Building a $5,000 AI Caller for SAAS Startups | Full Tutorial

hey guys today we'll be going through another AI caller use case this time it'll be an AI inbound AI caller for SAS startups and in this video I'll go over what it is um why it matters how you can build it and then we'll end off with a demo so on to the big picture of what this use case actually is so a quick overview is we are building a SAS customer service representative that can field calls 24/7 save money save time increase customer satisfaction and reduce term and so this is mainly targeted for SAS startups who may not have the money or willingness to pay entire team or out or Outsource their customer service um but also want more than just a simple chatbot or documentation page or a support email um for their company and they can actually have a phone number for their customers to call and get customer service from so here's how it does this if you see my other tutorials this will look familiar but we're using vapy which is a software that will handle the logic transcription and AI voice and then there's functions and in this specific build I made it pretty simple so that you can kind of get the broad view of this use case so we won't use any very complicated functions but for example that could be booking a meeting sending a transcript sending an email Etc and then data which is the information sent in the function so here's what you'll actually need to build it vapy like I mentioned for the creat of the actual Ai and the real guts and brain behind it make for the automations and sending end of call reports and that sort of data and then sheets for the actual data storage and now I would also like to mention that you can use any equivalent of these three if you would like to use sylow and zapier and Excel that also works but these will be the platforms I'm using in this specific tutorial so here I'm going to show you how this inbound assistant call Flow actually works so there'll be an incoming call and then vapy will say a first message which is kind of a generic hi welcome to X um how can we help you and then the customer will obviously respond to that and then vapy will detect which department the query is trying to access so if it's a technical question or a sales question or support question and then I'll use our custom knowledge based data which I'll get into in a second when we do the vapy tutorial um it'll use that custom data and then provide a response with that accurate data um that's specific to the SAS comp and then that will repeat until they're out of questions um and they either end the call or vapy ends the call then the call will end it'll retrieve the call data send the call data um to that Google sheet like I mentioned and then the flow has ended and so here's why it matters for you and the the big opportunity here is that SAS companies aren't perfect especially startups and so here's where it goes wrong SAS companies are constant constantly in development and they live off customer feedback um constantly throwing out demos betas seeing how people respond and reviews are really the lifeblood of these young sass startups but most of these companies don't have any phone support despite it being close to 60% of their customers preference um when first reaching out and you probably feel similar that you would rather talk to someone even if it's an AI over the phone versus talking to an AI chatbot or trying to search through all these documents yourself and here's why it matters because without phone support or even having phone support but it's inconsistent um not 24/7 SES companies lose out on valuables customer feedback customer Goodwill and important data on their customers preferences so now that we know the problem let's make the solution before diving deep into what we actually built um I just want to mention that the links to all these softwares are down below as well as a Google doc that will have links to the softwares once again as well as all the prompts I used um the make.com template and the Google sheet that I use so that you can create this for yourself um copy and paste all of it and then even add on top of it um as you want here's the AI assistant I created if you aren't familiar with vapy I'll walk through some of the key features so this cost is how much it costs per minute to run this AI assistant the latency is the average response time um so that's around 1 second and then here you really get into the meat of it the model is a AI model that is used to respond and so let me take you through um this first message in prompt real quick so you get an overview of what this AI caller actually does so for the first message it's just hey welcome to Meo so Miro is the software that I just used with the Whiteboard and all of that um so I just created a Mach AI demo the the one I built out here is kind of Representative of if that company were to have an AI caller um just so that I had a common company that I could use rep as representative of a SAS company and so obviously you can apply this to your own company or if you own an AI agency and want to build this for SAS companies um you just apply it to your clients's um software and information so here we describe the role of what this um AI caller does and we said they're incredibly experienced customer service rep for the software company Neo um their role is to answer calls for the relevant departments um depending on the customer's query and then resolve their query so their task which is a bit repetitive but is to converse with the customer um find out which of these departments best suit their needs technical sales or support and then use the information provided in the knowledge base to answer their question and then more on their task is to refer to the scripts provided as context for how each conversation may go um be very clear in how you speak explain things in a simple and Casual manner um so here's some like emotional prompting and basically talking to the AI and saying hey your task is super important and that actually makes it perform better so here's some context talk about the business what Miro does a bit and I kept it really simple here um let me replace this with just knowledge and if you did want to have a knowledge base that was more relevant to specifically context um you could plug it in here and specifically reference it in the prompt um but I didn't do that hey so in the original recording I forgot to show how I actually got the knowledge base so I just wanted to add that in right here and just a disclaimer this is a super rough um knowledge base but I'll just walk you through how I actually got it so what I did is I went to mirro.com which is a software I'm making the customer rep for and I scraped a couple of their pages so like their homepage um their price ing page and the like about us page stuff like that came out to like 23 different pages I just copy and pasted it all into this Google doc and then what I did is I went to chat gbt I pasted that document so I downloaded it and then pasted in chbt and then just prompted it to scrape it and create three separate um structured knowledge bases um for the AI caller that we're creating um one for sales support and Tech and and only include the information relevant to their own Department in each knowledge base and then um I do also like for it to put it in separate code windows so it's easier to copy and paste and so yeah that produced um these and then I asked it to give it a bit more meat um add some more depth to the knowledge bases and that came out to if I go here um these three different knowledge bases and if we look at the support one you can just look at C support issues some features um the type of credit cards they take um stuff like that and as you test out the bot for your own use case you'll be able to find where there's holes in the knowledge base and what FAQs um either the AI caller has a lot of trouble answering and you need more support for in the knowledge base or ones that are handled pretty easily and that you don't really need to adjust much but generally scraping websites and then creating uh structured knowledge bases um within chat BT is a great way to just get a rough draft of a knowledge base and get a demo of these AI callers out into the world whether um it's for your own company and you just want to test it out or if you do have an AI development agency like myself and you're creating a demo for a client just to show them um roughly how it would work um that's a fantastic way to do that so um back to the video and then short just what we do is help people collaborate digitally on our worldclass whiteboard Solutions and then here we have some examples of um short scripts so that the AI can recognize whether it's talk it's answering a sales question um a tech question or a more General customer support question and so here on the sales department script I do specifically mention to reference to the sales knowledge base and I forgot to go over this but we I did make a unique knowledge base for um each kind of branch of questions that the AI would get so one for tech one for support and one for sales and so by separating it like that and then specifically referencing to it in the prompt um the AI is better able to fish out information rather than having a 25 30 Page Long doc um that's just like a general purpose knowledge base and so in these scripts you can see it's pretty basic just customer High I'm interested in PCH purchasing a Miro subscription you can be like great we offer several um subscription options are you looking for personal team or Enterprise then they answer um with what they're looking for the AI will respond and help them get it set up and then it won't actually subscribe for them um that would be very complicated to have their bill and everything all within one phone call but it does know where to guide them and what pages on the website um to help them troubleshoot any issues with them subscribing and then here I have um function calls but there really isn't any functions I decided to do if you did want to include a function um at the end of these conversations then this is where you would input them for example if you had a function that would automatically subscribe someone say if you have free newsletter um for your SAS something that does not include billing um would be entirely possible within this AI caller and then lastly um I just have some notes and just um not to be repetitive just to note um all these kind of General scripts are the same whether it's Tech sales or support just for different use cases then these final notes um I just said for sales inquiries try and capture the product they're interested in P purchasing but not to be pushy um if they don't want to share that and then to ensure they follow the script closely we don't actually need that um it can kind of go off track cuz those are just examples this says if a customer asks a question deviating from the script um answer the question accurately then try and flow back to what the original purpose of them calling was and then um if a customer request to speak with a live Human for the call using the forwarding function and then I'll get to that in a second second but vapy does have a built-in function where you could input um say you do have like a Tex support team but they can't answer the volume of calls that you do get um it will only for the ones where people request to speak to a live human and then this last one is just respond in plain English with no numbers or formatting just because sometimes when I was testing it out I would get it talking instead of saying 2500 or 2,500 say 2500 um or it would go through and say Dash and like explain bullet points and stuff like that out loud um where you just wanted to talk in a more casual like humanlike tone and so publish just to save that then next on the transcriber I just use deep gram Nova 2 it's pretty common for the voice I use 11 lab Sarah I just find that 11 Labs has the best voices and Sarah's one that I like and then I didn't actually change anything over here functions I do not have any functions um I will enable actually the end call function that allows the assistant to actually end the call versus the customer always having to end the call and then here is um I'll have to block this out but here's where you input a foring phone number then onto Advance you do want to enable audio recording um this likely wouldn't be um on video at all so you don't need to enable video recording I left all this pipeline configuration the exact same typically these calls will be a bit longer actually I'm going to dial this down closer to 10 minutes just over that cuz I I highly doubt there would be a 30 minute support call and if there are um I don't have much trust that um the person's actually going to get their problem resolved and likely human would have to intervene at that point so messaging here we have this web hook um and I'll show you once we get to make.com what this is actually used for and then below this um we just have a voicemail message um you actually don't need an end call message and then um these I just left the exact same and then in this analysis tab is where we actually did some cool things summary um I didn't do it sometimes I like to say organize it in bullet points or what have you but that's complete personal preference of how you would like the transcript summarized um if you leave it empty it will still be summarized but not in any specific order um or structure and then this success evaluation is this is pretty cool because I said um there an expert call evaluator is how I prompted it but basically this will analyze the transcript and return true or false based on if the customer's problem has been resolved um so return true if the query has been resolved and false if the customer's problem has not been resolved um and then I did some more stuff with that in the Google sheet but basically this will let you know at a glance when looking at the data um what sort of problems the AI caller isn't handling and that will then allow you as an AI agency or um if you work with someone building an out like myself to go in and adjust for those uh use cases so that then is able to resolve those problems over the phone and then for structured data this I just had an extra little um data which basically identifies the department whether it was um the sales Tech or support department that was queried and in the Google sheet we're actually going to track um where the most either complaints or calls are regarding so if you have a bunch of sales questions then perhaps your messaging on your website or in your ads isn't as clear as you would like it to be and therefore you're having a lot more questions about that from customers um so that'll basically allow us to track that so how I made this was I just added a property named it Department uh the description is just this is the the department that the call is routed to then the prompt for the structure data is just Department function um you'll extract the data of which department the call is related to um it could be one of three options either sales Tech or support like I just explained now that we're all done with vapy uh let's go over to make.com and I'll show you the very simple function that made to organize the data so here we are in make and this is a super simple function it's just a web hook um with a router that makes sure only the end of call report data is sent to the Google sheet um and that no function calls are called and so basically what you this web hook does think of it like something that detects whenever an action is taken on a website so basically this web hook which you will create by hitting add and then copy the address to your clip board will be pasted within in vapy in this Advanced section this is a URL vapy will use to communicate for retrieving context function calling and end of call reports so basically this is the phone line between vapy and make.com which will send the data and what this router does is basically filters out any function calls or data that we don't want and only keeps the data that we do want which is the end of call report which we then store in a Google sheet and how you get this Google sheet is um you configure this by connecting your Google account here and then you connect the Google sheet ID which is just the name of your Google sheet and then you choose the sheet name the rest of all these values will be automatically filled in when you copy the make template that I made in the Google Docs down below and once again I just like to mention that if you would like me to build an assistant like this for you um or even one much more complicated than this then I'm more than happy to work with you and just book a call down below using my booking link um it should be right next to the resources below and if you do want to build this for yourself I'm always happy to um help you if you need to troubleshoot any issues and so back to this that very simply is all we do within make is that whenever a call is made it will notify this web hook and then send the end of Call report to a Google sheet now I'll go over the Google sheet right now because that's a bit more complicated than the make.com function so here we are within the Google sheet and you can see um I sort of made more of a dashboard than simply storing data um and it has a couple of cool functions so first it will store the date and time of the call the from number the summary of the call the transcript so that's the raw um transcript versus the summary is a summarized version of the transcript as you could expect a recording of the call that true false of whether the customer's query was resolved or not is stored in here and then I made it so that if it's false it will turn red and that if it's true um or if it was resolved and true then it's green and then it'll also sort the department whether it's sales technical or support and then from these two columns I also made a count of false versus um true like resolve versus unresolved um queries and the percent of calls by Department um and one other thing of note is that I also created an extension through the app script every time there's another false um value which is basically an unresolved query it'll automatically move it to the top because those will require more attention from the company and their devs and customer support versus you don't really need to pay much attention to the um ones that have been resolved because although they may be issues um they have been solved and have been proven to be solved by the AI caller when you copy this Google sheet that script should copy over if it does not um I will leave the code that I used to make it um down in the resource document down below and you can just copy and paste that and within make you can ad just what data you get so this call info block I actually deleted but say if you wanted to track the reason the call was ended so whether that's the customer or vapy ending the call or it just disconnected you can do that if you want to collect more information from the customer you can make extra rows in here and basically what I'm trying to say is this is not a comprehensive build um in the slightest this is meant to show you the power of vapy um as well as make and Google Sheets when you combine them all with this voice AI architecture and a really useful use case um for SAS companies where before it might not have been affordable or reasonable for them to have customer support um phone lines and people helping out whereas now they can field massive amounts of queries for a pretty inexpensive cost if you look 18 cents per minute especially compared to minimum wage is is not a whole lot of money especially considering that it works 24/7 it never forgets any information um it's incredibly consistent and so let's actually get into a demo of what this caller sounds like and you can see firsthand what a conversation between a customer and this AI would actually look like all aose AI demo hi welcome to Meo what can I help you with to today hey um my mirror board won't load for some reason I'm sorry to hear that you're experiencing issues with your mirror board can you please provide me with your name and a bit more detail about the specific issue you're encountering iide you through some troubleshooting steps to resolve the loading problem thank you uh my name's Tommy and um I don't know my power went out earlier but everything's back I'm on Wi-Fi but my mirror board just won't load any of the changes I made thanks for the details Tommy since your power went out earlier let's start with some basic troubleshooting one make sure your internet connection is stable two clear your browser cach and cookies three ensure your browser is up to date four try accessing miror from a different browser or device can you try these steps and let me know if the issue persists yeah I just restarted my Wi-Fi and it um the mirror board loaded thank you that's great to hear Tommy I'm glad the issue is resolved if you have any other questions or need further assistance feel free to let me know have a great day you too so there we kind of role played uh what a successful call will look like or would look like and so now let's actually look and see if this data has transferred over so um the Google sheet for some reason isn't working now so let me kind of troubleshoot this problem oh okay and so um I just troubleshoot it I accidentally did the formatting so that it only started at F3 if it was true um so now they're both the same and so let me do another demo call where um it doesn't actually solve the problem and I'll show you how the falses will actually get bummed to the top because of the script I wrote Paul aose AI demo hi welcome to Meo what can I help you with today hi I'm trying to cancel my subscription how can I do that I can help you with that can you please provide me with your name and the details of your subscription plan yeah my name's Tommy I had a team subscription hi Tommy thanks for providing your name to cancel your team's subscription I'll guide you through the process you can cancel it from the Miro Dashboard please log into your Miro account go to the billing section and there you will find the option to cancel your subscription if you face any issues or need further assistance please let me know hey I I'm sorry it it just didn't work I have to go and so now that we did one that doesn't work um we can see it update here okay so the false isn't moving to the top but once again that came from me deleting one of these columns so I'm going to adjust the script um real quick so that it's um in reference to column f and g if you want to adjust the script I'll actually show you guys this part in case you guys want to adjust this script um for yourself so here's a little bit of code I wrote um so it'll actually should be column five for column F then is there six anywhere else in here nope so let me try running this and boom um as you can see I just had to make that small adjustment um and all of these false values move to the top and once again I just had some placeholder ones like these false ones and these two true ones on so you can see how these graphs work and um like I mentioned these update in real time so you can see as we've gotten more queries that these have both been updating in real time you can see that all the false queries have gone above the true ones even if their Tim stamp is different and so that should be it for this tutorial once again if you'd like to work with me um use the link below to book a a quick 15-minute call and we can chat as well as um all the resources and software I used will be linked down below as well as on the resource doc where all the information I used in this build will be located and thank you so much for watching make sure to subscribe if you want to see more builds like this

Channel: Tommy Chryst | Voice AI

Views: 357

Keywords: automation agency, how to create ai chatbots, how to create an ai caller, vapi ai tutorial, ai tools, ai automation agency, ai, voice ai tutorial, how to get leads with ai, vapi ai, how to sell ai chatbot, how to make ai chatbots, full ai caller tutorial, ai automation, vapi voice ai, add gpts to website, how to use ai to make money, how to call with ai, ai sales assistant, how to, bland ai, ai voice clone tutorial, ai assistants, how to start an ai automation agency, AAA

Id: pr-YJxRrFJ4

Length: 26min 41sec (1601 seconds)

Published: Sat Jul 13 2024