EP. 259 AI-Powered Customer Insights: Boosting Lifetime Value with MongoDB & Generative AI

0:06Hello everyone and welcome to the MongoDB Podcast Live. I'm Shane McAllister. I'm one of the leads on our developer relations team here at MongoDB. Today's show is entitled using AI to increase customer lifetime value. I'm joined by Samir Agwal, Chief Product Technology Officer with AG Meta to take us through today's topic. So with that, let's get Samir onto the show. Samir, your very welcome to MongoDB Podcast Live. How are you? Very well, Shane, it's great to be with you.

0:35Samir, what I'd love to do before we get into the meat of today's topic is learn a little bit about your career path to date. How did you get started? Where have you been? What companies have you worked in? Before we get to the AG meta story, tell us a little bit about your background. Absolutely. Yeah, I'd love to. So I'm a second time entrepreneur, have run global engineering organizations, run global product organizations, managed PNLS, worked at startups and large companies.

1:02That tells you a little bit about how old I must be. I've done all of. Those things, listen, there's always a history in everyone and I think. Exactly. Right. It's just statistics, Amir, about the best. The startups are generally founded by people who've gone through an awful lot before. We tend to think of startups by the, you know, the young 20 somethings surviving on a diet of coke and pizza. But I think the successful ones are much, much more towards the other end of the age spectrum.

1:30Absolutely. There's the I think it's a mix here. Who knows, but beers that may. I used to work for a startup, sort of an early employee at a startup in the golden age of the Internet year of the circa 2000. We were acquired by Nokia, spent several years at Nokia, including some in Finland. Another story for another time, when somebody moves from barrier to Finland. That's it. Back when mobile phones were all different. And all the. Models and and back before there was some interest there.

2:02Except for currently, since 2007, everything's been a rectangular piece of of glass and plastic and metal, right? Absolutely. So I saw some of that in the making, which is really interesting. So lots of history there. I, I also worked at other product organizations, security, but really and, and, and in AIA company that was focused entirely on AI, which gave me insight seats to seeing what works and does not work in AI. But the really interesting sort of part of my journey was when when I ran P&L centres and really sort of felt and saw first hand what customer loss means, What happens when a customer churns out and what happens when you don't know why your customers churn out. And as I spoke to others who were people like myself and customers, I saw that there was a problem that was always present, didn't know why their customers were leaving. And most times they would find out much too late that the customers had left and, and by then there was really nothing to be done.

3:07And I think this was something that kind of stuck in the back of my head. And then an opportunity came along and I said, you know what? Why don't we solve a problem that focuses entirely on that? And that's kind of how Agmeta came in to be. So like the best startup stories, they're founded out of a level of personal insight, a level of frustration, and a level of I think I can solve this problem or at least make it better, right? Absolutely, I've got a personal anecdote to share on that sort of front. I've been with the same Internet service provider and cable TV provider for either 20 plus years and as we. All do because we we were loads to move and to change. I know exactly.

3:50Right. Yeah, exactly right. And, and so, you know, it just continued. And then about about a year back, my bill goes up by 35% and I, I, I was prepared for the drill. You call them, they tell you that your promotion has expired. They put you on a new promotion, they charge you another 10 bucks a month. I knew the drill down pat. So you call them, and this guy was really nice. He told me that your promotion has expired. And I said, you know, we're following the script here, this is going to be fine. And then he comes back and says, but there is no other promotion. And I'm like, oh, this guy doesn't get it. Yeah.

4:27So. Because usually you expect the fact that you've phoned up and taken the time to, you know. Get the next. Promotion a year on or get some savings right? He told you there was none. There was none and I said you know what, green guy doesn't know what he's doing, you know, let me just call back again. So I called back another three times over the next day or two and each time I was told that there is no other promotion and I decided to leave them. After 20 plus years, I decided to leave them. OK, OK.

4:59And I moved on a full week later, 4 calls and a full week later, I get a call from this ISP and they go, you know, we noticed you've been with us for more than 20 years. Would you mind telling us why you left? A full week later, they still didn't know why I left. And if that happens to me, you know, $3500 give or take a year customer and they didn't know why I left. This must happen to other customers as well. And I think people can relate to these kinds of conversations.

5:31What if this could be made so simple that these guys could have seen it on day one or day 2 that this customer is going to be leaving them. They could have made it right then rather than a week later when the ship had sailed and they call in and say, you know what, we can take you back to the previous prices and stuff like that. It was just too late. And that's the. Problem you see a lot of the time do you know, like even when you unsubscribe to an e-mail newsletter, for example, and then you get a sorry to see you go, what was wrong?

6:01It's like, well, they never asked that before and it's the same thing. I suppose going back to your original example, you expected to get a better offer. You've been with them for 20 years at whatever many thousands of dollars per year. You're a, you know, you were a locked in lifetime customer and three calls later you went, I've had enough of this, right? And they didn't. Know. They didn't know and they didn't know. They didn't know. And you know, I have to say for somebody at that size, that scale and technology savvy company that it was, if they didn't know, what does that mean for the rest, you know, other organization, other companies that may not be, that may not have the wherewithal, you know, to be able to do that.

6:43So what if we could make it dead simple for them to see which customer is at risk of churn and also very importantly where there is an upsell opportunity. And that's the other part that gets missed out because everybody is so focused on churn. Then you forget the fact that there is the two parts to the customer's lifetime value. There's one about reducing churn, of course. The second is about getting more from this customer. And if you miss that part, that's money you left at the table. So that's what we wanted to address. Yeah.

7:15And and I suppose given that this is everywhere, Samir, how like, and this is a huge market, right? Every company selling the service has, you know, phone lines or online chat, etcetera. What was it that kind of other than your frustration, did you see, did you see going, hold on, I can, I can use my technical expertise, my knowledge, what I've what I've worked with to date to maybe solve this problem. What was the other piece of you kind of going? I think I can I can work on this. Yeah, I, I think the one is just the sort of the technology evolution, you know, things that seem kind of far away, things that seem like, you know, you're going to get answers that using technology, you're going to get answers that are not really close enough or precise enough.

8:04They're moving to a point where you said technology makes it easy for me to do this and and why people are trying to solve stupid complicated problems. Here's a simple problem. This, this problem doesn't go away. You know, it never gets out of fashion. You can replace humans with with AI based agents. You know, AI doing the talking, you can have AI chat bots, but on the other side you still have humans and their customer experience matters. If they make an investment in a company, they pick up a phone. It's not easy.

8:38I mean in today's day and age, I personally don't like the idea of picking up a phone and calling a business. You know, I'd rather just do something on the website and and be done with it. But if I'm. Spending You and me both, I think yeah, look, it depends. Obviously I'm more comfortable, you know, using a chat tool. My wife, on the other hand, definitely picks up the phone. I think she does better than me most of the time when she talks to these companies. But, but I had like you, you mentioned there, your ISP, my, my deal is up, you know, last month or something in December and, and yesterday I went, OK, I better, exactly as you said, better phone them trying or get in touch with them, try and get a better deal.

9:17So I did what I would do, go to the website, click the chat button. And then I was told, no, you're not through to the broadband. You're through to the mobile or the cell phone. And I went, no, well, I clicked the button on the broadband page. Why am I with the mobile team? And then, and then they said, Oh, we'll pass you on. And I got passed on to the home broadband and I went, no, no, no, my package is business because I need it for work, obviously, etcetera.

9:41And they're going, Oh, you have to phone us. And, you know, that was 30 minutes of crazy cat stuff, you know, So yeah, I think and we talk about a lot of topics on the podcast live, but this one, I think everybody who joins us in the comments has a story to probably tell in the same space. So I'm really looking forward to kind of, you know, the comparison and and what you're solving in AG meta. But talk to us a little bit about the the old way because, you know, when I finish one of these calls, I usually get a survey or, you know, a, a text or an SMS to go.

10:16How would you rate your experience? How, how that obviously didn't work for them or it's not uptake or anything on that, do you mean? No, it is I, I, I think at a, at a gut level, we all understand, every business understands they need to keep their customers happy. But I think there is a, a gap, there is a chasm between how the various parts of the organization work. And, and I think this is where we started to sort of hone in on and, and say, you know, can we make a difference here? I'll give you a typical sort of journey. Somebody calls into a contact centre. Contact centre metrics are very different from the metrics of the guy who's trying to retain the customer in the marketing organization, for instance.

10:58OK. So somebody calls in, just like me. I called in, the guy was very polite, very nice, and told me he could not help me. From his perspective, he resolved the problem. You know, the call was a resolution. There's nothing we can do for this guy. That was the resolution he reached. From my perspective, he was 100% lost. Stuck to the script, he couldn't do anything for you and and that was that, right? But but, but from his point of view, this was a resolution. Nothing could be done.

11:29Was the resolution. From my point of view, it was 100% lost. There was no resolution whatsoever. My bill was still very high. And this is the problem that I think continues to exist. And everybody says, you know, you know, you've got to get right customer experience. The customer experience needs to be good. We all sort of swear by customer experience. The challenge is how do you how do you really measure customer experience, right? And this is a big question. And typically the sort of the shorthand version has been if you had a positive sentiment on the call or on the chat, then this must be a good experience.

12:05And I just think that is not the full story. That's part of the story. And the analogy I like to use is, you know, the blind man sort of feeling an elephant for the first time. 1 catches hold of the tail, another catches hold of the leg, another catches hold of the trunk. And they all have a different picture of what that elephant looks like, you know, what that elephant must be. And I think customer experience is somewhat that way. We just kind of, you know, feel the different things and say that's customer experience. But what is customer experience?

12:35How do you quantify it? So what we've done is, and this is where we've filed patents and such where we say, look, we will quantify customer experience using CSAT and that's a measure that the industry uses. But what do you how do you compute the CSAT? The customer satisfaction score? And this is where what? CSAT stands for customer satisfaction. What's the SAT? OK. Satisfaction C satis customer satisfaction. But what we have said is. Surely a customer unsatisfaction score would be much more appropriate for some of these? Industries.

13:14We like to believe that every business wants good high customer satisfaction. But what we do is we compute a customer centric C set rather than a customer satisfaction score based on it. This call was a polite call, positive call. You know, all of those things are component parts to the larger picture. If I as a customer don't feel good about this call and don't feel satisfied, it doesn't matter how good the call was, it is still not a happy call for me.

13:44And I think this is really the thing that we compute that C SAT score and then we bring it back and marry it into other stuff. And we can go back and say these guys are at a risk of churn. These guys are opportunities for upset. And this is something that we do in a unique way. We believe that makes a difference for the businesses. Perfect. It sounds great. And I know in a while we'll we'll see a demo of this in action Samir, which is brilliant. So for those tuned in and adding to the comments etcetera, we're we're going to keep chatting at the high level for the moment, but then we're going to start sharing the screen and actually see how this plays out. But tell me a little bit about obviously that C SAT score, right?

14:24But with the advances that we have in, in machine learning and AI and everything in the last couple of years, how is that really layered in on top of that data that we're getting back from these type of customer interactions? Absolutely. I think Gen. AI has opened doors that we didn't know existed. At least for me it was, it was revelation and I've been following it for the last sort of two to three years. But there are things that we can do now and do it much more easily and more precisely than we could have done before.

15:01And I think we've, we've tried to leverage Gen. AI to the hilt, but we've also used our own sort of secret sauce, if you will. We've also used NLP machine learning to make sure that we pick on other things. And I'll give you an example. For instance, if I as a customer have a contract with the provider, an ISP for instance, and I'm, you know, committed to being with them for 24 months or I pay through my nose, I'm unhappy. But I might still be unlikely to leave because I don't relish the idea of having to pay those large sums of money as a transit. On the other hand, if I'm in month 23 of that 24 month contract, it's like, you know, I'm out of here. I've I've had it.

15:45And, and that factors into the whole story. So we do use machine learning along with our CSAT and, and other things to come back and say, you know, this person is at a risk of churn. This other person, not so much. And, and that's the sort of information, but really the key, I believe Shane, is do businesses have this information in an easy to consume manner, in a timely manner? And if you have that, I think there is action to be taken very quickly where somebody can keep their customers.

16:16And you know, in the example that I shared with you, if they, if these folks had come back to me the following day and said, look, you had a conversation or more than one conversations, we noticed that you were not happy with this thing, here's what we can do for you. I would absolutely have stayed with them. And that's where we must repeat 1000 times each day. That's 1000 customers walking away each day and nobody is any wiser. So. Yeah. And that's it. And as you touched on earlier, not only, you know, retention but also potential upsell and higher lifetime value and stuff. The you, you mentioned, you know, sentiment and, and obviously, you know, sentiment analysis using, you know, vector search, et cetera.

17:01To do that, we have in in Mongo DB amongst the developer relations team, we've got a few team members, few colleagues of mine who've built some really good sentiment analysis demos and they're brilliant, you know, and that's the key thing I think that Gen. AI has brought to the table is data has existed for ages, but it's getting the meaning behind the data that I think is, is beyond the ability, as you say, to surface that to the top and turn that into actionable insights as well too. I think is is really, really, really powerful. But Samir, you also mentioned when we were prepping for this call too, that you don't need all the data. You don't need, you know, an, a mother load of data to come in on this.

17:42You can get these insights with some, some kind of very kind of bare, not a bare minimum of, of data, but you know, really, really kind of niche areas that you're looking for in your solution. Right, good. One of the big learnings for me when I was sort of experimenting with AI and, and was part of an organization that delivered AI based solutions, the challenge was less of AI and more of data. And companies would tell us all the time our data is scattered all over the place. It's in bad shape.

18:17And yes, we love AI, but we're going to come to it when when we've got our data story sorted out. And that sounded like something that might happen 10 years rather than something that would happen, you know, two weeks out or two months out. Yeah. So from get go, what we have focused on is do not start that conversation with saying I need all your data and then we will create value for you. We have taken an approach where he said, give me your basic data, give me your conversations, your interactions with your customers.

18:48Give that to me and I can already create value for you. At that point, I can already create value for you. And if you can give me a little bit more data, I can create more value and I can refine it further. And if I can get more data still from your C SAT and such, you can see even more value, even more precise sort of results. And and so we've taken an approach where you can bring in more data piece meal. It becomes part of the overall so the equation and it gives you better results.

19:14It gives you more complete results, but you don't have to wait for an eternity while your data gets sorted out that you'll start to see results. So on day 2 of being operational, you're actually seeing results. And then you can of course make it better. Excellent. So, so as we all know these conversations if their phone call are recorded, they always tell you that on the way in, right. And and obviously in the chats that you might have are there's transcripts available too. So you're basically saying to your clients and customers that we'll work with what we already know that you're recording before we go into the huge minefield of what's left of, of all of the other data that might require, you know, it's all in different silos, as you say, it's all in different places. You might need to transform it and it's very unstructured, right.

20:02So you're you're working with what's easy for the client to give you straight away and then enhancing that if they can give you more in the future. That's exactly right. I mean, I'll tell you, I was speaking to this is not in the context of Eggmeta, but in a previous life, I was speaking to a major airline here in the US and they painted such a scary story of their data that, you know, is put a fear of God in me. It's like I don't think I can fly this airline ever again. So we wanted to make sure that that doesn't happen here. OK, Yeah, that's a good one.

20:37I think people can can certainly resonate with that. I see a lot of people joining us again from all over the place, from London and Charlotte and LA and Iran and Brazil. So you're very welcome. If you've any comments or questions for Samir, please drop them into the chat and we'll try and take care of them as well too. Usually when make Samir, when we get to the demo portion, that's when the comments start to come because people can see what you're sharing and have a look and they have their own questions. So we've spoken about the data and we've spoken about how you approach towards that.

21:08But tell us a little bit about how you ended up on the the Mongo DB podcast live. So when was this? When did you first start to use Mongo DB and how are you using it in AG meta today? Yeah. So, you know, when we started our journey, we clearly needed some place to store the the results and the the first sort of obvious conversation is one of the relational databases and very shortly there. For most developers or people to start right, it's like, yeah, exactly right.

21:37These have been around for a long time. Let's just do it that way. Exactly right. And and very shortly thereafter we also realised that that's going to be the wrong solution for us because the data is going to be very unstructured. Different customers are going to want to see different things at different times and a relational database is just not the right answer for us. And so we were looking at no SQL databases and we had a few choices there. I'd clearly, I knew of Mongo DB, I'd partnered with Mongo DB in the past.

22:08You know, I had them in great respect. But the important thing for us was we also knew where we are going with this. We do need, we do need capabilities in terms of vector database capabilities, in terms of draft DB capabilities. And the choices that I had to make were, do I take a no SQL database, then shuffle data from that database into a vector database and then into a graph database? And I'm going to be doing nothing but, you know, moving data around between all of these databases, which is not clearly the goal of.

22:43Yeah, that's a lot of heavy lifting and there's a lot of, as we call it, ETL, extract, transform and load and and keep keep that data In Sync, keep it concurrent as well too. So which is a lot of extra overhead and and generally extra cost as well too. Absolutely. And and our thought was look, if there is a solution that can solve that problem, that's the one to go after. And at least our research shows Mongo DB is a great solution. That way it solves all of those problems.

23:14Is a check box for each of these things and the folks are easy to work with, which is never a bad thing. Love to hear that. Love to hear that. And probably timing was good. So we launched Atlas Vector Search back a year and a half ago now I suppose. So that probably coincided with when you were building out your solutions as well, too, right Samir? That's correct. That's exactly right. And in in our journey, we are sort of just at the point where we are saying, you know, we are using the vector DB enabling that use and also looking at graph DB capabilities to see if we can marry all of these and get a graph rank solution as we take the next steps in our. Solution.

23:57OK, OK. And you mentioned on our prep too that you're obviously using GCP as part of this as well too. Tell us a little bit more about that because obviously they're a huge partner of ours in terms of of hosting as well. Yeah, GCP when, when we first started with an idea and not much else, it was just a thought with nothing else. No company, no people, no nothing. It was just a thought. And we started with just the Google infrastructure really just just go back to collab and try something out and see how that works.

24:32And then we reached out to someone at Google startup and, you know, they helped us secure a good number of credits. And then we started to feel like this was the right platform for us. It did all the things that we needed to. We also use Google's Gemini Flash LLMS and everything just sort of kind of clicked and, and Mongo DB works very nicely with it. You know, it's all integrated well. So for us, it just turned out to be the right solution. And we've stayed with GCP and we've stayed with Mongo DB and it, you know, it just helps us not have to worry about the infrastructure piece and having to sort of spend our time on that. It leaves us time to go focus on what value we can create for. Our customers excellent, excellent and you were were you start a part of the Mongo DB startup program as well too, Samir?

25:21Very much still amp yeah, good, good and. For those that are are not familiar with the startup program, essentially it's a, it's a guided program. There's a very straightforward application. I'm part of the team that reviews a lot of those applications. And essentially you're getting Atlas credits, of course, to make it super simple to build your umm, you know, your first iteration or your MVP as people would call it as well too, and but also insights and, and kind of extra Technical Support from our teams here as well too.

25:54And then obviously if things progress, we get an opportunity to do a case study, get you on the live stream, etcetera. So there's a little bit of quid pro quo, but I think in general it's it's in your favor to certainly join that as well too. Rajiv gave us a question there, but I think this is what we're going to see in the demo, which we're going to get to next. So he's on querying as to whether it's a real time solution and yeah, and how quickly the provider can know if the customer is a high risk churn.

26:24Yeah, this. Is this is, yeah, this is a great question. We, we struggled with this. I see a lot of contact centre solutions today that do real time stuff. And real time meaning is on the in the conversation you're doing stuff. And what typically happens is they they sort of hone in on certain keywords and then based on that they do something. We've made a conscious choice to not do this as a real time millisecond based stuff. I just think it is a little overrated because the contact centre operates on different rules and a marketing organization that has responsibility for churn and customer growth operates on different rules. OK, using the example that that I used before, I didn't need somebody to tell me in a millisecond if they could solve my problem or not.

27:18But if they had come back to me the same day or the next day and said, you know what, we're going to bring down your bill by 40% or what have you, I would have happily stayed with them. I didn't need an answer right that second. I just needed somebody to take care of me. That's what that's what we decided to do. Instead of saying we will do this in real time, we'll solve it. I just don't think the contact centers and the the growth organisations necessarily work in that sort of tandem at all. The case in point, I spoke to several marketing leaders and I said how much of your contact centre information is being used?

27:57And they said not very much. Typically what happens, Shane, is after every contact, every conversation, chat or otherwise, you get a survey in the mail and you know, and that's a great way to collect. So to get information from the horse's mouth, you really want to know what the customer felt like. The challenge with that is the response rates on those surveys are three to 5%, which means that the you're happily waiting for a response when your customer walks away and you've got incomplete information there. In our case, we replace the survey entirely because we can generate that customer centric information using AI and that makes a difference.

28:41OK, I see, I see. That makes it. Yeah, because I don't answer those surveys most of the time. I'm yet to find one that says we do. Yeah. No. And I suppose look, the ones I might answer are the ones that are only one or two questions. If they go 3-4 or five questions, I got you know what, I'm not doing this. I'm just going to, I'm just going to drop. Well, look, I think you framed the problem really well so far. You framed the insights as to why you went about this.

29:09You framed the fact that the technology has caught up to get those insights and you know you can leverage GCP and Mongo and vector search, etcetera as well too. I think it's probably a decent time to transition across to let's see this in action, Samir. Let's show us, show us what you have in the demo so hopefully people can understand it a little bit more. This is something that a customer would never see. You know, the, the agents, the AI agents would process calls as they come in through REST API and such. So I'm doing this in a browser.

29:39It's kind of a force thing. Just makes the demo easier. And that's what I'm so I'm going to pick up an audio file. I'm taking a customer, telco, customer service billing issue. And we'll process that and let you listen to this thing. And then we'll also sit and see where we thank. You for calling Sprite. My name is Rhea. How can I help you today? Hi, my name is Francis and I have a question about my bill. Oh, hi, Francis. I'll be glad to answer any questions that you have about your bill. May I have the account number or the phone number of the account? OK, the account number is 12345.

30:19Great. Thank you. So may I know what your question is? Oh, it's like this. Last month my bill was $25.00, but for this month it shows that I need to pay $35. What's going on? Oh, I see. I understand. Sometimes looking at your bill can be quite confusing. So let me check your account and let's go through this together, OK? By the way, do you have a copy of your bill right now? Yes. That's great. Let's go over your bill together and see the difference Upon checking your account, yes, you're right, the bill last month was $25. However, this billing period there is an additional charge of $10.00 for a data usage that totals to three gigs. OK, I really thought that my plan included data. Well, no it does not, but I will be happy to add a data to your plan.

31:21For only $5 more you get 10 gigs of data and it's more value for your money. Considering the current situation. How does that sound to you? OK, that sounds good. That's good. So I'll add the data plan to your account right away. And since you are a valued customer, I will also take care of the $10 charge for you. And that's great. So let me just process the reversal real quick. OK. And it's done. So let me just recap everything we have talked about the $10 charge. I have added a data plan of 10 gigs to your account. So that's going to be $5 more in addition to the $25.00 that you are already paying every month.

32:09And the $10 charge for this billing period was already reversed. It's strictly appreciated. Thank you. You're welcome. And just to let you know, you can also check your data usage by texting data to 777. So have I addressed all your concerns today? Yes, you have. Well, it's been my pleasure to assist you, Francis. Once again, my name is Reya. Thank you for calling Sprite and hope you have a great day. Do the same. Bye, bye, bye. Bye, very good. And here if you notice, can you read the screen or Shane if you can give me some direction. Yeah, I I think it's big enough for for most of our viewers, I'd say.

32:51Anyway, if anyone has trouble, we might just drop it into the comments, but certainly clear to me that that's that's even better. That's that's perfect. And one other thing, Samir, do you see the little alert from Stream yard, the platform we're on? You can hide that. Do you see the stop sharing button down below? Yes. And then the hide beside that. Absolutely lovely. Yes, thank you. Great. In this particular instance, we created or computed AC SAT customer satisfaction score of five and there but right below that you see the reason and this is a customer talking to you.

33:25So you know you would have received a survey where you would have answered the question and and provided your comments. This is what this is in a way. So I gave a 5 because my issue was completely resolved. The agent Tria was incredibly helpful, went above and beyond to assist me. Not only did she explain the unexpected charges on my bill clearly, but she also practically offered a solution that was better value for my money. Waived the unexpected charges. She did everything that you would expect from an agent. You know, this guy clearly was unhappy about the $10 charge on the bill and she addressed that.

34:00She put him on another plan. She took care of things and This is why it was a fight. Now had just as a discussion point. If she had only sort of put him on another plan but not waived the charges, there's a good chance this would still be a positive call, but just not nearly as positive. And that might have been a CSAT score of a four where you are happy with the outcome, but not super thrilled. And then this particular instance, this guy is just really, really happy that, you know, the charge is gone. She's you put him on a different plan that is better value and so on and so on.

34:33And so this is really what we capture here as a customer. Is there a resolution from a customer's point of view? Absolutely there is. Yes. Agent's effort was great. There were no unanswered questions. There was no follow up promised here. So we do capture things like those that usually get left out, right? Because you know, an agent tells me I'm going to call you tomorrow afternoon. I patiently sit by my phone. Nobody calls me. I'm not a happy guy and this is the sort of thing that we would also capture.

35:04Then there is somebody here for the entire call, which is a 3 minute call. You know, we've got all of that here. That's fine. The sentiment was positive. We also captured the reason and the intent of the call and that's going to be helpful in terms of understanding where the business is seeing the most traffic. So we don't keep that information. This is kind of a summary in terms of a number of different things. But the really interesting part that I wanted to bring your attention to is the upsell track.

35:32So one of the things we are tracking here and this is all customizable in our platform, we said did the agent try to upsell? And the answer here is yes, there was no mention of anybody buying anything or selling anything. You're not once did the agent say I'm trying to sell you something or do you want to buy another plan? But she just quietly said, you know, can I put you on another plan that's going to solve your problems, get you more credits, whatever.

36:00And there it is. So new monthly plan that was part then what were the issues? Why did the customer call? And we use all of this information in what we can then show at an aggregate level and at an individual. So this is really what the call is. We capture this information and then in after all of this, we do do redaction as well. So if you're going to be keeping this information, we want to make sure you're on the right side of privacy laws. So we take care of that.

36:28Understood. Yeah, I know that that's very clear. You mentioned earlier and back to the question, it wasn't real time and there's loads of reasons is not real time, but how long does it take to produce this what you're showing? So I processed this in real time for you after I received the audio. I, I, you know, uploaded the audio and I said let's go process. This was a 3 minute. You processed in about 21 seconds. OK, OK, it is near real time. It's not real time.

36:57Yes, right. So there is if. Somebody wants to take action as a business within seconds they can. I think what I have seen in my own experience is these things get done in batches rather than being done as one offs that that at least might take on it. Let me see what this looks like from a dashboard perspective. What, how do you consume all of this information when there are hundreds of calls or thousands of these calls happening all the time? So what we do is a few things.

37:28One, we show you Intel, sort of the actionable intelligence part of it. That's the Intel here. There's also analytics from an aggregate perspective. And then the individual interactions, they all show up here. So the first thing we show is the organizational NPS score. This is the Net Promoter score, which goes from -100 to a + 100. And what we do is we extract the information out of the individual interactions and then we compute the NPS scores based on how many times the interaction happened over a period of time.

38:01And based on all of that, for the the sample of people that did interact, we can show what that NPS curve looks like. And in this particular case, you see these dots and the trend line, it's trending upwards, which is generally a positive sign it's. Going in the right direction, yeah, it's. Going in the right direction, then we also extract information out of the customers calls and customers interactions, which are the biggest issues that the customers are facing over any given period of time.

38:29So here we are saying these are the major issues. I go here, this is a telecom setup and the four big issues that they're seeing are competitive offers coming in, network issues, billing issues, handset refresh issues. They have got an old phone kind of thing. If you look at the competitive offer, I can go down and say, you know, there are three big things here. People are saying I'm getting free handsets, I'm getting free extras. This can be streaming services and such, or I'm getting lower prices from the competitor that you are not offering. And now you can see how many people called, what was their experience like.

39:05This is not so much saying I can look at the individual call details, it's just not useful. But knowing that red and green sort of mix here, you kind of say what's going on in my past straight for each of these particular things and I can do a drill down into your network issues related stuff or whatever it is now the operating. Highly visual and the and the traffic lights work really well in terms of absolutely understanding what things you're you're taking me back this you're you're the same vintage as me perhaps where we remember used to have to defrag your disks and you could see where all the counts of storage where this is super taking me back Samir. But.

39:45Yeah, it works really well. It's like those stock tickers as well too. But yeah, exactly right. It it does. I'd never thought of it that way, but it really does. The the interesting thing here is the the chief operating officer's office is looking at it and saying, what are my customers telling? And this is not guessing, this is really knowing the stuff. And this is again, the point where, you know, does real time really match? And the answer to me is not so much you really want to see what your customers are talking about. And it doesn't have to be in a millisecond level because you're not going to change your competitive offer situation on network issues in milliseconds.

40:24But once you know what's going on or where the customers are sort of yelling the loudest, you know what you need to do. So that's one part of it. And just before you advance off that Samir, would it be the case that there's much as much insights to gain from the green and the positive versus the the yellow and the red, For example, would they tend to look at what's working well versus what's working bad or just concentrate on the bad? No, no, you're exactly right.

40:52This particular graphic is mostly about what the customers are bringing up. But I do have another graphic and I'll show it to you in a second on a different tab, which tells you what's working well and what's not working OK. So I will absolutely talk about that as well. But in this particular instance, the the most important take away is what are the customers talking of? What are the customers? What's causing them to to pick up the phone or to get on a chat?

41:21And and that's the thing that sort of comes to the fore here. Excellent and Kelson has spotted it was a lot of green. So this is obviously not a bad company you are dealing with here, but still work to do, right? Still work to do, otherwise nobody would be picking up the phone or chatting here is the churn risk part of it. And what we've done here is we've taken just a tiny bit of data information, right? I said give me the interactions and I've taken a little bit more.

41:50I've said look, don't tell me what your customer revenue is, but by the customer tell me are they in bucket one or bucket 2 or bucket three or four or five. Bucket one is your least important customers, Bucket 5 are your most customers, you know the guys that pay the most or what have you. And now I can tell you how many of these customers called today, how many of these customers called in the past week. So there is something going on with these guys and I think just the five here in terms of saying the most important customers, I'm only interested in looking at those here.

42:25And now I'm showing you information. This is again we've taken the all of the interactions we've looked at the past week. We take the data from the C, the CRM system or what have you. And then we'd run machine learning algorithms and come back and say here was an interaction, this was the account number, this was the customer. When did they call? What was the customer satisfaction for that particular interaction? What were their major issues and what's the probability of churn for these guys?

42:53And we produce this information in its entirety here so that you can see what's what went on. So if I click on Mary Todd here, Mary Todd, this is the account number, a very important customer in the bucket 5, our computed NPS score is a three. On a scale of one to 10, that's a three, which means she is absolutely a detractor. And what's the reason for that? These are the interaction details. Over the past week. She had called in the last week. Major issues were network issues, excessive buffering, Internet connection. Customer satisfaction score of 1. Called again in the past week, same issues. Customer satisfaction score over 2. All again today, same issues, Customer satisfaction score over 2.

43:42Stands to reason that she's thinking about other options yes. And this is the sort of thing that we can show very easily and say, you know, she's a .87 probability she's going to be gone. Now what we do is we set it up so that you can receive an Excel with the same information at 4:00 PM every day or you know, at three times a day or what have you. And this can go into some sort of a system that says I'm going to send out a message that says we will make it right, or you know, you called, you were unhappy, something along those lines. But solving the problems based on the issues that you have here.

44:21Then I do a similar sort of a thing with upsell where somebody did have a good experience. Oh, OK, makes sense. And, and I'd say they they are very likely to buy from you because they had a good experience with you. So if I click on on one of these, they called, we solved their problem. They had a high satisfaction score. They were AC set a customer value three kind of middle of the road customer very happy. This is an opportunity to see if they can become a more valued customer because we can sell them more services.

44:54They just had a good experience. Rather than wait, go back to them the same day or the next day and say look you called into the centre, we noticed that you had this problem. We were glad we could resolve it. And by the way, we have this another service that we are offering 20% discount on for our valued customers like yourselves. And this is an upsell opportunity that usually gets lost because nobody saw. So we solved that problem quite nicely as well.

45:23And again excels get sent out and such in our next Rev. What we are thinking about is automating that process with an agentic setup where we can take this information, pull it out of Mongo DB and then we can say set it up so that based on the available promotions, based on the guidelines that have been set by the organization, we can send out those messages directly, bring back the results if somebody clicked on it or not and and close up on it all without having to touch it.

45:56And then only escalate things that need to be escalated. So that's kind of the thought process. That sounds ideal. Obviously you know, a lot of people now are in that phase of agentic solutions. This is a a prime example of how to use that. And I love the fact that you're, you know, you're using it to reduce the churn at one side of things, so retain that customer. And then you're also using the same understandings to, to look at towards the upsell as well too.

46:22So it's, it's a win win. And then obviously from the clients that you speak to Samir, it's huge proof of value to them to be able to to show them this from the extractions of their audio calls or their interactions, right? Absolutely. And and I think this is what we found interesting. They had these audio calls forever and they still lost customers because a, they were waiting on surveys or they had sort of festival journeys mapped out AD nauseam, but not the basic information about what the customer felt. You know, how do you put that to use? How can you be responsive within the timeframes that you can address something?

47:03And I think this is where we are creating that value. I'll go to the next tab here that talks about the more sort of cumulative analytics, if you will, that shows you several different things. Overall, if I look at my customer satisfaction scores over the last month, am I generally looking green or red or yellow? The details are almost not important, but it's kind of important to see how am I doing overall? Am I more green? Am I more red? Where are we? Same thing by sort of dates, you know, over the last month, am I generally doing OK?

47:37Am I generally not doing OK based on the customer satisfaction source? And then you can get into the details. And this is something that touches upon what you had said earlier, Shane. When you get interactions from customers, they always have an intent. The customers call you because there is an intent. You know, I have a billing problem, I have a return to be processed, I have a, a product support issue, whatever something. And now you can see and say, look at it very easily and say how am I doing qualitatively even just how am I doing on intent and resolution? Which of my intents are the least number of resolutions? And here you can very easily see that I've got product support, that's a problem, I've got return and refund, that's a problem.

48:18I've got a bunch of things that have issues and other things. Like that? Yeah. Intent and CSAT, Which intents have the lowest CSAT scores versus the highest CSAT scores? Which intents have positive versus negative versus neutral sentiments? So you can slice and dice this in multiple different ways, including looking at agents, you know, which agents are doing well, which agents are not doing so well in terms of the seasides intense. So there's a ton of stuff here in terms of looking at that information and being able to tell where things are going right, where things are not going right. And then we're being able to take some action around. And then final piece here is looking at the data for the calls of the interactions themselves, where I can say look here was the call that we made.

49:06This is the details of that call. It shows you every bit of information here, yes, absent tried, you know, so on and so forth. We can look at the details here, the summary, the C sat reason, other things. We have the transcript, we have the redacted transcript. All of this data is basically available to the customer to take action and then also go do things of their own that they wanted to do if they should want to access this data and do something with it.

49:34And that's really what we are doing right now. It's the nutshell piece and the details piece that you're using the LLMS to generate and create and store, right? Absolutely. We are using LLMS there. We also use some of our proprietary information in terms of the customer centric C sat. This is where we've filed a patent for because we think most times the LLMS will come back with a vanilla answer that doesn't reflect the customer centric. And now we've we've tried this out with quite literally thousands of interactions and our hit ratio in terms of getting the customer centric view of C sat is fairly. So essentially using the.

50:19Standard LLMS, The Gemini LLMS, are we using a fine-tuned version of them or anything somewhere? No, we we are not using fine tuning. We do use LLMS and we also are extremely careful in our use of LLMS because they do tend to help in it. We've got to be careful because we cannot provide customers wrong answers. So we've tried to build guardrails around it. We've tried to use the features available within LLMS to make sure that creativity is minimum and we stick as close to it as possible.

50:55So that's what we are doing. Understood. While you're still on the dashboard screen from Agneta, Rajiv had another question. You captured data point that gives some sense of timing between the last interaction and when they churn out. Is there, is there a notion of that? Maybe there was and I wasn't paying attention as you went through it, but is there something there? Yeah. Absolutely. So there is a, there is a time received time here, OK, that that absolutely has every interaction captured by time and you should be able to filter it, you should be able to get any of that information.

51:28We we do provide filters for that sort of thing as well. OK, absolutely. Given Rajiv's question, he's probably thinking, going back to your original, I suppose you know the issue that you had, you phoned three times, you didn't get a resolution, they didn't follow up within the day, so you churned. Whereas if you have the timing of all the interactions, you could say, hey, if we don't get back to this customer within 24 hours, 36 hours, they're going to go. So let's try and make sure that we have a no. Absolutely correct.

52:00OK. So in this example that I have on the screen with with Mary Todd, again, we've we're showing you her interactions in the past week and a week as sort of that unit of time beyond which I think it just became still, you know, I cannot say she called four months back. So this must be relevant. You know, it is not anymore. And the upsell opportunity also walks away three months later. So this is all talking about how can you be current within a day or two days, you know, something that's a short window of opportunity where you are looking at it and saying in the past week, this is what happened.

52:36She or he called these many times and were we able to solve their problems? What kind of an NPS score was generated out of this thing? And and is this customer worth keeping? I mean, let's be honest, right? There are the the customers that spend a ton of money and you do want to keep them. And then there are those that fall a million times but don't spend any money. And maybe it's good returns, I don't know. Yeah, yeah, they have those as well too.

53:00Most definitely, yeah. But this? Has been a great demo. Is there other areas we haven't touched on yet? Samir, OR. I wanted to show this just easy to do. And then as I said, we are working on some of the things that automate the full cycle where the information sort of travels, the the agents pick up the information, send it across, process it, come back, and then you only get to see what worked and what did not work and then a human can do something with it.

53:32Perfect, perfect. Well, look, I, I think as, as demos go, I think that's been highly illustrative of, of everything that we discussed at the beginning for anybody. And I, we're not again, we have a huge audience from all over the place as well too. For anyone with any questions, please drop them in the chat. We'd love to take care of them if we can. And I suppose the key thing then is if anybody works in this industry or works in this space, how do they get in touch with you and, or how do they try this out or, or how does it work for AG Meta when you're looking at new customers? Yeah, we've tried to make it simple instead of deep integrations to be built and you know, a trial that's very difficult proof of value.

54:17What we've said is look, sign an NDA with us. You know your stuff is going to be secure with us. We'll process 100 of your files, audio files or chat files, whatever you give us, we'll process that. We'll stand up a dashboard just like what you just saw here, and you will start to see if there's value here. If it's not solving any problems for you, if you already have solutions that do all of those things, wonderful, You don't need us. If on the other hand, you see value that you didn't see so far, then it's all worth doing and it's a very simple integration.

54:49All of our stuff is REST API enabled, which means it's super easy with Mongo DB. What we've also done, Shane, is we've taken an approach where we are not super possessive about data. We bring in people's audio files, bring bring in their chat files, process them, and then we delete those source files. So we don't keep any of that and I'm saying you can use my Mongo DB instance and that's fine, or we can also use your own Mongo DB instance. Brilliant and. Keep the data with yourselves.

55:19We'll just keep the basic billing data. Everything else stays in your sort of autonomous system boundaries, and that way life becomes a lot easier in terms of doing stuff. Yeah, that's ideal. I suppose. Look, obviously we hear the stories all the time, customer data leaks, etcetera, etcetera and people are cautious and rightly so. And, and look, obviously within Mongo and particularly with Atlas, we take care of most of everything, but there's always a case of, well, my data has come in, say if they are already on Atlas going to you and then back and, and that, so that makes a ton of sense. I, I love that idea.

55:53That's a nice way to go. And they, I love that approach to a demo. Give us some of your data, not lots, but some, and we'd be able to prove our point, which is really nice way to go about it as well too. So they just go to AG meta and and just kind of get in touch there, Samir or what's, what's the process? Simple, you see the URL up here. At least I hope you do. Go there, there is a get in touch with us or request a demo or or something similar. Just click on it, leave your contact information, we get in touch with you and in two weeks you have all the information you ever wanted and see if this is a good match. Excellent.

56:34Well, listen, that's a that's a really good way to round out the AG meta story. Thank you, Samir, for for all of your time and effort into that. I wanted to just double back on, you know, the founding of this and and kind of scratching that itch and seeing that things could be better. Any words of advice for We've got a varied audience who joined these live streams all the time on the podcast live. Any words of advice for those looking to break out, to do something on their own to create that start up from you?

57:03Somebody experienced in this space? I I think there there is a proverb look before you leap. I think that has done a great deal of disservice to humanity. I think it's better to leap before you look because more times than not we walk ourselves out of opportunities. I think we tell ourselves, we convince ourselves that this is too difficult. It's not going to happen and there will be difficult times. That's a given. It's a roller coaster ride for sure, but take that step and you'll find out. The other thing is just the commitment. It is, people think of start-ups as sort of straight lines that do really well, but it's it's actually an up and down like this that goes on.

57:49But it is a fulfilling journey. It is an interesting journey. You have only one life to live, might as well try it. So I love it. And I suppose going back to what I was saying at the introduction was, you know, obviously I was trying to say that the successful startup founders tend to be on the older side of things, but the, the younger startup founders exactly that they've leaped, they don't know what they don't know, so they just get on and do it.

58:17So I think that's, that's incredible, Uh, piece of advice. And I, I had my own startup for about 13 years, probably a bit too long to be called still a startup, but, uh, you know, uh, but I agree wholeheartedly, you know, just jump straight into it. And if you've and it's to scratch, and particularly if it's an area that you're keen in, it's yeah, it's hard work. Don't get me wrong. I would imagine. And as I said to you before we came on air, it was a holiday in the US yesterday, but you were still emailing me. So this is the way, this is the way it works. Any final thoughts or words for our audience before we go? And I'm just watching the the chat that comes in and we're just getting some nice compliments. And I appreciate that.

59:00And anybody who all the viewers who joined in. So it's been great to have you on board and and do join in for future shows as well too. It's really nice that we can share these type of stories. And you know, I think for me, Samir, you really, you know, you shared the problem very straightforward, very eloquently, but it's a problem we generally understand, but we don't see the nuances within it. And I thought you did that really, really well. I love the fact that you've jumped, you know, through kind of the problem and then actually how the more recent technology of LMS and using GCP with their Gemini models, et cetera has helped. And then you, you gave us some nice thumbs up for MongoDB and Atlas and particularly Atlas vector search and storing the vectors of your data alongside them in, in the same document essentially.

59:50So there's no round trip, there's no extraction, there's no transforming and loading data here, there and everywhere. So you did my job for me there in the middle of our conversation as well too, a bit of the pitch, so I appreciate. I'm a fanboy so you know there we go with Mongo DB. Perfect. Perfect. Yeah. And for those that want to learn a bit more about MongoDB, if you're if you're not so familiar, as I said earlier, we have sentiment analysis demos, etcetera and you can go to developer.mongodb.com to check those out. It would be remiss of me.

1:00:23Most of my colleagues create content there all the time and go in there and search for for Gen. AI or search for semantic search etcetera as well too. And then obviously the startup program that you mentioned, just search for MongoDB startup and you'll see it there. You'll see what we have on offer. Super simple application form as well too. Samir, it's been a pleasure to have you on board. I hope you enjoyed yourself while it's here. It's been a great guest. And I have said this a number of times and I'll say it again, the had a boss who said the reward for good work is more work. So Samir, as AG Meta progresses, I want to get you back on the show to show us what you've done, particularly when you get to those agentic applications that you say AI is going to go off and do some of the heavy lifting for you. I'd love to see that in in action because I think that's the next step.

1:01:14We've we have our chat bots, etcetera, etcetera. But getting the tools to do something to take the what I would call the mundane or that the drudgery out of out of the normal work and allow people to do the high value stuff is for agentic AI solutions are all that. And I very much look forward to that as well too. Perfect Shane, it's been a pleasure, really enjoyed it. No, the pleasure's been all mine, Samir. So listen from me in pretty coldish West of Ireland and Limerick to you and Samir in San Francisco.

1:01:45Still cold, but not quite as cold as the rest of parts of the US. It's been a pleasure to have you on board. For all our viewers who joined in and left questions and those fabulous comments towards the end, thank you so much. Keep an eye on Mongo DB's, YouTube and LinkedIn for future shows. Generally, most Tuesdays you've got to listen to me and A and a informative guests such as Samir. So from me, Shane McAllister. Thank you so much Samir. It's been a pleasure to have you.

1:02:11Do take care and have a good rest of your day. Thank you. Take care folks.