides you see here are recommendation only plus or minus you can add but you know these are just recommendation it doesn't mean it has to be two slides if you can explain it in one slide I'm okay with it if you want to take three slides I'm okay with it but do not take too many slides because I'll give I'll tell you the time limitations for each one so each if you just take the business problems like 25 sles it's going to be a problem I know that not that anybody will do that I'm just taking an extreme example what I want you to do in the first one is the what is the business case and why are you considering it what is what impact does it have in the sense you know what is it that why is it important why are you considering it next if it is if it is not if it's a green completely Green Field and brand new case nobody has thought about it then there's no how with it been soled today I'm assuming a lot of the cas cases are today it's been solved in a particular way AI will help it to make it better or give a new dimension to it or open up new avenues so one thing I want to know is how is it being solved today next how gen AI or ni can be used to solve the business case when I say how I'm not expecting you to again this is not a technical presentation I'm not expecting you to say I will use you know random Forest I will use you know uh svm I'll use this model if you want to say it say it but more interested for me is basically I want you to walk through the use cases that to say this is how it is there this is the block where I'm going to use AI once I use AI this is how it is going to be done so I want you to walk the use use case with and explain to me where AI is going to be used and what is the impact that does that have that's what I'm looking for not to say that I'm to be using this model I'm going to tune the alpha value to this if you can say that yeah say that but don't waste too much of time on that now what will be the ROI if J so how is it better what is the return on investment I invest this money what will it be will it help the bottom line will it help the top line and don't say it help the bottom line explain how it may help it if you have numbers or percentages definitely use that if you do not have numbers or percentages I'm still okay with it but explain to me how is the ROI used what is Roi then what would be the impact what do I mean by that do I need for the existing employees does it make their life better do they need extra training do they need some you know so I want you to explain that what would be the corner cases that you would be worried about right what is it that you would be worried about if I put if I put AI what are the things I need to be worried about how do I make sure that you know these are the things I can make sure about it and any other thing that you would like to add any questions or comments on the business case guys next slide I talk about the duration and things like that but before that any questions or comments on the this slide so sorry Prof uh what's the difference between you know how is uh this being solved today and what the impact how is it been solved today means without a it's being solved let's say manually how would be the impact me to say if AI comes into practice does it mean that uh you know existing employees need to be have some special training do they need to have a new interface do they need to understand what should I say say the output do you need to train them on the how AI is going to give an output do you need to train them to say if AI says XYZ it means this what is the is there any additional training that you need them second what would be the coral cases so so excuse me quick one you you mention from any any domain any field can can we do industry agnostic for example um NLP chart BS as a customer service for any any but if anyone you can do it but then if you take one particular field it will be for example Roi can be used if not you can't use an Roi right right yes so so you know what I'm trying to get here yes yes absolutely okay there were a couple of uh chat things let me explain uh Nick is asking is this is the audience VC investors probably Nick if you think about it you know let's assume it is either the VC investors or assume you're presenting it to the chief business officer of your uh company so keep it at that level assuming your your CBO you're basically to the CMO or the CBO any other questions or comments excellent questions any other questions or comments okay it is going to be a group presentation each one will have 15 minutes to present and then I reserve not I we reserve anybody can ask a question guys we reserve a five to 10 minutes of Q&A from each group either one person or if you want two persons to present or everybody to present I don't think everybody I I leave it to you if it is one person to present I'm okay if two people want to present I'm okay you guys decide who will present I'm not going to nominate a person from each group you guys can decide each group can decide X is going to present or Y is going to present it may be that why is going to present but if there are any questions anybody else can answer obviously answer it's not like only why has to answer any other questions or comments this is the question what will be the ROI of Genera is used so in this case we can just use the Assumption if you are taking any absolutely right you but if I ask you how did you come up with that assumption you should have some basis for your assumptions basic assumptions you just can't say 100% my sales will be increased right I'm just I'm not that you'll be saying it you know what I mean to say it will be in percentage kind of things we can absolutely you can that's why I told you absolutely but if I ask you this question you may say you know what efficiency will increase by 5% you may say how I how do you think that you may say you know what today it is taking you know this six days to do this I'm assuming because of this I may reduce it to four days so at least 20% extra efficiency right I'm just giving you an example you know that level is for that one basically yeah that yeah some that at least don't say I think it will increase 100% why I hope so right that will might be a Char but that's what I'm not expecting I'm not going to probe you and say prove it to me today it will decrease by four days two by two days but at least you can say you know based on what I think you know a lot of time is wasted on this this if AI can speed up this process right instead of taking one hour it may take you know 10 minutes it will be helpful some some assumptions raw assumptions I'm okay with you guys that's it I'm not expecting you to what I want you to do is think that that uh what should I say that I want you to start thinking that way that's it that's the most important for me any other questions or comments okay I'm assuming there is none right um okay good let me go back to my slides next week also I will have another 10 minutes like this so if you have any other questions I will just again next week I will have the same thing so that you know you can also if you have any other questions I can still answer it so I'll have these two slides next week also not that it'll change it'll be the same two slides but I will just spend another 10 minutes next week ah where was I ah so this basically sort of you know goes into detail into each one of this guys so basically business understanding data preparation modeling it goes into you know what are the uh aspects of each one of it so for example if you look at business understanding it says project scoping objectives priority you know stakeholder management communication how do you make sure during a project you communicate to your leaders don't you know just obviously standard business practices domain understanding business so it goes into each one of it you know each one of it it goes into slightly details very very important that it goes into a lot of these details so I don't think I should go into each one of this tabs please review it if you have any questions or comments next week if you even tomorrow you can ask me not why next week sorry tomorrow you can ask me I'll be happy to go into each one of you okay now what I will do is I'm going to spend a little bit time talking to you about a few projects that we have done again today sorry I'm sort of taking HR and bfsi nothing to be let me tell you honestly why I chose bfsi not that I didn't want to choose bfsi um one of your students and I don't know if he wants me to name his name had a conversation on B and he wanted to do his thesis in bfsi sort of the finance domain I was thinking about it and said okay let me we have also done a few projects let's talk about it so that's it again what is bfsi bfsi sorry some bfsi stands for banking financial services and insurance B is for banking FS is for financial services and I is for insurance so banking as you know is traditional banking right Financial Services include your credit cards may include you know loans like you know or it may include any Financial Services basically your brokerages all that comes under financial services and then there is insurance sorry I think bfsi let correct me if I'm wrong isn't that is that an Indian term or is that a global term guys I don't know I'm asking you that question I genuinely don't know is that what they use in us also bfsi or is that something Indian thing it's Global it's Global right yeah okay I exactly I I was not sure about it I just wanted to make sure okay so I'm just giving you a few examples and then I'll go to one or two projects and we can talk about it and and then we will so one thing some sample examples right Enterprise big level data road maps what I mean to say here basically you know how do I basically combine the data into you know one place you know that's a very big important Enterprise because data is always into multiple places design of analytical networks setup of Enterprise data legs for for example Next Generation call center enablement which is a very very very big thing right not just in bfsi or anywhere right for example I showed you an example last time of customer Genera with agents right that's really a Next Generation call center enablement right whatever they they you know the customer was able to see a video rank take that video dat a shirt transfer it to a call agent call agent was able to know this was happening so that is the next generation enablement or you know cross- selling right obviously you know how do I cross we talked about it retention you know what are the chances somebody will renew your auto policy with you itself versus somebody else customer lifetime prediction right value prediction lot of examples a lot lot of them not just apply to here it appli to a lot of other area for example you know this one retention prediction of data you know of customers right or employees I we did a project in I don't know whether I told you have not one of the things that you know we did a project for one of the insurance companies in US large insurance companies in us um um so uh very large insurance company sh I definitely talk to you yeah sure sh definitely yeah sure I'll talk to you so um this insurance company basically what they wanted to know is see in India we don't have it slowly it is coming in India if you look at health insurance primarily it is coming but it's very it's slow not all of them do it but it's slowly coming in India health insurance only covers hospitalization to a majority extent right if you visit a doctor or if you buy medicines your health insurance will not cover it nowadays slowly those things are coming into the they're adding it but still it is something far in between it's not the common practice in India the only thing your health insurance covers in India today is your in hospitalization it'll cover it whereas in us it is slightly different at least in us I can talk about I'm sure Europe I don't know much I'm sure somebody can comment on it or even in for example Hong Kong or Singapore I do not know much I'm sure somebody from the in know region if you're familiar comment on it but in us your health insurance not not only covers your hospitalization it also covers a lot of these things it covers a lot of your tests it covers your Pharmacy which is your medicine it covers your hospital I'm sorry your doctor's visit yes there's something called copay copay is basically the minimum amount you have to pay for every visit and it depends upon know the plan you have and the company there's a lot of things it depend but it does cover sometimes they cover only 80% of the visit 20% you have to pay out of pocket whatever right but it does cover your Hospital your what should I say your doctor's visit it covers test and it also covers your Pharmacy in UK now the new trend is emerging they they even cover your health and fitness uh needs so so they prevent you from becoming sick exactly much more profitable for them oh very good and I will tell you very related to this itself is my example I'm because we did the project for them absolutely Ely good so um so what one of the things that you know you see there are certain illnesses or ailments see there are certain ailments like for example if you have an ear infection what happens you need to take the tablet for five days and then that's it right more of 10 times antibiotics are pretty strong nowadays and you know your air infection goes away but if you have blood pressure or there are diabetes there are certain diseases where it is a long-term what should I say medications right you don't take a blood pressure medication for one day right you take it once you start a blood pressure medication I don't know how many of you know that it's start of for a long period right guys you don't just take it for a day or two you just take it for a long time similarly there are certain elements where you take it for a long time now this insurance company wanted to know but when what happens is when you go to a doctor to they call it you know refill your prescription basically when once you go to a doctor the first time they write a prescription for the medicine let's say ABC medicine whatever that is they write it for 30 days then what happens you have to go back to the doctor and you have to refill it and they write it for the next 30 days then they have write it for the next 30 days blah blah blah whatever right 30 days 45 days I'm using 30 as an example doesn't make a difference what the duration now what used to happen is see what happens for the patients some of them actually get an appointment on the 25th day go to the doctor make sure it is done they get the refill some of them do it on the 31st 32nd day some of them you know for whatever reason right 37th 40th Day they actually get the appointment so technically they will not have medicines for 10 days because the refill the fill is only for 30 days and let's assume you take it every day and you don't miss any day for the next 10 days you don't until you go back to the doctor or you call your doctor for a refill whatever right you do not get the medication now as I'm sure all of you as he rightfully said right go right rightfully said it is cheaper for the insurance to give you that 7-Day med pay for the 7-Day medication then you developing some complexities and you being hospitalized because the paying for the hospital is 10x or 20x more than paying for the 7-Day medication but how are you going to make sure that you take the 7-Day medication right so one of the things that they wanted us to do is based on you know certain features factors and everything to make sure what is the probability that this person will come back on the 25th Day come back on the 30th Day come back on the 35 they wanted us to assess that thing and if there are any circum so they wanted us it's a continuous model that says you know that this does that and um they wanted to target the people who are not going to come chances like that you know chances are they wanted to Target them they wanted to make analysis and everything and and they wanted to make sure that you know and so we took a lot of factors right for that matter how far is you know do they sometimes they had to come to the doctors how far is the doctor so we took a lot of factors and anyway we did the project so retention is not just from an employee perspective here it is you know you can consider it in healthcare is know making sure that so they wanted us to categorize them into green ABC right green yellow red or something like that right categorize them so that you know they can start sending messages to people who are you know always in the red to say hey you need to do that you need to do that versus people in the green maybe you know what send them less often messages things like that they wanted to do a campaign and things like that so there's a lot of things that you know I whatever even though I use the word bfsi here it doesn't have to apply for bfsi it can apply for many you know uh what should I say uh what should I say areas so there are a lot of examples here you know things I'm going to talk about one two examples that we have actually done but before that I want to spend a little bit time a little bit time talking about and this a very it's a very sort of how should I say simple but nice architecture to say when I say to you right now that we putting the data into a single place taking the data to Cloud what does that mean I just want to show you you know how is this basically storage of a you happens so I want to give you the architecture of for storage I'm going to go into it technical but you know I'm not going to go too technical into it but I'm going to use certain things so that you get an idea if I'm sure a lot of you are working in this area so you may add or subtract and you know if I say something you may say no no no it may be doing like this nowadays so please add and subtract guys you know so on the left what I have basically said was internal and external sources so the word internal may not be 100% true I can see internal and external sources there are a certain things here because I because I took this also from a certain you know this one this are India specific and I will explain to you what do I mean by India specific the first one you see here is called cibil cibb report civil is a credit rating agency just like you have TransUnion Equifax you know again one more thing right my knowledge is either India or us because those are the only two countries where I spent a lot of time so I'll use those two examples but I'm sure every other country has an equivalent to that apologies I may not know every country's you know Credit Agencies and everything so I'm going to use these two examples apologies for that but I'm sure you can translate that into your own credit rating agencies so Cil is the credit rating agency in India it's called civil score just like you know in us we call it a FICO fic right FICO score they call it nowadays there are other scores are coming up lot of other scores have come up also but FICO is one of them very popular basically it is it was for I think 850 if I'm not mistaken so they said how is your score anything about 750 was considered to be good you know uh 600 to 750 was average okay and then below 600 you know you are in the red zone or you know so based on that you know even when you went for a loan or something right the rates also differ if you have the really good scores then you know you got the best rate preferential rates for your mod cages if you don't have the right rates then you know they may ask you for other details or they may ask you for a lot more information something right so similarly you have something called civil in India and so you have the word civil now you see the word c kyc and e kyc kyc is basically a term again used in India called for know your customer what does it mean know your customer is basically to who is my customer know your customer doesn't mean I have to handshake obviously it's not not just hand shaking and say how are you sir my name is WESH or something but more than that what documents do I authenticate the customer or know are the person and in India We There are two major forms of authentication that has been created one we call it Adar Adar is nothing but a fantastic scheme right where each citizen or each person in India has a unique uh I think it is what a 16 digit number other correct me if I'm wrong guys 16th or 14 digit number I don't remember I think it's a 16 digigit number I may be wrong whatever x x digigit number every 12 digits 12 digits okay 12 digits sorry it's not 16 12 digit it's a unique 12 digigit number for every person in addition to the 12 digits one of the additional things that we have done that has revolutionized that is basically we have every person with the 12 digit has at least one form of other identification or another form of identification asso Ed with it the one that is most common is your fingerprinting so we have in addition to the 12 digits they have taken the fingerprinting nowadays in addition to the fingerprinting a lot of them especially if you if you want to get a new AAR or something even your retina scan is being done so now the beauty of that is it's not just see for example in us and again if you look at the social security number the social security number was created in the 1960s Adar was created in the 2000s so it's it's basically apples and oranges I can't but irrespective of it a social security number is 642 4 so 3 + 2 5 plus 4 nine digigit number right because I'm just thinking my social security number but anyway okay so nine digigit number it's a nine digigit number and a name but there is no physical identity to other than other IDs you can show your driving license there's nothing actually bodily you can say this number belongs to this person because it was in the 1960s guys it was an old I'm not I'm not I'm not saying it is bad thing but the beauty of Adar is that they have made sure that you know all your fingerprinting all your retina scan is associated with it now the beauty is the wonderfulness about that is now the moment I put my finger on a scanner automatically my Adar comes up and so C kyc eek kyc is electronic know your customer that is done at the individual level what do I mean by that individual level means let's say you want to open a bank account automatically they want you to what should I say um keep your finger and automatically your bank account number all your details about you know they they can authenticate that you are venkatesh or I am wenes not you are in terms of you know first party I am venkatesh I don't need to do anything else now automatically I'm authenticated as venkatesh that is called ekyc ckyc Central kyc Central kyc means to say it is a central record keeping so if you do your kyc once and authenticate yourself let's say Insurance wanted the data somebody else wants other agencies wants the data they can ping the same database that is called basically ckyc similarly you know for example cam cam so we have another agency in India this is for mostly for your brokerage and everything right all your brokerage records and everything there are a couple of agencies that have basically in order to help you in order if you want to set up a brokerage account and that linking of the brokerage account cam is one of them so it is there so basically this is Indian data guys just to make sure that you know just to give you a few examples but doesn't make a difference what I'm trying to say is since it is bfsi personal data and everything right all the sources and in sometimes it could be external data also these are all internal sources you may be also adding external sources right and all of that is now sucked into in two formats the first format could be batch what do I mean by batch every overnight I get the data I load it into my database or real time I'm getting the streaming right every time there's an transaction I'm getting a streaming so either it could be a batch format or it could be a streaming format that tools I put here right for for example you know Kafka which can be used or stream Flume all that can be used for either bat or stream there are tools that are being used so from the sources it is pulling this databases this is pulling this data into a central storage and processing unit this is sort of you know the first thing that has to happen you need to identify your sources and try to put them into a central location and that is what it is being done and for each one of this there's a connector that you have to do so either it could be in for a database it could be somewhere else it could be an Excel sheet doesn't make a difference you need to write a connector to get the data out of it so that is the sources and the data acquisition part any questions or comments before I go to the next part now that you have pulled in the data you had to create certain business rules what are the business rules the business rules is let's say you getting the data every second some data do you really need to keep every second of data or can you say you know what I am going to aggregate it for every uh 30 seconds or you may say whenever the data comes you know what I you know I want it to be aggregated at the consumer each customer level or each what should I say product level you can write lot of business Foods how is the data going to be stored is that business R is it going to be stored like blindly every data that comes or there has to be in addition to that are you going to create new uh rows or columns which didn't exist from the existing data those are all called business rules which you need to write so you start with raw data and then based on the business rules you basically and then there's the DQ check what is the data quality check making sure that obviously right there's no that you know quality of data and everything once you do the quality of the data check then it basically does the processing of the data what is the processing of the data if there are missing values do you want to fill it if do you want to aggregate it do you want to create new features by adding two or more data points things like that happen and that is called processing data and basically guys I'm following the arrows raw data to business rules to data quality check to process data to data transformation data transformation again basically means to say am I transforming the data what do I mean by data transformation certain times in models you can transform the data you can square the data you can multiply because it it helps you in modeling so you will learn that when you do basic thing you can transform the data a lot of times in order for you to help in modeling sometimes you can transform it into also data transformation may include for example right you may make sure that all the data that you get it I I only store it in what should I say kilo what should I say grams if it is something or I only store it in single currency whatever right you can do some data Transformations and then basically it goes into analytical data mark this is where you know the data it can be used for actually your modeling and everything over it you have the data science engine which is the modeling and everything you have the automation if you want to do certain automations and things like that and finally an application right that uses data science you can create an application so that is the storage and the processing piece of it and then the query later if you want to get the data and query the data query is basically get some information and then you can come basically the output of it can be put into a tableau what is Tableau creating dashboards things like that and then you can also basically this query lay data is where people are saying right get me chat GPT kind of know generate don't make me query the data in a very you know raw manner or it can be something where automatically the query is something that is inbuilt every time the data is ready I basically push it to to or if I want to collect dat so this query later can be either prompt based or it could be automated and then you have access control data security governance infrastructure and everything like this is sort of the total basically how a project or a data science project if you have to do all these things has to be taken into consideration until and unless you're just playing with on your laptop that's a different thing I'm talking Enterprise level data science project questions comments any questions or comments on this professor I have a question this is right yes please go ahead so if we look at the storage and processing so we have received the raw data we've applied the rules we've done the check and you know then we are processing the data for let's say missing values or whatever so when you see automation because this data is going to keep on streaming you know based on whether we are doing it uh year time or whether it's a batch process so how do we take care of you know correcting the data so are we going to write certain tools or or you know it has to be tools where whenever data is available comes into a dump I take care of it hello so that's why you are saying Auto automation engine out there right exactly automation engine or you could Al say right see sometimes what you may do it automation engine could also be decision could not be human based they see it automatically the Deion is taken care by the machine so for example I gave you right the moment you know I gave you an home automation example the moment I detect water level in my leakage shut down the water supply there's no human involvement so I get a sense data from my you know water leak sensor sends me some data that says there's a water leak shut down the water supply that's automation right okay got it thank you there could be automation anywhere and see what I'm trying to say is not every case has automation I'm just giving you certain cases May automation some cases may not have automation excellent question so there could be so I'm showing all the pieces doesn't mean every project will have every piece so Professor if we want to put a gen aai lens here where are the components that absolutely can give us incremental success in implementing gen exactly so there are couple of places where you can use gen so first of all in the development of the model code development you can use J right I showed you right you can use that's a simple but mostly in the query layer here uh uh where you know you can use gen so that now when you're doing a query you can ask people to do query on a prompt based you don't have to be basically always or it could be in the data transformation what do I mean by that remember that same examp I'm going back to the same example right remember one of the examples I showed you where the person said you know what Hey listen um I'm going to compare these two documents with my uh compliance of my office that is basically I have all the data here processed ready and you know I'm comparing two layers that as at the query layer or at the application layer basically end of the day all of that happens at the application layer Hello Prof uh could you give me some explanation the difference between data infrastructure mops and develops absolutely uh is that somewhere here or you're just asking me in general I I help you absolutely I'll help you in those three things is there is that somewhere here in this it's not here right but I can explain to you absolutely data infrastructure mlops and what was the third one you asked develops devop devops right okay um okay uh maybe I'll do that okay infrastructure means to say it is the data infrastructure is basically this you're seeing here it is the hardware the connectivity the sucking of the data all that is the data infrastructure how do I get the data where is the data available uh how do I get it do I get it on a weekly basis monthly basis as a batch basis streaming basis setting it up cleaning the data even okay those okay even before clean getting the data those are all the infrastructure the pipelines I need in order to do this let's take each one of it so that you understand any questions or comments on the data infrastructure part any other question did you understand any questions on the infrastructure now now that you have done the infrastructure devops what is devops devops is basically you know getting the data in maybe you know cleaning up the data making sure that you know the data you know connectivity happens transforming the data all that comes under devops mlops in addition to all of this there is one more factor that you need to understand which is model selection modeling how do I model which model to do it how do I create what should I say new uh what should I say features things like that are very important when you do uh what should I say uh uh in mlop so devops plus mlops includes the modeling piece of it that is what infrastructure devops and mlops are hope I've explained to it if you don't understand please ask me I'll be happy to explain it thank you okay now there was a question from Vin that says you know other use cases business rules of processing absolutely so now for the business rules uh thing uh V one of the things that nowadays they're asking is you can write see for example instead of defining new rules can I ask the data to create new rules on the go right you might have never thought about a new rules can you now go maybe you have a query that says Hey by the by I want this model to in plain English I want this model to create these take these to what should I say uh things put it together and create a new and you know add this as the new data point and run the model that is a new business rule that J can help you out which you know if you had to do it earlier you have to go tweak the model now you can can I just ask the question and the model automatically take that put that new data point create that new data point and create the new column and add it to that this one uh for the analysis Professor can you explain this again are we saying that gen can be used to add that feature in the data sry absolutely so if you can say hey can you create there people are looking can you create this new feature and consider this new feature while modeling okay got it but the data has to be available you can't just say I don't have the data go go figure it out the DAT you can't do that got it I've been reading some um that you might be able to use gen to generate synthetic data to train the model ah so that itself is a Sony brings up a very good aspect of it uh that's a new can of worms so you 100 I'm not denying what you're saying is not true okay how many of you know what is synthetic data guys anyway you brought it up I'm going to talk about I'm not that I don't want to talk about it but anyway since you bring it up how many do you guys know what is synthetic data data well it is made of data is you are right Goro but it doesn't mean that it is wrong data two different things it means the data doesn't exist you have created the data for your ah exactly for example personas customer personas customer personas are new synthetic data you can consider or so for example let us take a picture of a word I give you simple and complex examples very good questions that you bring up guys I take a picture of a very simple examples I'll give you I take a picture of a bird now what I can do is I can create another picture from that bird where I chop the bird beak from Full beak into half beak I will now create another picture where I will basically say one of the Wings is out the bird maybe both These Wings May done the original picture the second thing what I will do I will slightly tilt the picture into a side different ways so basically I'm creating different versions of the same picture all these different versions didn't exist from the natural source I got this picture from I only got a one or two WS of a bird maybe bird sitting on a tree and another one know maybe bird flying but with that I can create you know I can darken the bird wings gra the word I can create 20 different aspects of the same Bird right I'm just taking bird as an example all of those 20 aspects can be considered as synthetic data data that you have created from the original data that is on text now from text what I can do I have a pieces of line I can write the same thing in multiple ways or I can remove certain certain make a few spelling mistakes I can create different versions of it similarly audio I have a beautiful piece of music I can basically add a little bit of noise change the treble change the boss I'm just making up right whatever I am creating different versions of the same that is one of the ways of creating synthetic data now there is another way of creating synthetic data people are talking about which is slightly more according to me you know talking about it this I think is everybody accept as the right way because that way I can train the model in different things that didn't I I didn't have data but I'm creating new data that's okay because it's still the same Bird right for whatever reason bird broke it slightly it's beak I will be still be able to recognize it as a bird so it's a great idea or know for example I have a human face what do I do I let's say I had a head injury I have a bandage over it I have that over it I have you know ey patch over it whatever right I suddenly grow a beard like I only have a mustache I have you know you know different types of mustaches I have different types of beards I that's it but what people are saying nowadays is and one of the things that is also scary is and this is a debatable thing and I don't I don't want to say it conclusively one of the things people are saying is today what people are saying is I am exhausting the see the rate at which data is created versus the DAT rate at which data is consumed the rate at which data is consumed is a lot faster than the data rate which data is created today there is so especially with this large language models they're saying it is I need more data I need so much of more data that you know uh I need so much of new data that you know I may now start looking into creating some scenarios which are similar to existing scenarios but with all made up slightly made up you know data and different forms of dat this is what people are saying and this is a little scary part right because how are you going to make sure whatever synthetic data that you created is really good and accurate you know that's the part of this so when Sony use the word it is slightly know absolutely is right but that's a little scary thoughts questions comments what they are saying is for example if I obviously I have trained with every known Picasso painting can I now start using you know data that says you know I have a dog if Pico I painted this painting how would you paint it this way can I now start training those model those data which I created say okay I use it to I use gen to create this is how because painted this model and used that as my training data if Pico painted you know uh picture you know or you know Wango had a painting if Pico painted a Wango painting how would you the same painting right how would he paint how would it look like and use that to train say okay this is how and give it to some experts and experts say yeah this is good enough or you know okay and then use that good bad I don't know I think that's the conscious decision but it's also happening unconsciously now as well because so much data has been generated by gen and it's automatically being picked up by these LS for training anyway mhm absolutely so the second part of it is exactly as I told you right it is process data you can call it gen again there a feedback loop I'm feeding what I've been you know saying good bad I don't know the jury is not still out on that guys I'm sure you guys have your own comments I'm not saying you're right I'm not saying but this is happening absolutely he's right he's happening so this is what we call it a synthetic data any thoughts questions comments okay now that we have that uh this is not okay I I okay I will take up huh I'll take up this examples couple of examples I will talk about it maybe you know what I can uh continue this next week also I'm sorry tomorrow also few more examples and then do a few things this example I'll set up the business case and then you know this was for a Latin American uh restructuring firm unfortunately guys I can't use names apologies what does this company do this company basically picks up assets when I use the word assets it is basically [Music] um you know either you know loans that are late loans that have you know less greater than 30 days not paid loans that are you know either may have gone into default or you know companies that have gone into debt not repaying so bad assets or bad debt assets you can call them and they obviously right they get it for 20 cents on the dollar 30 cents on the dollar or 40 whatever right depends upon it and they basically pick up that loan or that asset and they want to basically say can I re return it around or can I collect it from the finally collected right if it is know 1020 cents on the dollar and I collect another 30 or 40 cents good enough right I made a 40 cents uh profit I paid 20 cents and I eventually got 60 cents so I made a 40 cents profit I'm just taking it up right for example what they wanted us to basically understand is depending upon their own you know existing loan portfolios that they have done in the past depending upon public available data what is the basically you know uh chances that you know this the what are the probability basically this asset will be you know that they pay for this asset or you know this is basically they wanted us to rank the assets a b c d which will help them in the bidding processes because while they're bidding right it's not just one assets so many assets are there they wanted us to basically even before bidding look at these assets look at what previously they have done on uh their idea put that come up with a model that says you know if you pick up these assets based on your probability and some publicly available data and other things you know what are the chances that you know you will make money on this what are the chances that the person will repay if not full 30% 40% whatever has been they wanted us to basically make the prediction so they knew I should I bet on this asset this asset this asset and the prediction not only said that but also gave them a range you you know the payback could be you know 30% or 40% or whatever right x amount and all they you may get you may maximum get between 30 to 40% of your money or you may get 50 to 60% of the money now they know how much they can build it for right if it's a $100 and if a model says you know let's say you know you get for $50 they will only bid for $30 or $40 right they will not going to bid for $60 you know what I'm going right obviously didn't give you a number we put them in a Range so asset restructuring very common example questions comments what is the meaning of Mae 33% better Ma on the uh right side the asset evaluation time uh oh um what do reduced uh ma is margin on assets Point margin I mean it's 33% better better margin but AE has some word when you just Google it I don't remember you know but it's on margin okay any other questions or comments I'm going to show one or two examples guys and then you know um if there are one or two left I may take it up tomorrow another one was you know basically easier one delinquency basically they when a person applied for a loan what are the chances of he person with default and things like that and you know what is it that they need to do how risk of that person basically they wanted to do the risk profile of that person and things like that so we have done quite a few work in South America Latin America whatever you want to call it right in this case see in some cases I can put the name because they are okay to put that name the third okay I will again this is a loone one I want don't want to say this is an HR analytics one which we did which is two types of analytics they wanted us to do one basically they wanted know for existing uh team members or members or the company what is it that you know what are the probability that the person may leave the company the next six months next three months next you know they basically 369 is what the on 12 they basically anything beyond 12 it's very difficult they wanted us to what are the probability and obviously right they wanted us at two levels one at the mid level and senior level management level which is a lately you know I have to be careful and then even at the lower level because they wanted to make sure that you know what is happening the second thing they wanted us to do and this was more difficult and it was not very not very successful also when we are interviewing an employee can we say what how what are the probability he or she is going to stay with us you know they're not going to churn in the next three months six months and N9 months we did certain things you know based on past you know people similar who have we have H they have hired and what are the probability that you know these people is tired and then we also asked we made them ask a few particular particular questions based on that answer we said you know to be honest guys we were about 35% accurate again these were not geni models these are traditional AI models some e ethical ethical issues could be there yeah could be ethical issues right but but you know what but the company really wanted to know right what are the P CH that this person will AB live see there's nothing wrong in answering that question right but yeah but we were not very successful to be honest the ones that were with inside the company right we could see because of their you know we could know because of their certain things like for the performance and you know how long they have been in the same position and you know what was that and also the customer feedback there were a couple of things that you know if you are in the front line we were able to there we were able to about know I think 53 to 54% I think 57% I think was our success rate here it was very bad I don't know good bad depends uh so Professor what model was used in this particular case basically it was obviously right it a deep learning model and the data was taken it and it became basically you know uh what a classification right you know a b c d three months 6 months 12 months uh which which category they come under so so if if we are if we interesting to develop a model as a company uh would you be able to help us uh well V in the sense yeah obviously I can recommend commission commission upgrad yeah I can recommend a what should I say a consulting company to you absolutely thank you I can recommend a company to you absolutely I can do that but then you'll have to talk to them and you know thanks uh absolutely uh anything as so we were able to do it guys it is it's not very sometimes these are not easy solutions let me be honest with you these are not easy problems to solve even humans have not been able to solve these problems right it is definitely not an easy problem to solve we were okay I'm not saying we were very successful not bad fraud detection is one more thing you know graphically you can take a look at it and things like that you know basically you know fraud detection is another big area in this one guys what should I say uh nii because you know it has to be really good right so there are a lot of things they look into it one of the things they use actually is graph Theory but then they develop a models basically to say you know probably you know what is the probability that this transaction is fraud right and that is very critical that you inter very little time you have for intervention right you basically um you have to stop at when the transaction is done why because if the if you let the transaction go you may say I may detect it later and warn you but it's too late because the goods have been exchanged let's say I I am doing fraud transaction in know and I buy let's say $20,000 worth of or $10,000 worth of goods now I may detect it a little later but by the time the customer would have walked away with the $10,000 worth of goods right I can't go back to the customers say give it back yes then you know the problem with with the credit card companies and especially in us and I'm sure it's also in India is you know the end consumer can only you can actually maximum do is50 as long as you know second you can also if it's a very fraud transaction even the the goods person that you know they can't you know you the C the credit card companies must eat it up a lot of this number value this what should I say amount and so it has to be in real time you cannot uh you know stop at a later time especially in retail transactions because the the goods walks away from the store store or whatever right then it's too late so they wanted us to sort of look into it we did that it was okay I'm not again you know but nowadays they're very sophisticated models for example you know most banks and I'm sure if you walk in the banking industry they have some very sophisticated models for this frauding right and they're pretty good you know sometimes they may be over you know they might have been a little bit of over this one but you know that's okay right sometimes you have to live with it and I think this is the last I have a few impacts on gen which I can talk about tomorrow guys any questions or comments I think the time is yeah okay 9:05 that's not bad any questions or comments one more question let me ask you and this is something I want to remind you has everybody turned in your assignment one yes Professor even I take the example in the insurance domain preventing Brent insurance claims using I Ai and I2 uhhuh okay fantastic so what I will do is I'll try to finish your valuation of your assignment one this week guys this week as in Monday to Saturday I try to finish it as much as possible I just want to get it done with and for some people who for whatever reason you have not turned it in please turn it in this week I you know please turn it in this week I know May 1st was the date fine for whatever reason you have not turned it in please turn it in this week okay I'm I'm doing the evaluation this week right I just want to get it done with I don't know how many have turned it in but I will do it this week and I will just get it done I'll try to get all of them done this week okay no issues uh guys if you if you turn it in you know a little late absolutely no issues don't worry about it please turn it in I'm the last person if you for whatever reason you turned it in late absolutely don't worry about it even if you have not turned it in please turn it in this week I'm okay with that yeah the class is done guys thank you Sam I just wanted to thank you guys thanks a lot thank you sir thanks bye