— So, let's start with the customer experience agent, right? One thing I want to really clarify in the beginning is that this is not customer support because I've seen so many like tutorials or like creatives like talking about customer support, right? It's a big use case. But there's a difference. Customer support is being reactive. It's when things have gone wrong. When the customers want to, you know, they need help with something or they just have something to complain or they want to troubleshoot something, right? Whereas customer experience agent is being proactive. is trying really hard to really keep the customers within the customer journey without them basically just leaving halfway through. It's actually that's the major like use case. If you literally imagine a business, let's say a pet grooming company, right? This is actually based on a real case study which I will dive a little bit more in details as well. But for now, just imagine there's a person called Sarah in that example and she has a dog called Bobby, right? And let's say she wants to arrange the dog for a dog grooming company. What Ruby does is basically they have a company that drive vans around the country and they try to basically try to clean and wash out the dog essentially and let's say she's a member of that company and she wants to book an appointment for Bobby right so she sent a text message through SMS to the company I want to book a another session for Bobby now the problem is they are quite a small team so they basically are member a team of five to six members and one of them is on the phone all day dealing with conversations s were basically 100 plus different members at the same time with basically the same questions. Now, we're all human, right? Like we work like 8 hours a day. There's only so many conversations you can have before you basically just you know what it's too much for me, right? And this is exactly what's happening with one of my clients which is a packing company. It's like okay I just it's way too much information like too way too many conversations to handle and a lot of the time they found that the response quality actually drop and they actually can just do something like oh just fill in this form right and there waiting time it's not great like the experience is just not great that was kind of the picture now imagine you have another agent that can basically personalize the conversations to a point now when Sarah test texted the company I want to book another session for Bobby it will say hi Sarah How's Bobby doing with his anxiety? Right? Because the assumption is that sometimes with some of the BS, they usually have some sort of anxiety issues or they sensitive areas that basically needs to be a bit more careful with during cleaning. So, it will say that if there's an issue, right, it will say, "Hey, how's doing with this anxiety? " I can see from his last session really well. — This is so interesting that you've gone like I mean the platforms don't really offer this stuff right now. Like I think HubSpot's been trying to do it saying that oh you can have full context of the past interactions with your customers but seeing this actually implemented on a small business scale is uh is very interesting about how you can have these kind of persistent memory across different conversations um with the same customer. — Yeah, exactly. And that's actually what customers want, right? Remember the business is a premium brand. So like you don't want to have a feeling of like just oh you're just another number, right? You're just number one out of like 100 customers, right? We want them to feel valued and that's where to make them feel valuable is by remembering little things like the anxiety or like oh would you like to book his usual food grooming for next Tuesday morning right that's being attentively saying hey my my customers actually boo like his bad for grooming every Tuesday it will remember that that's where the customer a good customer experience should have so maybe I can generalize that a little bit um into maybe like what kind of businesses with womb. — Yeah. — Basically small B2C's company like business to c consumers company that's kind of drowning in customer communications right usually again they usually have a small team that's handling a huge volume of conversations and they usually have their service businesses that have very strict like brand voice requests. Again the reason why is because they want to be premium. That is why as an example for this pet company they have very strict language guide. As an example, we literally have like a 20 page like voice system which you can actually I have shown kind of examples of it in the guide but just to give it a snapshot of it. There are few keywords that the agent should be able to say for example instead of customer client you should say member instead of owner you should say pet — like this level of detail is what really business is looking for to keep that premium feeling and keep that good customer experience. And the problem is that chatbots are give us such a bad vibe of doing that. Like it give off such bad like vibe where it just keeps on saying like the same stuff over and over again. It's it feels really cheap. It feels really general. Right? So that's the challenge. How can we automate that experience? Right? That's the use voice for this. And maybe I can show a walk down a little bit how we kind of break down that um tech architecture. I think um you have some data up there on uh on the CX use case. I think this is just a really interesting thing. This is like the main takeaway um for this section guys is that uh improving the customer experience is one of the most like validated ways that AI is providing an ROI or making a meaningful difference in the business. And I think this is a huge takeaway for you guys. We need to be using AI in areas that are natural to it and help you know you're not forcing a fish to climb a tree kind of thing, right? And I think sometimes people try to get AI to do these crazy tasks that uh it's not really that well suited for and in this case the improved customer experience is showing in the data that this is where it's really providing a meaningful difference to businesses and it's like the fast response times it's more personalized messages it's better availability and these sort of external or like customerf facing systems um that takes it beyond just that customer support to being uh proactive I think is a really important thing for you guys to look at uh when you're coming to maybe niching down your agency or or working with these kind of systems like you are Edin. So really interesting to me that uh these I mean these are going to be everywhere in a few years right always so interesting for me to see as you guys will see in a little bit how Edwin's built these but when you notice that there's some major thing like this okay well in a few years every business is going to have this but right now a guy like Edwin is having to use voice flow and air table and like cobble together all this stuff it shows that we're still so freaking early in this that there's not a centralized way of doing this — by the way this I think this came out like maybe last week or last two weeks I can't remember exactly but this came up very recently and I didn't obviously I didn't know about this beforehand but like a lot of the report by the way I recommend actually having a read on the report and yeah basically this basically confirms a lot of what I kind of experienced in the market right so as you can see the top gen AI impacts across business area the second use case customer experience this is just behind productivity how crazy is that right far ahead of marketing which is what you'd think yeah — you'd think that would be like that's what everyone's trying to use it for Yeah. But like actually there's like these three at the top. Yeah, for sure. — Um maybe now I can break it down how we think about building this kind of system a little bit. This is the overall architecture. So we have the incoming message. Let's say hey just want to know where my dog's next appointment is. All these conversations are through SMS and we they use the client use over which is basically a Twilio variant and it passes through the message passes through to voice flow and voice flow acts as the braid. Now what I mean by that is two things. The first thing is persona. So basically giving the LM an identity. So you might know you might notice from prompting we always say something called identity prompting right? You are a uh customer experience agent right? But we take that a step further. We give them a name, a backstory, what it's trying to do, how many years of experience it has, right? And try to just be as specific as possible. Try to make it as almost like a character essentially. And yeah, implementing the brand voice within the persona. — So that's that doc that you mentioned before, right? — Yeah. Exactly. Right. And we have the conversational flow. We have a flow for new member which are the new customers and we have the existing member which are the majority of the use cases. And most of the intents of the conversations usually have two things. It's like general questions or appointments, right? And the way that we can actually have that personalization is have all the customers and pub profile data saved in a database called Zeno. Now Zeno um is a database like to store all this like pub and customer data. And the way we do that is we retrieve that entire like information about the customers. their name, the email, this is the basic stuff, the address, but the more important stuff I guess is the dogs themselves like the when's their birthday. Like we literally have systems like if it's your birthday, say happy birthday to the dog, right? — Yeah. — To for example. So like and we have like the sensitive areas is like okay the sensitive areas I said before is like the face muscle. So we the agent should be able to take all of that information and draft a personalized response to the user. That's what we meant by personalization, right? — That's awesome, man. Yeah. And um and for the questions because I know another thing is the customers have a lot of question answer pairs. So for example they have a set of predefined questions and brand aligned answers that they want the agent to copy exactly as like a template essentially for the LM to generate a response. So the way that we do that is rather than like using a knowledge base right which if you know how rag works is like basically how to retrieve things semantically we actually wanted to retrieve this exact phrase like 100% of the time how we do that is we basically have each questions we give them a question ID right and based on the question ID like we semantically search for the question ID and then it search goes through the air table retrieve based on the question ID retrieve the questions and the answers and all relevant information like instructions etc. So it gets a little bit more complicated than that. But then the core idea is it pulls back all the information back to voice flow which basically has is this super brain that has all the information that he needs to generate response. — I'm going to call you on that on uh technicality here. Do you think it's necessary to have a retrieval system for a relatively small set of question answers there? Could you just stuff like the tokens are cheap nowadays? There's a huge amount of content you can fit in. So, where do you draw the line of like going to the rag system? Because there's a question of like is rag dead because of these massive models that are so cheap now. Um, is it a latency thing that you're trying to solve for? Like what is the difference here? — I just think that there's a misconception that rag means factor search which is not true like — maybe our mind of the dummies needs to have that explained like what do you mean? — It's basically um it's rag retrieval augmented generation. Yeah, I guess it doesn't have anything specific to a vector store, right? It's retrieval augmented. It's like context engineering is I think the buzz word they call it these days, right? — Yeah, exactly. Like what vector search like rack initially like basically to give you some context where the vector search comes from is when rag was becoming a thing a lot of basically a lot of the big vector stores company like pine cone um chor etc right they really try to push that product out and suddenly like everyone is like talking about vector it's like oh semantic search is the best thing right it's like these are things that are semantically similar to each other it means the retrieve accuracy is bad must be better but that's not necessarily the case what we're really trying to solve is like a search problem. It's like basically matching keywords, just searching for the on the user. — Yeah, like cool — the answer, right? So, yeah. So, rack is not necessarily just like back to search. And two, where do I draw the light? Well, because we're trying to build something like that's actually scalable, right? So, right now it has over I think 100 different questions. — Okay. I was thinking you just had like a little 20 Q& A — on this though. That's obviously a lot of work for the client to do. So like I get if they've got a quick F FAQ on their site like how do you how have you found this when you're asking them hey we need now we need a 100 question and answer pairs from you um do they push back on that at all? Do you have an AI solution to it where they can just like upload their knowledge what would have been the knowledge base and then you're converting that with a basic automation to convert it to like a com CSV file that you can imple import into here. — You know what the good thing is I'm very lucky they already have that set up in a Google sheets. — Okay. I guess that's for the existing support team, right? You can just probably find the existing support docs if they've got a support team, which you'd hope they do. — Exactly. Because But what they literally do, they literally just go to the question. Okay. The guy basically the person who's texting the customers. Okay, this is the answer. Okay, I need to copy these answers. — Okay, so you've just automated their existing support process here, which is uh which is great. I think uh SMS is a channel. Um it's always been an issue of what's the interface. It's like, how do you have these fully integrated end to end like customer experience things? Um, it's not like you're going to, hey, download our like dog grooming app. Like, you don't really want to force people to do that. Um, what you're going to send them to a website, then they're on the browser on their phone and they're going through this interaction. — It needs to be either like through voice or through SMS or WhatsApp. And those are really the best entry points people have. But then obviously SMS is sometimes like much more a US-based thing. WhatsApp's probably a bit more common for the rest of the world. Um, so it's interesting that you've I mean I think that these channels are are essential if you can have these sort of fully contextualized conversations across these channels. It's really powerful businesses and they're all going to have it soon. — Yeah. And I think the key thing is we want the least again we're back to the customer journey like customer experience side right like the way that you want them to build them is like what's the least resistant path for the customers right as you said I don't want to go to the website into another app like SMS just feels natural as well as WhatsApp right — yeah I mean if you could just make phenomenal SMS experiences that are like this for businesses like you are on some built right over a gold mine — yeah for sure — um if you just want to show just that uh quick glance at that voice flow thing there. Um people can pause if they want. You guys have access to this board if you want to uh have a cruise around. Um but that's an example of what it looks like when you're setting these up on voice flow. Obviously a lot more uh logic based. Um but I assume there's aic parts within that. We can jump onto the next one to keep this moving. Another example for you guys is um another case that I have is for two way like basically this link with the board will be in the description so you guys can actually see for yourself if you want essentially this is a more customcoded solutions like using Python um Bangraph and Pantic AI which is more customcoded frameworks and it's just another kind of a more advanced example
of what you can actually push in terms of building these conversational systems and the second agent really want to talk about is the revenue I call it revenue recovery agent right like before I kind of jump into like all the techn technicalities. Let me just try to again I use Sarah again as another example and this is a very cool use case is because this company is quite unique in what it does. is an education consulting company and what they do is there are many students who want to study abroad in the UK right and they don't really have an idea of okay what schools do should I go to now the use case is when the students go to do these kind of consultations and also do the admission test that data being saved somewhere in the database but a lot of the times the CEO found that none of them actually got back to them like they took their stuff they took all the good stuff and then they just don't want to proceed like it's like oh I'm not too actually too serious about studying in the UK or like I've gone to another competitor who's cheaper whatever right whatever reason and they're just — okay so what they're doing an admission test submitting it to them or they they're getting sent the resources and then they have to do it on their end like what would they why would they come back to is like at the end of that admission or consultation like hey let us know within let us know soon if you want to move forward is that kind of they just get left on that — yeah so it's actually an inerson test that you have to do so like — okay — yeah so it's like an internal thing within the com within the education company where based on the schools of the test they can kind of see where the level they are at and they can recommend right schools for them to go to essentially — gota — and yeah a lot of the students take test and they never apply now one thing I need to kind of get out is like kind of the architecture right now is kind of not very systematic in the sense that there's a lot of students almost like 300 students to handle for each consultant which is a lot now 300 including students who hasn't got back to them like all the different cases right it's a lot and it's a nightmare And they don't simply they don't have the time to follow up and organization to follow up on those students, right? And what we come ahead and said, hey, what if there is some sort of followup in terms of agents that can just automatically send them a WhatsApp message that can do some sort of follow up. For example, if they don't respond after 3 weeks, you can send them a WhatsApp message. Bases basically saying, "Hey Sarah, from the taste test that you got, you got 90% maths and 67% English, right? Here's a full breakdown of your performance and the suggested improvements, right? This how we think that you can improve. Let me know if you have any questions, right? Something like that just to give some value like basically some sort of like um automations that based on these test results and then if they still don't respond, — you basically have another follow-up sequence which is basically saying, — "Oh, hey Sarah, would we love to connect you with Tim, which is another student who has the same scores as you do, right? " He double down in pro improving English with and ends up being in Harrow, which is a if you didn't know, he's a top UK. So, so this actually gives them incentive to reach back to us and like try to finish the customer journey. Again, — you know how I'm dividing kind of these all these three agents, but they're all linked together, which I'm going to talk a little bit in the end. — Yeah. I mean, just looking at that there as well, the um don't underestimate how like if someone sends you a message like that sounds like they've taken the time to write it to you, in this case, you personalize it based off the test results and you maybe signed it like thanks from like Jane or something or like Tim at the bottom. And then you follow up with that one again, which again is very specific, like, hey, it looks like you had a weak thing when it comes to English. I can connect you with this. Like it's a pretty good offer, an interesting thing. And again, it sounds like it's coming from a real person. When you have messages like that, like there's a you can kind of play on human nature and what how we've been programmed after so many years is that ignoring people like that, we don't really like it. Like we feel like a bad person for ignoring someone who's taken the time. So there's a like a reciprocity thing, especially when they've gone out of their way to connect you with something uh with someone like in that second message you have. Um, so I think when this sounds like the illusion of it being a real person and through a very personal channel like WhatsApp, I don't think many people are really going to be just flat out ignoring a very personalized and a message that comes from seemingly a human account. Um, so I'd say this is extremely effective um at re-engaging these people. — Yeah. And it's really getting the attention of the students as well, right? Or the parents in this case. It's actually a lot of the time actually parents because they're quite young relatively and the parents like really you know like the parents are really like serious about their scores right they want to know how to do right so that's another thing good for them it's like what we learn is that this is really effective for businesses with kind of a complex customer journey and they have a kind of a high drop off it during a customer journey and usually that can involve businesses with you know kind of sales processes that have kind of a lot of different steps um that can not just educational services right maybe some high ticket purchases that kind of needs them time to think about, right? And the highest ticket items they are like the people are going to take more time to consider it and you need to send more follow-ups on them, right? And — filling in those gaps — in the sales process for sure. — Yeah, those time gaps — and sometimes you just have leads sitting in CRM. It happens to the guy like they're literally people like their data like leads in the database that just sitting there. No one has like all the following up with them. It's just such a wasted opportunity as well. and — and they might have spent money on ads to get those leads. you know, it's like they literally sitting like $50 $100 in the hole per lead and it's just sitting there doing nothing or you like they've come off ads and then like they've fallen through the cracks and they don't even have a system in the on the front end like in the near term if they fall through um what to do with them like how you can solve with this — and I think the biggest pro the bigger problem is for this specific case study is like not having a very organized way of managing those leads which is where CRM comes from we actually introduced into CRM So that's like that's actually like a bonus I guess. I would say like for this specific use case like around 70 to 75% actually drop off like after test as I mentioned and there was no lead tracking and everything. So that's why we're like okay can we do some sort of automation around follow-ups and the reactivation sequence so the students just don't just ghost you like right and that's where we kind of think about setting this solution combining high level which is a common CRM and voice flow. So maybe I can break it down a little bit again what it looks like. So on a high level no pun intended um it basically there were different opportunities. So if you have used high level before that basically that's how you manage different stages of the leads. As an example there's a pipeline we set up which is called as test. Well what that means is people who have taken the test but they've just kind of just ghosted like you guys. for example more than two weeks they haven't you haven't heard back from them and there are different stages like for example they have no valid WhatsApp numbers to begin with or you've message them no replies finished interested et goes it goes on but basically what this does is you're basically organizing the leads very systematically right across different stages and the beautiful thing about high level is that you can actually set up automation workflows inside them so essentially we have the students who you know who their flags were you more than two weeks that we haven't contacted and because in high level you can actually save information like this right is for example with custom fields you can have very specific information about students the department like education course IGC GCSE is like a UK like course so there's all these different information but basically again the key to personalization is details and data the more specific data you can collect from the leads the more kind of ammo you can use like to craft this personalized message right because you have more data both so this is where the dynamic personalization comes from the custom fields of the CRM again pushes through the voice flow as the brain and the brain will send the first DM through WhatsApp now there's two cases if they reply cool right they probably like they're still interested or they're not interested and if they're not interested they want to be out like it's a compet it's basically competitive server I'm not going to name them, but then regardless if the results are good or not, like they're interested or not interested, they're going to push back to high level and say the results. If there's no replies after 24 hours, there's still no replies. Then we then set up like we basically go through the follow-up sequence, sending a second, third personalized message again through the audio. So, it's sort of jumping back and forth between a high level triggering voice flow and then waiting like you've got a trigger or yeah, I guess a 24-hour trigger back on high level again to kick off another like round of personalization and stuff. So, yeah, I think this orchestration is where it gets trickiest. uh especially um when you have I don't know maybe they've been talking back and if it's conversational it's asking questions back and forth when is that conversation considered dead or like stale and it needs like when is like okay now that's technically a conversation done we need to like re-engage them with you know it gets quite tricky when you have uh say like Instagram DMs is a is an interesting one when you have like maybe appointment setting going on in DMs and within that thread it can get quite messy about like when does the agent think that that's a stop um when does the follow-up message go out. Um so it's interesting to know how you're managing that with voice flow across sessions. — Yeah. So basically um the good thing about voice flow is that they have something called like — session IDs. Yeah. — Yeah. Exactly. So there's a dialog manager API which you can call through the specific session. So each phone number is the session ID. Gotcha. So that's how you can uniquely identify each user and once you start like you use a basically within the voice flow dialog management API once you initiate a conversation it stays within that session. You don't have to again you just have to kind of do the back and forth conversation. Now where does it happen? Right? That's a really good question and usually there are like people usually there three types which are really serious like they just say oh sorry I haven't got back to you. That's that one thing right or one thing is like oh sorry I just it just you know my students are just not ready yet for the studying UK which is very common reason. Second reason is I've gone to like I've gone to someone else, right? — And these are just like hard though. They're just like and then we will just end up wrap up the conversation be like no problem like let us know if we can like help you anything help you in the future, right? That's it. But there are sometimes conversations where like yeah I just need more time to think about it right now. There's two like situation where we do is that we actually flag it to one of the consultant if that's the case and we actually have the consult to call them because this is the moment where like you should like the human should be doing the work now like this is like it's now or never they're going to be gone like it's like a sales issue now. It's like can you create urgency and can you push them over the edge and that human in the loop aspect for these systems. I know that's probably the two issues. It's like how are you orchestrating it and making sure that like the the sessions and threads are not kind of overlapping or there's issues with um like not being contextualized about what just happened and then it's like okay how do I handle the human handoff when it needs to be done and that's always tricky because there's the back end right like I guess you're going in through the high level kind of conversation manager to be able to jump in this so if a human does need to jump in they're jumping in on high level — yeah so within high level you can actually like have a custom view where you can assign which like human like for example you can assign bot which is voice flow in this case um but you can actually change the assenee to one of the consultants right and because we again from the custom fields you already know which assigne originally assigned because — at the beginning of all consultation each student would have been assigned a consultant so it would just change back the assenee to the original consultant that was yeah sweet all right that's um that's the what revenue recovery agent Yep. — That's okay. What's the last
one we got here? — The last one is commerce agent. — Sweet. Let's jump in. — Okay. So, let's talk about commerce agent now. Actually, like commerce agent by definition is going to be kind of heavy retail heavy, right? And just from the report that we're getting from Google, it's actually like 50 getting like 51% like AI agent adoption rates, which is like crazy. It's like on top of financial services and media entertainment, which is kind of what you expect. Telecom. Yeah. and healthcare is actually not that high because you know with like compliance and stuff as well and you want manufacturing — manufacturing and automotive is an interesting one. — Yeah, that's that is another thing to look out for actually now that you mentioned um in public. Yeah. But the key thing is like retail CGP is got quite a high like adoption rate which is not surprising I guess because the use case is there right and maybe I can break it down like kind of what I'm seeing from my angle. So again, if you imagine like we've got a restaurant chain which is like one of our clients, let's say they want to order from let's say McDonald's, right? They want to do some sort of like catering event, right? And if you know for catering like it's messy because like you got some sort of large budget, you got large amount of people that you need to manage and like within the catering menu like this is just a sample that I put together. It's like this is not the real thing of course but like you've got like all the different appetizers, you got like chicken, you got like all the different food types and it's messy because they come in different sizes as well for cater. So like you got like half trays, full trays, single wraps, like it's messy, right? And the problem is that when currently the person like call them up to try to do the catering order, it's like I have a budget of $500. I have 20 people I need to feed. I want some chicken salad. Two of them are vegans and I need to buy next Wednesday and then like I'm at this location. What should I do? And a lot of the people pick up the phone on the restaurant. They just make the food. Like they don't really like basically this is asking them to okay let me take the [ __ ] out. Let me actually do the maths in real time. Okay, $500 20 people like okay this is half a wrap feeds that many people and it cost that much. Like did they do the math? It's a nightmare basically and it's causing a lot of friction. And not to mention as well a lot of the people who pick up the phone at least in this specific restaurant chain their first language is not English. So it's or it already has a language barrier that's creating friction when people trying to do a catering order and catering order usually high like relatively higher ticket items within like the whole franchise right so this was a problem like where like okay what can we do to eliminate that friction and that's where we have considered like basically developed an agent like because within the agent you can just give the same question and you will be able to recommend like let's say three trays of wraps, two trays of rice, two trays of salad for this like kind of specific people, right? And here's the four closest restaurant to your location. Let me know which one you want to choose from, right? And within the force flow front, it looks a bit like this, right? Spoiler aling like the recommended package like the wraps, like how many quantity, the price, right? and also the total price and the cost per person as well. So, — okay. So, you've just created a is this just a literally a fully contextualized prompt that has all the information about like what's on offer and the unit price for that or are you pulling in from a database somewhere for this stuff as well? — So, a conversation really like it's again try to help customers to buy what they want and guide them through the purchase. Right? So, again it's the entire like transaction process and try to and that includes recommended products, right? We have two cases for use case for that. Whereas like for example there's like a you know uh basically a multi-location franchise right to recommend them the catering package. There's also like a retreat company trying to retreat like retreat packages essentially. So it's obviously completely different things but it's the core logic is more or less the same. It's still recommending these something. — So when they've got like quite kind of custom orders that need to put together. I mean, you can do this for like my auntie has a framing business and I know this as like, hey, I have all these pictures that I need to have framed. Like, what do you think? It's like, oh, we could put you can either like put together a couple different packages and say in her case, you could use AI image generators off the back of it and like the it kind of like you say kind of blends into that customer experience agent. when you're like really improving the customer experience, you're going to be instead of waiting for emails back and slow back and forth, you could create this great like commerce CX agent that's able to help them to identify like get proper packages pitched, they get real tight, like very quick responses on how much it's going to cost um as well. So yeah, there's huge use cases for those. — Exactly. And actually one thing I want to add like I was going to mention a bit later but I actually want to add is that you can combine the right and the use case would be loyalty. That's actually another thing that we I'm like we're currently having a discussion with client about like — loyalty program — loyalty program. So for example, if the right now there's no system that's kind of you know for a normal order at least no one is tracking like okay this customer order this food that their favorite like food item is this um etc right that their order patterns like we're discussing internally like okay what if there's a use case where you can kind of map out like the ordering patterns what their favorite food is basically data about the customers and basically build an entire customer profile based on that essentially and yeah try to basically um recommend and package that way. That's another like for that's another thing that we're having. — Yeah. It's almost like driving sales with personalized offers. It all requires collecting that data um through I I've always said that I think these conversational uh channels um whether it's across Instagram DMs or whether it's WhatsApp like WhatsApp chat bots or whether it's um on the on your website. I think all of these are going to become when you're talking to what you perceive as an AI chatbot on a company's website, you're giving kind of raw unfiltered, you're not really thinking people are going to be digging into it too much, but um you start to get really good like intent signals, context on who they are, what they want. And that being like data feeds into the business on one being able to aggregate it and see what the trends are and who the who their customers really are, but also on an individual level being able to like kind of weaponize that against them but and make much more personalized offers. And that's clearly the way things are going. So I think this is such a great thing for people getting into right now. — Yeah, it's such a good use case and there's not a lot of traction right now on this. So maybe a good opportunity. So that's why I jump into it, right? So um yeah, maybe I can break down — it will be now. — Yeah, it'll be now. — Um yeah, maybe I can break down kind of the text a little bit like this. — Um so yeah, we no surprise we first look again and but this use case is basically an order capture conversation flow. So all the time we're just trying to collect information. Again, it's a it's more like a data capture exercise more than anything. So we have the budget for example, what can I get for $200, right? That's the one that's one thing. Customers information like I'm not going to go through this pretty basic, right? Name, email, and stuff like that. — Yep. — Anytime or the event, when is the catering event, — the location because they need to know like because the franchise need to know which individual franchise should be like should be producing the food essentially. That's why I didn't even So the way that we do that is through an open cage um API. So what this does is you will capture the user address. Then you will call the open cage API. Now we already have in the back end all the possible franchise locations and their latitude and longitude. All we're doing here is basically kind of the closest um distance between um the users address which would translate to latl long and the individual locations lat. So you said this is just a pure like coded function. Where are you hosting? So you've got a program running somewhere. Are you like — hosting this on your own little server? Um what's the current like just Python script that's running in the background? — Yeah, it's actually through superbase edge functions. — Ah cool. Yeah. — So um the client set up an account there. So we just built the functions on their end. And um it's basically fetching the four closest location, right? And then the users had the choice to like, okay, these are the four locations I think that's closest to you. Uh, which one do you want? And then they will be like, okay, maybe I like that one. And they have the freedom to choose that one. Okay. — And that's the location. Now the food recommendation is the kind of the hard spot. It's basically a coded a complete like JavaScript coded solutions. Again, another superbase edge functions. So on a very like high level, right? It's pretty much just saying okay I basically collect all the data that I previously should have collected based on the budget the amount of people like the like basically is there any special dietary requirement etc and we actually within the code we have like basically a priorities into which we should prioritize like there's certain parameters that are optional right so like they might not always have dietary requirements right they maybe some most of the cases like okay I just want like I have a budget that's the amount of people go do the maths, right? And it then runs through the script and it just splits out like basically the um all the food options and the prices and basically just a recommendations like to the user. It's like this is totally like totally how much it's going to cost. Um this is the price and this is like what's the food items look like. Now um this is again I think I said it already is basically another super base edge function essentially. So this is where the food recommendation comes in, right? And after all that information, it's then being pushed to like a solo CRM which they're using, right? All the leads like are saved in there. Generally, they don't save leads on CRM because there's really no point if it's just like single orders for now. — But like for these larger catering like orders, they're doing so. Now this is still in development right now because they are still figuring out the kind of payment deposit collection kind of like this basically what we want to do the angle of this right now it's just saving everything in the CRM right the angle of this is to be able to essentially like for catering events you need to collect 50% of the right so it should be able to save this in the CRM and automatically triggers an automation that can send SMS with a dynamic payment link based on this order, — right? — And then you can set up follow-ups and you can, you know, once it's confirmed and it sends back and notifies people. Yeah, that's freaking awesome. I think being able to take handle this like end to end like the first touch like I have like you'd know from the accelerator the solution sphere and how like on the ex like the outer layer you have these these first people that like the interface between the business and the and the uh and the customers and that's where really where generative AI particularly these conversational agents can shine and you can handle everything from that very first uh touch all the way through to like the start of service delivery and ideally you can start to take that even further into service delivery and using conversational AI agents to walk them through that part, all the onboarding and things like that. So, yeah, this is awesome, man. I think everyone can take a lot away from this.
