hey guys so in this video I'm going to show you how to build AI agents on WhatsApp that can do personalized Outreach respond to customers use databases can notify humans to jump in and can understand voice image and even documents now I really wanted to make this video because I think WhatsApp agents are a huge opportunity in the AI space right now there are now over 50 million businesses using WhatsApp to interact with their customers and almost three billion users and the large majority of these businesses are still using humans to attend clients on WhatsApp meanwhile the large majority of these interactions can be automated with AI in my agency we've gotten in dozens of requests for these types of agents and have delivered several of them by now and you've probably heard all these softwares like agentive uh chat base and many chat who make you really easy to deploy your AI agents on WhatsApp and although I really like what these guys are doing the biggest issue at least for me has been that most of these softwares still lack vital features to actually make them useful for many business use cases like for example human handoff doing outbound messaging uh some don't have chat management or chat history some don't have the ability to integrate tools to for example save data to a CRM and uh they don't process voice image and documents yet so in this video I'm going to break down my setup that covers all of these business needs and I'll show you a demo and a full step-by-step breakdown of a WhatsApp agent and I'll also show you an option of how you can set a system like this up even without the official WhatsApp business API if you don't know me yet I'm I'm Ben I've been implementing AI agents and AI automations into businesses for over a year now also run a community with over 4,000 members and if you're business looking to adopt AI or Solutions like this into your business you can also book us in for a free call in the description below so before showing you a demo and a breakdown of my setup I want to go over very quickly the opportunity because I really believe the opportunity is huge and second I want to give you an idea of the softwares that are out there to build WhatsApp AI agents and give you an overview of the advantages and disadvantages of each one uh and give you my recommendations for each specific use case so you can decide better what the best tech stack might be for your specific use case or for your client's uh use case so let me start with the opportunity so of course the WhatsApp Market is huge as I said uh 50 million businesses now use WhatsApp and of course there are three billion users uh and it's still growing exponentially especially in the WhatsApp business uh uh part more and more businesses are adopting WhatsApp to interact with their customers and honestly it's no surprise because the engagement metrics on WhatsApp versus traditional communication channels are just ridiculously better uh here is an example right WhatsApp has 98% open rates versus 21% with email right WhatsApp messages are read on average within 15 minutes uh WhatsApp has 60% clickthrough rates uh on links versus 3% with email and 68% of customers or users say WhatsApp is the easiest way of contacting a business now of course this all depends on the continent right in the US uh WhatsApp is not adopted yet by the masses so of course that's not a uh a good market for it but the Europe South America and Asia are are huge potential markets for this and really AI agents AI WhatsApp agents are a no-brainer for these types of businesses because of course with AI agents instead of humans we can have 24/7 availability we have faster or immediate response times therefore an improved customer support uh we can handle way more or customers can handle way more inquiries they have a reduction in costs and they can redeploy the staff they used to uh uh engage with customers to interact with customers for more strategic tasks so why do I think there's such an opportunity for uh AI agencies or AI Freelancers uh to build this for companies is because in the end it's still quite a complex setup uh for companies themselves to set up and that's Al always where there's an opportunity for agencies or Freelancers to jump into so setting up WhatsApp agents is technically challenging right usually they require customized Solutions right we need to actually uh set up and integrate it with their specific crms or customer uh management tools uh we need to import knowledge bases and this requires AI expertise right AI expertise really is necessary to make them work well enough uh to to Really reliably deploy them inside of a business so that's the opportunity for agencies now what are the most common use cases now of course one of the main one is customer service right now second one is inbound sales and appointment setting which is very commonly used with WhatsApp too many companies actually run Facebook ads or Instagram ads and the call to action is directly to a WhatsApp chat uh so this is very common use case too then we have marketing and Lead genen now this is actually a really good use case for WhatsApp to because of the high open rates and click-through rates this can be very powerful for businesses but has to be done right right because of course meta has some guard rails in place to avoid spamming people over WhatsApp right so you need to be strategic here in the end it's very similar to upbound email we don't want to be spamming people with irrelevant messages right but this is a very strong uh uh use case for WhatsApp agents if it's done right and I'm going to exp uh show you some examples later too then we have notifications which is another big one for companies right so we for example you can imagine order notifications right order status updates uh payment confirmations uh things like that very useful to do that through WhatsApp again because of the high open rates uh and click-through rates right and lastly we have uh feedback and Survey collection now what are the most common niches that WhatsApp business is used for uh a wi variety of Industries but the most common ones are retail and e-commerce of course online stores and local shops local Services uh rely a lot on on uh WhatsApp too right salons James repair shops Home Services Etc then we have hospitality and travel like hotels restaurants and tour operators Education and Training schools universities and online courses real estate and property management and uh Professional Services like law firms and accountants so many uh types of businesses using Whatsapp if you live in a country where whatsapp's used a lot you will know right I live in Brazil and whatsapp's used for everything now what are some of the key requirements for businesses in my experience uh these are really important from what I've seen uh to actually make them useful inside a business and the first and the big one is human handoff now most of the time we can't actually completely automate this in the background yet uh we at least want to have the ability to jump in as a human for example when our AI doesn't have an answer stored in its knowledge base which happens uh when we when a customer is ready to book uh we want to actually jump in and close them ourselves uh when a customer asks for a phone call when there's a problem that's not exactly accounted for inside of the prompt or inside of the base there are many scenarios where we still want to sort of be there as a human now the second big one is outbound messaging and notifications now what do I mean with outbound messaging it basically means that the company is able to do active Outreach to the customer instead of just being able to respond incoming chats now why is that important because for many use cases the company actually needs to be the first to outreach to the customer for example with these notifications like payment confirmations uh but also for sales related tasks like someone filled out a lead form on our website the company needs to reach out to the customer first over WhatsApp right uh and also of course for these more marketing and sales related um use cases they need to be able to do active Outreach to the customer now why am I saying this because most of these third-party softwares actually lack the ability to be able to send uh outbound messages over WhatsApp uh then we have chat interface and history access which of course very important too for companies to actually have an idea of what's going on in these chats uh and to be able to jump into if they need to and again I'm saying it because a lot of these softwares lack uh this right now the fourth is knowledge integration of course we want our AI agent to be able to respond to questions uh from the customer about the company uh so we want to have a knowledge base uh to perform rag on to actually be able to uh answer these queries right and uh tool usage of course is very important too right just like a normal customer support rep needs to have their CRM at the side to actually update state look for uh the status of of orders things like that our agent also needs access to that right and to also be able to actually save data to a CRM or update data of the CRM and lastly of course many people use voice messages images and documents to communicate with companies uh so it's very important uh that our AI agent can also uh understand these media types which brings me to the next part which is a little bit of an overview of the softwares that are available right now to build these agents right and the types of WhatsApp agents that I've seen right so I basically I got gotten in three types of requirements right the first one which is the most common one is uh inbound plus outbound agents so companies that need an agent that can actually actively reach out to customers uh and also be able to respond to uh queries from customers coming in right but I've also seen uh outbound only for example uh companies that just want to do sort of personalized Outreach or marketing campaign through WhatsApp but doesn't actually need to automate uh the the conversation itself right and then the last uh type is inbound only right for example the appointment set I just mentioned uh a company that only needs to respond to incoming queries from customers so basically for the softwares we have three uh types of options right the first one is the third party provider so these are basically the softwares that I mentioned before and these softwares are basically meta official partners that can provide the official WhatsApp business API for you so the big Advantage the big upside of this of course is you don't have to go through the hassle of actually setting up the official WhatsApp business API and the second thing they do of course they make it really easy for you to build AI agents and deploy them on on WhatsApp in basically five minutes right and also they're fully compliant with matab because of course they're metab partners and you're working with the official WhatsApp business API and most of them also have knowledge integration so you can give your agents access to knowledge to actually be able to respond to questions now what are some of the downsides the first one of course is uh usage costs because because they're providing this for you this service they have to make money right so they charge some extra costs on top of it of course and the second thing is most of these softwares still lack some of these key features that I mentioned before now I'm not hating at all I think they're all doing a great job and I'm sure they're working on these features but uh at the moment for some of these use cases they are not useful yet for example agentive at the moment uh they don't have uh actually chat history yet they don't have a way to do outbound messaging right you can't have media interpretation and there's no human handoff right now chat base is a very similar one to agentive uh now agentive actually does have tool usage which is a big one uh chat base very similar to ative they actually do have chat history but also no outbound no human handoff no media interpretation and I don't think I'm not entirely sure but I don't think they have tool usage either right then we have many chat which is more like a traditional uh chatbot Builder right so we have a lot less flexibility with our AI agent uh but we do actually have bound messaging but uh and and chat history Etc but it's a lot more difficult to have knowledge and tools Etc in a platform like many chat and we have drxy which is another one which actually does have a human handoff feature but doesn't have outbound uh and media interpretation and also no chat history I think yet and uh then we have B bot press which gives you a lot more flexibility but from my understanding uh they don't have media interpretation yet and also I don't think about messaging but I'm not entirely sure on that now the second option which is my preferred option especially for the complete Solutions right the inbound plus outbound is going with the actual official uh WhatsApp business API now the big upside of this is we can actually start working with our most flexible AI agent Builders right like nadn relevance AI voice flow make Vector shift uh flow wise we can just integrate our own official whatsa business API into these platforms and therefore we have a lot more flexibility and therefore we can do things like media interpretation right we can do out found and notifications now keep in mind that with meta with official WhatsApp API to do outbound messages we need pre-approved templates from meta uh we can only send those out again because I try to prevent spam but we can still do uh quite some personalization because we can work with variables and through those variables we can still keep uh uh messaging very personalized for example in about marketing or sales right I'm going to show you that later too and then most of these softwares here have uh also lower cost right then those others right and again you're fully compliant with meta because you have the official WhatsApp up now what are the downsides again you have those pre-approved messages but again you can do quite a lot with those right and I'm going to show you later how you set that up to inside meta uh right you have to set up the official WhatsApp API now again it's not that difficult right if you've done it once you know how it is and the second time will probably take you 15 or for 15 or maximum 30 minutes right then uh the other downside which you have with them too is you need a separate number right you can't actually use your actual phone number because you'll lose access to that and you can't use WhatsApp web with these WhatsApp official WhatsApp business API numbers right and most of these platforms don't have the actual uh right UI or the front end to actually have the chat history and uh the human handoff feature so I'm going to tell you in a second my setup to actually have uh both of these features in there too and then the third option is actually to do it without the official WhatsApp API now this is a little bit more of a hacky way because of course MAA doesn't uh wants people to use their official WhatsApp API it's also how they make money right but there is a way uh to do it without the official API and you can actually do it on relevant AI so there are a few upsides of this and the first one is of course you don't have to set up the official WhatsApp API but second you can keep your existing number so you can do this on your personal number or your existing number and you still have access to Whatsapp web right which is Big because you can still jump into WhatsApp web to actually take over the conversation which is not in general not possible with the official WhatsApp business API right then the second is we can do 100% personalized Outreach because we don't need these pre-approved uh message templates that we need with official API so we can actually do 100% personalized Outreach and it's also the cheapest now what are the downsides we can't do media interpretation through this uh uh setup it is a bit slower because inside relevance they basally check every two minutes if there's new messages so it might take two up to to 2 minutes before your agent actually responds there might be a little bit more errors because of course it's not uh the official sort of setup we don't get a direct web hook right and uh there is a small risk of meta blocking your number right now I don't think this is a huge risk if you don't really overuse it uh and starts uh sending really irrelevant spam to thousands of people uh then probably you don't have to worry about this but of course it is a risk so what are my recommendations and take this with a grain of salt this is just my setup there must be way more setups uh this is just what worked for me right so for outbound only I have actually had clients like that that's actually what I do right I go without the official WhatsApp API why because clients can still keep their own number and we can do 100% personalized Outreach right and then they can just take over the conversation inside of the normal WhatsApp web or whatever chat management tool they use to actually respond to the customers if they respond right so that's my setup for outbound only for in only where we only respond to the customers uh I do like agentive right uh chat base also pretty good but I do prefer agentive because of the tools we can integrate uh so I would go with agentive but only in the case that you don't need human handoff right if you need human handoff then you can also use uh these these platforms very easily like naden make voice flow uh relevance whatever you want right because of course with a gen if we don't have that human hand off we don't have the the media interpretation Etc so you can set up inbound only also through naden make.com if you seen my last video I'll make sure to link it up here too I basically set up an inbound voice WhatsApp agent only through make.com for example and then for the complete solution with all the key features I mentioned that's where I use a combination of relevance a and make.com uh basically because through make.com we get all the Integrations necessary uh with the official WhatsApp API so we can uh get documents voice images uh everything we want and with C we basically get the front end to get uh the chat history and also the human handoff feature right so if you don't need the chat history in the you handoff again you can do it only make.com or NAD wherever you prefer but because relevant AI basically already has that sort of front end build in that's why I use that uh combination so I'm going to show you a breakdown of this setup and this setup because I think for outbound only there's many interesting use cases too especially if we can work with these 100% personalized messages and if you're interested in inbound only uh I also have a video on my uh Channel where I show an inbound sales agent that I build on agentive so if you're interested in that you can also check that one out but in this video I'll go over this one and this one so let me get to the the setups of what I'm going to show you today so the first one I'm going to show you is outbound only right without the official WhatsApp API uh now we can do this like I said only in relevance AI right and the big advantage of course here is we can keep our current number we can keep using Whatsapp web and we can do 100% personalized Outreach because we don't have to work with the pre pre-approved messages now we can with this setup actually also automate inbound right right inside relevance say we can all do this inside a relevance say normally it's not my preferred setup again because it's a little bit slower right it basically pulls every two minutes to check if there are new messages and it can't interpret uh voice image or document right but I'm going to show you this anyway because if you maybe not don't have uh a business right now and you still want to try this out then you can uh basically set up the entire system uh like this right and you can basically automate your own personal WhatsApp or any other WhatsApp you want to you want to automate uh so we're going to do this only with relevance Ai and then the second thing I'm going to show you is the complete setup that I usually use for clients that need sort of the inbound and outbound right with the official WhatsApp business AP I got to take the use case of a travel agency because it's what I recently uh built for a client uh so I sort of replicated it for this it's not entirely the same but you get a good idea right so for this first one we only need relevance AI if you're completely new to relevance AI it's a no code AI agent Builder I have many other tutorials on relevance too if you're interested or if this is going maybe a little bit too fast uh I have some more beginner friendly ones and then the second one is a combination of relevance Ai and m.com so let me first show you how you can build uh personalized Outreach through WhatsApp tool right in relevance AI uh now again I took this travel agency as an example right so basically just to give you some context what they wanted is they had all these past clients that booked holidays with them they basically wanted to re-engage them uh with a personalized message uh basically recommending a new promotion of a new destination they had and they were wanted to sort of uh recommend relevant ones uh so making sure that they're sort of similar to the previous ones they went to right and basically whenever then they respond right they've also wanted to uh so you can see here uh basically we have a Google Sheets with the name the past holiday and the number and from that we use a rag right a knowledge search and I'm going to show you this in detail to find a similar destination with a promotion right we're going to send that personalized message to the to the customer and as soon as they answer they also want it to um automate sort of the answering right so that's where our AI agent takes over right to actually respond to the customers and our AI agent there in relevance has access to five Tools in this case so we have the send WhatsApp message tool so it basically can send message is back right it has uh a deals database tool so basically a database full of all the special deals or all the deals the travel idency had available at the moment so it could basically get that information and send it back to the to the customer if they have questions or whatever a general knowledge base with uh questions and answers from the actual company general information about the company then we had tool four which is the human handoff tool which is built into relevance a AI right and it's basically what they call this delegate to manager tool right and basically we can instruct our agent when to delegate to a human right or to a manage or how do you call it and we can do that through slack and email and basically in this case we instructed our agent to uh delegate it to a human as soon as uh the agent identified that there was intent of booking right so human could take over and get them a call basically to to finalize the booking uh the second uh case was if if an answer was not found inside of the knowledge base right then the human could jump in answer a question and there therefore we also had tool five which is actually save an answer to the knowledge base so as soon as a human came in to answer a question because it was not available in the knowledge base then it will also use the safe answers to knowledge base to actually update the knowledge base and the next time a lead will have that question it will actually be able to answer it so it's a sort of self-training uh tool right so that's that was sort of the setup uh for for this now again for this use case our agent can't understand voice images or documents right so I'm also going to show you this setup which is my normal setup of course right and in this case we use a combination of make and relevance AI now why do we use make and make we we basically use to interpret these documents voice images Etc right and also because make has a direct web hook with WhatsApp so we don't have to wait and pull every two minutes whenever message comes in we get it directly in uh so we can actually respond as fast as possible so the way that works for the inbound handling of responses is make.com receives the message right through a web hook in the WhatsApp module right then we can identify what type of message it is and if it's an audio we go voice to text first right if it's a document we download the document we transcribe the document right if it's an image we we pass it through a an image uh L&M right that can actually see what it is or if it's a text message and based on that it will then be sent to our agent in relevance AI right so that's how we make sure that we can handle all of these different media types right and then the setup is very similar right and you could even in this case we didn't do it but you could even make it send back voice messages or images Etc uh you simply connect your send WhatsApp to also to make.com right now that's it U now for the for the personalized Outreach in this setup uh we need to work with pre-approved messages which I'm going to show you later right so let me first show you this tool and how it's set up in relevance AI so let me get to relevance AI uh right so here we have my personaliz WhatsApp generator Tool uh for the travel agency right now if you don't know what a tool is if you're new to this uh tool is basically a workflow or logic based uh automation where we can string together a few steps to automated process right now for this personalized WhatsApp generator tool basically uh this uh agency that had a list with data of their old customers who've taken a holiday with them right they basically had uh the price of the old holiday they had the old holiday destination the name of course the phone number and the type of holiday if it's like a beach holiday or a city city destination Etc and basically on these three data points like the destination the price and the holiday we're trying to find out of a database of their new deals the best sort of recommendations for these uh customers and that's how we sort of personalize the message so let me show you quickly how this is set up all right so we have basically five input Fields the ones that I showed you right price destination name phone number and type of holiday right and then what we do is a knowledge search right and in this case we're basically querying the knowledge base uh almost like a Google Search right and we're saying find similar holiday destinations too right and we give the name the name of the destination right the type of holiday right and the price right and our knowledge search will retrieve the most similar results to those so if you want to this is for example I just set up a dummy one right but we have the destination uh the disc count duration the price the link to book right uh and the type of holiday right so it goes through that and finds the most similar ones right so I'm going to show you I I'll show you in a second an example so basically this outputs some similar results right and we put that into one more language model step where we basically um try to get it down to the best one because here we get three results back I think yes three results here you can Define how many results you want back right in that extra AI step here and again you can read The Prompt in detail in your own time if you clone my template right but it basically finds the best one right we output that best one and then we're going to send uh a WhatsApp right now I actually send three to make it look as real as possible so the first one just says hi name right how are you right then we have a slight delay to not send them all exactly at the same time so we have a 8sec delay right and then we send the second message right it's baned from outlas Journeys right just wondering did you already book your holidays this year right again we do a little bit of a delay right 10 seconds and then we actually send the personalized message we have an amazing deal on right and the destination the best M matches their uh past holiday basically I think you'd love it after your you know and then we personalize that again with the last destination they went to let me know if you want to learn more right so let me show you how this works uh in action and then I'll show you how you can actually run this in bulk too right on a long list so let's say the price was $1,200 right the past holiday destination was uh let's say uh Paris right uh name let's do my name Ben put in my phone number and type of holiday uh CT right so we run the tool and you can see you found three results that are similar right it I put it one Vienna right now I should have received the first message so you can see yes the second already too and now I should get the personalized one in a few seconds there you go so we have an amazing deal on Vienna right now I think you'd love it after your Paris trip right let me know if you want to learn more right so that's how it works and now the cool thing is we can actually also uh automate the responses right so I'm going to show you that in a second but first I want to show you how you can actually run this in bulk so you can actually run it on a large list of people to do personalized Outreach for so uh you just go back here to use right you go to run in bulk right and here you could basically just upload a CSV uh with a column for each of the input fields and then you can just map them together right and run them on the whole spreadsheet so uh I don't have a spreadsheet with that right now but I can show you a quick example uh you an existing data set I just made right and here you can see I have the right let's say these are the columns which your CSV then you can basically just map it right so you can click generate with tool then you can map the input fields of your tool with The Columns of your CSV right so in this case price right destination name phone number where's the phone number and type of holiday right and there and then you run it in bulk on the entire uh list right so very useful uh feature I think especially for this WhatsApp outage now so let me show you that our agent can also respond right if someone answers and I'll also show you the human handoff feature now this agents actually connected to my own personal account right now so I'm going to change it to my other WhatsApp account because of course it won't get triggered if we send ourselves a message uh so I'll put in my other WhatsApp and then we can actually first send the Outreach message from here so you could basically click here right and it will run it on this row or if you click it on this you have a list it'll run it on all so you can see now it will send this from my own personal account you see right we're getting the messages and this is the nice thing of this right of course we can still jump into our own WhatsApp web to actually respond uh so I'll just respond back something like hey um I'm actually looking for a trip to Asia right so I just responded now this is the big Advantage right it's like of this sort of set up without the official WhatsApp I can still jump in here and actually respond to the client right now the downside is it only answers every one or two minutes because we only relevant C only checks for new messages every one or two minutes so you can see we got it back now and it did take 2 minutes and you can see I basically gave my agent access to the same uh data base right of the deals right so now I found some deals in Asia right Bangkok Thailand Kyoto SE right now uh let me show you that can also actually answer questions from the knowledge base right so it has another database with actually like an FAQ basically most common questions and answers uh so for example I can ask um you have discounts for groups I know that's in the in the knowledge base so here we got it back yes we have a 10% discount for groups of our 10 people if you're playing a trip with friends or family let me know and I can help you with the details right now let me show you the human handoff feature right so basically we instructed this agent to go for human handoff as soon as it doesn't have the information uh for a qu to answer a question inside of his knowledge base and also when uh the client shows High intent of booking a trip right because then basically the travel agency wanted to call them to help them set it up so for example I ask a question which is not in the knowledge base uh for example uh two trips uh do are the trips to Tokyo uh suitable for wheelchairs I don't know it's uh I'm sure this is not in the knowledge spacee so it should do human handoff and the human handoff basically you'll get notified on Slack and through email or whatever you can do both or one of those whatever you you prefer of course I have to send it for my other WhatsApp uh or I send it from the agent of course because I answered in here yes now did it right of course I send it to the client now from from hours so you can see we got a message here in slack now customer asked if trips to Tokyo suitable for real chairs right so then we get a button here basically where we can go to our agent right in relevance and you could go directly to that message right and then of course you can go directly into your WhatsApp web to actually answer the question or you can actually go in here and say Yes uh all trips to Tokyo um are uh wheelchair suitable right now you and basically the advantage of doing it here inside relevance is we have another tool here which is called save answer to knowledge so as soon as you can see here in the background it will save this answer to the knowledge base so the next time a customer sends that uh question the agent will actually know right and can answer it so you can see it responded back right yes all troops in Tokyo are wheelchair suitable if you have any qu questions or need more info just let me know right so that's how it works uh of course the advantage of this you know non-official WhatsApp API is that we still have access here to this WhatsApp web so we don't necessarily have to answer their inrev AI but for the WhatsApp business official WhatsApp business API that's very useful because we can't actually we don't have access to Whatsapp web right with an official WhatsApp business API so that's it and I'm going show you very quickly also if I show high intent for booking we also get notified uh right for human handoff so I can say something like okay great uh you know could you could you help me book The trp right and now again it will let us know through email and through uh through slack that's basically how it works and can be quite useful because you can basically just let this run automatically in the background and only when there's High intent of booking or there's any questions uh that that can be answered by the bot then you have to jump in quickly so you can see right again we got notified to our Asia trip request customers uh intent on booking trip right now and wants assistance right and again you can then just go through the relevant sayi thread of the customer find his phone number call in all right so that's how it works also remember right with the non-official WhatsApp business API setup um our agent won't be able to understand media files right so that's another downside of this setup but it can be quite powerful you can literally actually what I'm doing right now is automating my own personal WhatsApp uh which is pretty insane so let me show you now how this agent is set up and then uh after I do the agent setup and and I'll go over the tools then I'll show you how to actually set this up with the official what's business API and how you can also let besides all of this let your agent also understand media types like voice image and documents so I'll go over to agent settings very quickly and uh through the tools but again right the template will be available uh also in my community so if you want to check it out in more detail you can do that there so uh the agent profile right uh we have the name right WhatsApp travel agent non official WhatsApp business API now now here we have the Integrations right so here we basic have two options right we have the WhatsApp for business trigger so we can actually trigger it directly from relevance a with the official WhatsApp API but I'm not doing that in the next one I'm showing because then it won't interpret voice images documents Etc that's why I'm doing it through make uh but the second option here is the normal WhatsApp right which is my personal WhatsApp right and the way you can see mine is already connected right but the way you do that is basically here as you can see it's just like if you know WhatsApp right it's just like WhatsApp web right you just scan you link a device you scan this and it's connected so that's that's all you you have to do there um so that's it for the trigger right then we have the core instructions right uh I'll go over it very quickly um I use my agent a agent prompter tool usually uh to help me write them a little bit faster and quicker just check out my prompting resource you'll find it there so you're Ben an expert WhatsApp customer support agent for travel agency right with a particular act for blah blah blah right your goal is to help and uh and encourage people to book trips with our agency think step by step to uh and ensure you provide friendly helpful nonformal support right besides booking trips you also help resolve any questions they might have right now sop right think step by step through the following process initial trigger when a customer responds to one of our Outreach messages you are activated right giving it a little bit of context right now the first thing in this specific setup with the personal non-official WhatsApp API we actually have to sort of we can first retrieve the conversation history which is a tool which I'm going to show you later so the agent gets a little bit more context of the uh conversation that has already been going on or the recommendation in this case we already sent through the Outreach message then we uh always aim to chat travel deals and bookings to the customer right communicate with the customer answer questions right now here's the escalate the manager right which I'm also going to show you in a second how to set that up but here you can basically instruct it when to actually escalate it to us right so in this case always you always use the escalator manager tool in the following scenarios when you have run the knowledge retrieval tool and didn't get any useful information back from the knowledge base so that's the example of I said uh of my examp so so that's the one for my example where where I said uh are the Tokyo suitable for wheelchairs right in that case it tried to get the knowledge from the knowledge retrieval it didn't find an answer and therefore it started delegating it to us right uh the second is when the cost customer asked to talk to a human right when the customer uh intent on booking a trip right now right uh whenever you don't know how to handle question or inquiry from the customer and lastly when whenever you receive a message that you cannot yet read this usually is an image or an audio message right this is what I said before they can't understand that yet so in this case I also told it to um delegate it to us if they're sending that because basically you it will still pop up in the agent but he won't be able to uh do anything with that uh file basically and then lastly is the update to knowledge based tool right which I'm going to show you too use the save answer to knowledge tool when your manager has responded to a question that was not yet in the knowledge retrieval tool right so that was the example I said too right with the with the wheelchair that was then saved into the knowledge search so your agent can actually answer that questions question next time then we have the instructions right I toble down on that that rule for the escalator manager tool right again I give it some more uh context on the tools and an example or one or two examples in this case one example of an actual conversation with a client right and I double down on the important roles again in the note section so again you can check it out in detail uh in the right in the flow Builder I haven't used in this one right uh because not an extremely complex workflow uh it's more like a chat bot right so that's why didn't use the flow Builder now here it's important part in the abilities right and because here in the abilities we have the escalate to humans option right and here you can see allow your agent to escalate to you when it doesn't know how to act right so in this case you can choose both between email and slack so I did both you could just fill in your email here then uh connect to your slack and choose the channel you want to send the message to and then automatically it's basically added as a tool inside of your agent system right and you'll see that too then it will pop off if you use a slash here you'll see it basically appears as a tool inside of your prompt right so you can see I used it here escalator manager right uh and then uh you can see in the tools because it actually doesn't appear here right but it's still a tool it's just basically a buil-in tool from relevance AI now I've equipped this agent with uh five tools I'll go through them very quickly so first we have the retrieve chat history right so as soon as a uh customer responds we basically prompt our agent to always first use that retrieve chat history only the first time right because uh after that it won't need it because all the messages will come in the same thread I can I can show you it basically identify as you can see it's all in the same thread so we have a different thread for each of the WhatsApp chats right so for all the rest of the conversation you already have the context right because all the messages come in here so it's only for the first one so I'm going to show you quickly how that's set up so I'm going to actually show you through the example right so you can see I I uh responded to him hey I'm actually looking for a trip to Asia right and he first used the retrieve chat history right and basically gets the last five messages from the conversation so you can see hi Ben how are you right it's been from adlas Journey so it's the outre messages we sent right then we have the used travel deal search right of course because I asked for trips to Asia uh so let me show you first this tool in a little bit more detail so basically we can retrieve that through the chat ID right and the chat ID we automatically get uh as soon as a customer uh S A you see sends a a question right to this chat so you can see here at the end we have the chat ID so our agent knows how to fill out that chat ID right so that's our input field for this tool is the chat ID right you can see it filled that out correctly right and then we have a simple WhatsApp actions here and this is the key right so in relevance we basically have three options here for WhatsApp right so we have a sent follow-up message for WhatsApp business so this would be for the official WhatsApp business API this also send template message for the WhatsApp business API and this is for your personal WhatsApp right so if you select this one you can see I have multiple options you have to get all chats right you have get all messages from a chat right you have get user profile if you want to get their name for example uh or or their phone number right then you have send messages in a chat and start new chat now you can actually use both of these even if chat is already uh exist even if a chat already exists you could still Ed to start new chat and it will respond in the same chat so that's it so in this case I got uh the action got all get all messages from chat right and basically what we need there's the chat ID and then we get that information back right so if I run it you can show you can see It'll get back the last five messages right now then I have one more step which is an llm step that basically uh cleans this information up a little bit because we don't need all this information chat provider id Etc so all this and does please extract the last five messages in this conversation right and then it out puts it in a cleaner cleaner way as you can see right so we send clean data back to our agent which in general is good right we want to keep the life of our agent as easy as possible because he already has the plan he has lots of other responsibilities so we want always send clean data back to to our agents so that it for that first Tool uh then the second tool here is that knowledge base basically that I showed you before for the personalized Outreach too right where it can basically query a database full deals right to to find so in this case what is the customer looking for this query will be used in a vector search on database uh be specific and concise right so in this case Asia right it runs it through the knowledge right to actually find the deals we have available in Asia right so again I can run it quickly so you get an idea so again here and this General is really good uh for these um for these uh uh knowledge retrieval tools right is always have a step after the actual knowledge retrieval so in this case I uh in the settings I put you know to retrieve Five results right and then in this second step again I'm making sure that the results that I'm getting from that knowledge base actually match the query right so what I showed you before also in the personalized Outreach we're basically making sure that it actually matches what the original query of the user was right so you can see what it put out right is option one in this case they were all valid option right but sometimes one or two might not be or maybe the customer is asking for one specific uh City and it returns five then this step will make sure that it only uh uh outputs one right so again you can check it out in detail uh in the template right now we can check the the next one so this is the actual sending of the WhatsApp right I'm going to show you this one quickly again for sending a message back in a conversation we need to know the chat ID right which our agent again fills out right and of course the WhatsApp message to send to the customer right now I instructed my agent to be very uh brief concise uh in this message right because it's a WhatsApp message we don't want to have is really long AI texts right so be brief concise don't use white spaces or line breaks of special formatting do not use bold right talk like human avoid being overly formal this is a WhatsApp message it's supposed to be a personal Communication channel right so you can see it didn't use any bold Etc right so here's the message and then again we have that WhatsApp action actions module and in this case we have sent message in a chat now it ask for the chat ID right where I put in the variable and of course uh the text message which is this variable right and it sends that and uh yeah that's how and that's how it works so good let me go to uh the next one which would be the knowledge retrieval right so so this is another uh knowledge base basically with more General FAQs right so you can see uh do you have any questions or sorry do you have any discounts for groups right that's the query now what I've done here is I actually tried to optimize query before passing it into the knowledge search now why do we do that to actually it makes a huge difference how that query is formulated and sometimes customers ask a question in a way that is not EXT well optimized for actual uh search through a knowledge base right so very good to sort of put this L&M in front that optimizes the query for a knowledge search so you can see here the search query right I put that in and then users are asking for help about travel which uh which is a travel Agency for elderly this search query is free form text that we need to convert into an effective search query for fq database this means it should be concise and remove all F filler words that don't contribute to the search right so basically trying to optimize that query for this step the knowledge search right and you can see we pass on that variable inside of the actual knowledge search right so and then again right we make sure and this is also again really important we make sure that that answer actually matches uh uh the question original query of the user right and I'm going to show you this quickly because if it don't if it doesn't right we don't want to send that back to the to the customer right so basically what we instru instructed here in this last one is original search query right we pass in the original search query from the user that the revised one the one we we we optimized in that first step right information that might help which is the outcome of the vector search right so you can see the vector search brought back Five results right and some were not relevant right and based on the search query analyze the information that might help and extract anything that is relevant to the query if nothing is relevant please feel comfortable to return no useful information now this is really uh important right to instruct it to Output this why because with this we know our agent knows that we don't have this information available in the knowledge base right and therefore it will escalate it to a human then after right because if we don't do that and we just send this directly back to our agent it might just start responding it with some random answer here that we got back that actually doesn't match the query or the question of the user very well so you can see in this case it did find a relevant answer so yes we offer a 10% discount on groups over 10 people right so that's what it sends back then to the user but I can show you the example of where it wasn't right so in the next one right you can see in this case there was no useful information right so that was send back to the agent so the agent then knows okay this information is not available on the knowledge retrieval so we're going to delegate it to human right and you can see it did that right as escalated attach to you right now we then uh answer the question here right and then of course our agent's going to save that answer to our knowledge base right so next time it does that knowledge retrieval this information is actually in there right so I'm going to show you this quickly so what we pass in is the original question and the answer right those are the two input fields and those two input fields we then upload to our knowled set right as you can see so very useful right to you make your agent sort of train itself automatically uh that's it I think uh these are all the tools yes these are all the tools so this is how you set up uh the the agent for a personal WhatsApp right non-official WhatsApp now I'm going to show you how to set it up with the official WhatsApp API now before starting very quickly it's important to understand that if you want to work with official WhatsApp business API you do need to verify your or a business inside of MAA right so if you don't have a business right now it's probably not an option for you you can stick with the the other version um but if you do right then uh you have to set up uh the WhatsApp API first inside Mata and verify your business if you don't know how to do that or you don't have it yet I do explain it in detail in one my previous videos right uh check that when I I'll make sure to link it up here at the end I'll explain the step-by-step process of how to set off the WhatsApp business API so I'm not going to explain it in this one so assuming you have your business and you have your WhatsApp business API um I'm going to explain you the following things right how to set up outbound personalized Outreach through the official WhatsApp business API I'm going to explain then quickly how you can also do notifications and then I'm going to explain you how to automate the inbound right people who respond to those questions and this system is very similar right to the one I just showed you uh the biggest difference is now we can actually interpret voice images and documents so let me start with the outbound personalized Outreach now before starting very important to understand that through the WhatsApp uh the official WhatsApp business API basically meta wants to have some guard rails in place right to not send uh any irrelevant spam to customers so we can only do outbound messaging through pre-approved template messages so basically we have to fill in template messages uh send them to meta they have to approve them and then we can use them in our outbound messaging um now we can still work with variables I'm going to show you that in a second so we can still basically personalize them and then for inbound uh as soon as a customer texts us then we basically have a 24-hour window after that last message to respond to the customer using any message we want right we don't need pre-approved templates there so as soon as a customer interacts with us right then we can actually have the freedom to send whatever we want so that's why our agent here does have complete freedom to send whatever he wants it's only if we do the first Outreach uh right then we actually need the pre-approved uh template messages so I'll show you very quickly first how you set that up Ina so again assuming you already have your WhatsApp business API set up right you go to your business. facebook.com right and there you select your business portfolio and from there you click on the gear icon right from there you can go to your WhatsApp accounts right now multiple um but from there you can go to WhatsApp manager here right in summary and from the WhatsApp manager we can set uh we can basically uh set up these pre-approved template messages right so uh I'm going to show you here right message templates as you can see here you can click on manage templates you can see I have I made one quickly for this example right you can choose some they already have or you can create another one so you can click on create a template right you you have to choose which type of notification it is right which kind of message it is in this case I choose marketing and custom right then I go to next right you can name your template right travel [Music] notification now you want to avoid words like you know promotion or things like that cuz you might risk that uh MAA doesn't approve your uh template message but you do have quite it used to be a lot more strict now they give you a lot more uh wi wiggle room for also promotional messages but it's still it's very important to get decent uh reply rates to these messages because if you don't right uh then meta might uh in the longer term right might you might have a potential of meta blocking your account because they feel like you're sending spam right so we want to keep it relevant in general we want to sort of send it to people we've had contact with or Know About Us to uh to to keep our number or or quality rating high right so travel notification and then basically here you could just uh type in your your message right so in this case hey uh how are you right um uh I'm B right um I right and then we can put work with variables right same way as we do in uh Rel a with the double curly bracket right and that basically we can then later fill out right and personalize so I'm ban we have an interesting deal for let's say Tokyo I think trip right so like this we could still B personalize these messages quite a lot right so then you you have to give them an example of what you would actually fill out right so so they get an idea of how your actual message would look like but we can but you can cheat a little bit here right like you can you can say something that actually you can leave the variable for a promotion right while in the test one you actually you know don't you pretend like it's not a a promotion right and they do submit for review and usually right away they they approve it right and then uh basically it's going to be available right now let me show you how you set up the personalized Outreach so I did the personalized Outreach as you can see I set that one up in make.com now why because we the modules in make.com we directly have access to that uh WhatsApp business API here the whatsa business Cloud right and you can see basically very simple right um I have set up the Google Sheets just as an example right same sheet as I used before right my name the past destination online right what type of holiday orse the price and the phone number right and basically this so This make scenarios this is searching the rows in the sheet in this case 71 but it can go through an entire list right to to to do it in bulk right so I have my Trav agency client sheet name Etc right and then what I do is basic I pass it through that sort of same tool that I showed you before where we're trying to find a similar holiday destination right with relevance AI right now if you don't have that specific use case uh you don't need to do this with relevant CI you can also directly send it right uh with WhatsApp if you don't need to personalize it with the knowledge base Etc right but as I already had it set set up in relevance that's why I connected this if you don't know how to connect relevance AI tools um make sure to watch one of my beginner tutorials where I explain it in in detail right but again this template will be available uh also in my in my community so basically I connected this to uh I can show you quickly the find no agents here find travel deals so all this tool does right again right pass holiday price it passes in all these data points then we do a knowledge search and output a best matching s the destination right now we set that one back right and we then use that as a variable right so as you can see and I'm going to show you this one because I haven't set this one up yet so we go output there you go so I'm going to show you this so this is the relevance AI right for all we get from this tool is um the output of the the new destination we were going to recommend right and here we can then choose our message template right so in the WhatsApp module what's that business Cloud module right we have two options here which is to send a message where we have complete Freedom right but remember we can only use that one if a customer first interacted with us right so if it's outside of the 24-hour window then we can't use that one it will give an error so then we have to use in this case we have to use a template message right because we are the first one reaching out to the customer so we use that one again right if you don't know how to connect this to make.com I explain it in detail also in uh in the other video right how to connect make.com um uh to your WhatsApp business API so here we can of course select the sender right this will automatically be filled out right and the receiver now in this case the receiver right is in my spreadsheet right which is the phone number without Plus or you can't even use plus uh and then you can hear your your message template will automatically appear right so you can see I already did this before right so my message template already appears right and here you can see the message template right Hi how are you B from Outlast travels here was wondering if you have planned your holidays for this year already we have a great trip too M available right now I think you'd love it after your M trip let me know if you want to learn more right so very similar actually to that message that I used with a non-official WhatsApp API and then here we can fill out the variables right so in this case number one is the name right so I take that just from the spreadsheet then number two is the New Deal we want to recommend so I'll take that from the output of the relevance AI tool right so IP put this one as you can see the last time I run it it's VN right and the last one is their past holiday destination which I can also take from the spreadsheet right past destination so that's it right I already set it up in this one yes I already did so I can delete it but that's the way it works and now if I just turn it on right and run it once you'll see that I get that personalized message boom we just got it in you see hi Ben how are you Ben from Outlast travels here I was wondering if you have planned holidays uh we have a great trip to Vienna available right now I think you love it after your P trip so you can see this is how we can do in bulk also do personalized Outreach through the WhatsApp the official WhatsApp business API now you can imagine it's going to be very easy to also set up notifications right uh which is a big one for companies I've done it a lot for the clients which is you know payment confirmations status updates for customers Etc because we can basically just connect their CRM or their Stripe Right whenever stripe payment comes in right watch events we can then connect that to the WhatsApp business Cloud module right and send a temp uh a template message right for payment confirmation or whatever right but again right all these templates first have to be done inside have to be sent inside of meta right and you can see here you can actually track right the the statistics right you see you see message opened right blocked Etc message sent so that's why we don't want to send spam too much because if we get too many uh people blocking us after message like that Etc then our num is going to be in RIS right in inside of meta now let me show you the next part which is of course uh automating inbound right with our agent now our agent is very similar to the one I just showed you for the non-official whatsa business API the big difference is we are not triggering it directly from relevance we're triggering it from make to be able to process all these different uh uh media types so I've replicated the travel agent quickly but then for the official what's that business API so as you can see here I have the non-official one I just showed and this is the official now very similar right setup uh the big difference of course here I don't have the trigger with WhatsApp we're triggering it from make.com Scenario why because then we can actually transform these media types and send that data over to our uh to our agent now in this specific use case documents and images might not be very relevant voice I think is always relevant because people just uh respond to people through voice on WhatsApp but I think I just put it in there because I think documents and images for WhatsApp uh interpretation have a lot of different use case I literally know a startup that just raised a lot of funding that automate uh basically uh restaurants orders um because that's done a lot through WhatsApp here in Brazil uh with with images and documents and automatically extracting the data from from those documents and images and putting it into their CRM that's the startup and I just raised a lot of fund because of course it saves people a lot of time so there's a lot of use cases I think for those uh image and document interpretation now looks a bit complex um not that complex but uh I'm going to take you through and also the blueprints will be available uh in our community right so basically all we do here is in this first one it's a watch events from the WhatsApp business Cloud right so that basically triggers every time we receive a message onto that official WhatsApp business API now again if you want to learn how to set this up I explain it all in uh in the other video right so this one set up right uh it's basically a web hook right so every time we get a new message this is triggered right and then we basic have a router right whether it's a media or it's just a text message so let me just first show you a normal text message right so if it's a text message basically here we have a condition right that looks at the message type right message type uh and if it's text it goes through this route right and then this is basically our agent in relevance AI right so we set up that trigger from make.com so we send the message over to our agent now how do you do that I'm going to show you very quickly right in your agent you can here go to API right and from here you can set it up right so I put in the endpoint I put in the request body right I generate an API key right and and that's it I can go through it very quickly but again if you want to know this in detail I explain it in uh many of my other tutorials right so here I have the URL right which is the endpoint right the method is always post for these agents or Tools in relevance AI then the content type application Json is always the same then for the authorization you need to paste in your API key right and then below you have uh body type is raw uh content type Json right and there we pass in the request body right which again you can find here now this request body only has an agent ID so basically we know which agent to send this message to but in this case we also want to send each different conversation with each different user into uh uh one uh conversation thread right so our agent actually knows what's been happening in the past because if we don't do it if we pass in the request body like this it will start a new chat for every new message it received so we want to make sure that we send uh the messages from the same numbers to the same threads so how do we do that we add in one more header here which is the conversation ID and the conversation ID we can basically decide what what that is so we just take a unique number from the customer which in this case we can just take the phone number right we add that into the request body I can show you here right so I just added here another header right conversation ID and then I just passed in here the variable which is basically basically the number this is how we identify to send these messages these new messages into the same conversation thread so our agent has actually knowledge of what's been happening uh before right so that's it so from there it gets sent now one thing you might have noticed if you've set up my personal assistant for example is uh if this part I've added because what happens if our agent is still working this is what how relevant say I set up right if our agent is still working and it will receive another WhatsApp in that same thread in that same conversation then will give give us an error right because our agent is still working in the background so it won't be able to receive a new message and with WhatsApp it's very common that people send two messages very quickly after each other so what I added here is basically uh a polling request right that basically checks if our agent finished working right if he's ready to receive a new message yet right and now how did I set that up right it's a special endpoint of relevance AI right um there's a documentation page on this too right but again I'll share this right so you can do this and then uh we have that URL and then we want to add in the job info or the studio ID as you can see here and uh the job ID so it can identify If this job is finished now where did we get that from we get that from the output of that first module right so I just pass that in as you can see right we get the job info here right job ID and Studio ID so you just pass that those into this URL here then the method will be get right again you put in your auth authorization and that's it now in a scenario this will basically say if if our agent has completed his task or not right and based on the outcome of that we if it's complete we're just going to send it here to a set variable doesn't matter which module you put here this is just basically means the scenario is ended right so if it's complete we route it to here so you can see right if our polling trigger outputs complete it will finish the flow because then we can re uh start receiving new messages now let's say our agent is still working what it will then output is not equal to complete I don't know what it says but probably incomplete right and in this scenario and that's why I put a repeater here right it will actually go to a sleep module right and that sleep module will wait 10 seconds right and then we'll run this again right and basically it will check again right because it does the repeater again right it'll check this again checks if it it's complete if it's still not complete it will go through the Sleep module again right so basically wait and wait and wait until it's complete and then you finish this scenario so that's how we make sure that uh before it receives a new one it has always completed his current uh job right now important to actually make sure that works is to also add in here in the settings scenario settings sequential processing now what does that mean basically we're saying to make.com only process scenarios one after the other so only process uh a new scenario once the other one is finished now is this a perfect setup no um but uh it it works well enough and it avoids the errors of multiple messages right and it will still it will still then take that second message that the user sent and uh process that one too right but there is no ideal setup right now for WhatsApp yet and this works pretty well right so that's basically what it does here so I can go show you a quick example right so if we say um actually I again right I'm looking for to Italy all right now we probably we won't actually see it no we won't uh so I'll I'll show you the example later because I'm not I I didn't click on run once so it's just going to run in the background uh so you can see we got a message back I found a great great trat for Italy it's a 5 night Etc but I'll show you this in detail how this works in in make right now here the second option of course is Media right and with media what we need to do with WhatsApp business cloud is actually first download the media right very simple module right we I just pass in uh all the different types of Medias it can be right and it will it will download that then then I upload it to drive so we actually save that voice or document or image uh in our drive right and then I have another router which basically this one is for documents as you can see right message type document this one is for images and this one is for voice audio right now for voice what we do is create a a transcription right with whisper open AI module where we can basically transcribe uh Voice or audios then we pass it through a text parer just to clean up the result or the the the text before we pass it in to uh the request to our agent right and it sends that voice message then to the agent and he can then start responding it right and here it's basically the same setup as I explained before right we just make sure that he actually processed it before we start uh processing the next query of the user right here in the second option we have uh uh an option where we analyze images right so we actually look at what what's in the image extract all the text Etc again we pass it in a text bar to clean up the text for the API call we send it to our agent right and again we do the same process to make sure he finished process it right and here we have document right so we actually download the document then we convert it to a file uh right because we can't take it directly from the Google drive into uh extract data from PDF so I use dumpling here I think you could probably use something else too because it is a paid app but I really like dumpling it gives you sort of clean information back from PDFs or a lot of other uh sources uh so I have a p i generate a PDF here I extract the data from the PDF again I pass it through a text parer and then I send it over to the agent so for this specific use case these images and documents not very useful but I think many interesting potential use cases for document interpretation from uh document and image interpretation from WhatsApp so let me show you some quick examples so if I just click here on run once I first do text um do you have more options in Europe right as you can see right it sent it to the agent right and I was checking here I'll show you the output right it says in progress yeah right so that's why it went to The Sleep module now it was still thinking and now it repeated it right and you can see in the second one is also still in progress right so repeat it again and now was complete right after the fourth time and then it finished it right and you can see our our agent came with some other options in Europe right but uh that's how it works there in make.com now let me show you another example with a with a document right or whatever an image so let's just send a really random photo of my thumbnail completely unrelated but we're just going to just as an example right you can see now when the media are out it downloaded the media now it sends it to the image interpretation mod mod model and again it will send us send it to our agent and now we can actually see what our agent received it basically the description right the user send an is uh an image here's the description of the image the image shows a person smiling on the left the text reads AI agent SEO agenty Etc so basically describes what's in the image right so our our our agent has a better idea now again now of course now our WhatsApp agent transferred this or escalated this to us because we instructed our agent to escalate to manager as soon as it doesn't know what to do and obviously it's very unrelated image so uh that's why it's delegating it here to us right and again right here we can then tell our agent what to respond like thanks for the image I don't know doesn't make any sense but you get the idea that's how we have that human handoff with the official WhatsApp API which is at the moment very difficult to do with any other software so that's basically how how it works here and of course you can do voice too uh and documents here but again blueprint will be available so let me show you the agent quickly right uh very similar setup right again right we have this almost the same prompt right we have uh the abilities again right of course we have the escalate humans right in this case I only have it on slack uh and for the tools we have everything except for the retrieve chat history because we can't actually do that on this one but it's exactly the same as um the previous setup the only difference here is this tool which is to send WhatsApp to customer tool because in this case we're not doing it with our personal number of course we're sending it with our WhatsApp business API now I'm going to show you that one very quickly so in this case we're actually sending it over to make.com through a web hook again because we have that module there so in this case very important to understand too right the conversation ID of this conversation right so basically we saved each uh thread or conversation under the phone number of uh the user right now our agent knows always knows the conversation idea of each thread right so we already know that the conversation ID of this thread is going to be the phone number of the user so all I did here is tell my agent to fill out the conversation ID and he will fill out the phone number which we can use to send uh the WhatsApp to so that's what it would do right and WhatsApp message to send the customer right and then we have the API step where we actually use make.com to send it over now you do actually have a module here to in uh relevant sayi to send you can see you can send a follow-up message which in this case is the the case right because we're responding to a customer who reach out right or we can also do the template message now I'm doing it in make because it's a lot easier like you can see here with the template message we actually have to work with the Json object Etc uh it's a lot lot more difficult and with the follow-up message the problem is it gives it stock right it will send a message but it will get stock in here uh it's just a bug from from relevance AI so that's why I use make.com and also the reason is because we can actually send voice messages back which is pretty incredible to the to the user right so let me show you very quickly how this works right so WhatsApp right uh probably I have a few as you can see let me check if it's this one just by the web hook we can check it so it ends with mkt5 yeah that this this is it right so basically all it does here it receives that data the message and the phone number and it'll send it right so we have to send a message module here not to confuse with to send a template message because here we're actually just responding so we don't need to work with the template message so you can see in the send a message tool we just connect our our our uh phone number right and then we pass in um the phone which is basically the conversation idea which our agent filled out right and of course uh message type text in this case and message right so I'll show you again quick example hi there right and my phone number you can see it's send it now and it's send the message right so that's how it works and we can actually also do audio I can show you very quickly it's pretty cool so for example uh we put in uh first an 11 Labs right where I already clone my voice I can take any voice you want right from there right create a speech synthesis but then I can choose my own voice which I already trained I think this one right and we pass in the message the text right and then we actually instead of uh the text here we select voice right so we go audio uhuh we need one more module here which is the upload media right I already selected the 11 Labs one the sender ID we're going to choose right and then we're going to pass in here the media ID from the previous model we have to click okay here now the media ID will appear not yet click here okay now it appears right and so like this we can actually send uh voice messages back so let me again show you a quick example so let's say something like hi there uh my name is Ben uh how are you today I accepted right now you can see it's in the voice message now I don't think you're going to be able to hear this no but this in my voice basically that message uh which is pretty incredible so that's it uh again right all the templates will be available in my community uh if you got any value out of this uh I highly appreciate a like a comment and uh subscribe it took me a while uh to to film this video and uh thank you so much for watching if you have any qu questions let me know in the comments below I know there's a lot of information uh it's not the ideal setup yet of course because we're still sort of juggling between these two platforms but in my opinion it is the best setup uh we have right now to sort of have the complete solution right where we can interpret all these types of I we have the actual front end here in relevance AI we have the human handoff feature we can send voice messages can receive voice messages uh so that's it I've done the hard work here for you so I hope it helps uh thank you so much and uh have a great day hope to see you in the next one