The Missing Layer of AI Nobody Is Talking About | AI Special on The Mastermind™ Podcast

Ajouté : 2026-06-02

[00:00:00]believe that the more there will be AI the more there will be trust issues.

[00:00:03]>> This is not about a particular technology impacting one particular aspect of the business. Everything is going to be reinvented. The shape of organizations are going to change.

[00:00:13]>> Anything that we do as humans is a result of a conversation between three parts of the human brain. Right now with AI we here and the clock is turning and eventually we'll be settled when we reach here.

[00:00:29]when we reach here it will be a it will be a great time and and the general public and everybody will have a better life. That has always been the case. Uh be it about industrial revolution or or any other um with with the in invention of computers and whatnot.

[00:00:47]>> The journey is always going to be tough.

[00:00:49]Right? So um we are in this transition period right now and this transition is nothing like the industrial revolution.

[00:00:55]It's much much much more.

[00:01:00]>> Divy, you are an AI expert and uh every time I meet an AI expert in in today's time, I almost get spectacle, skeptical.

[00:01:09]Um you see the problem is that I I keep going to a lot of conferences and I and I exchange cards with somebody and then they say that I'm an AI expert. The moment they say they are an AI expert, I get skeptical about it. Uh but you got into AI in 10th grade, what did you actually think it would become?

[00:01:33]So back then uh firstly I don't consider myself an AI expert. Uh I I am frankly adverse to that that award myself.

[00:01:43]>> Um when I got back um to AI in 10th grade, I mean it was really fascinating.

[00:01:49]What fascinated me was, you know, I I used to love um uh movies like Star Wars, Star Trek, uh really enjoyed, you know, I always like to think about the future. I I like positive futures and uh there is no future without AI uh that anyone can imagine. So I I got very excited about what the systems are, >> can they be taught to learn and so on.

[00:02:13]But you know back then uh technology wasn't really there. Most of the work was around knowledge representation. Um so the idea was can you in fact just represent the knowledge of an expert. So for example, let's say if you're a homeopathic doctor, which is what I was trying to build back then, >> it seemed like an interesting problem because homeopathic doctors, if if you've been to any homeopathic doctor, >> u they they diagnose based on um um uh you know, they're kind of searching for for the problem, right? So they're asking you what do you have any stress in your life? Are you married or not?

[00:02:51]How many children you have? They ask a variety of different questions which has nothing to do with the ailment you have.

[00:02:56]>> Um but the idea is to find and search the the diagnosis and so it seemed like a great problem for me to work on back then >> on the cutting edge technology of the time.

[00:03:08]>> Uh things have evolved significantly since then. Of course we have a great deal of horsepower now. uh we have the GPUs, we have the RAMs, we have the data centers to be able to train large models >> and as you probably know um even the latest models have basically engulfed in the entire internet uh and they've been trained on that data. Um so things have shifted now completely from let's program and create the knowledge representation to let the model learn that knowledge representation based on all of the data available >> and that's been the transition and um I'm I'm very excited for where the world is heading um and AI is definitely uh going to transform humanity uh much more than any other technology before Interesting.

[00:04:02]Um, you mentioned about data centers while it's very very trendy and and people know about it. Define break it down for me. What what is even a data center?

[00:04:14]So data center is where uh just imagine um a let's say that you had to do this yourself and you had to buy a bunch of computers >> um which you would then hook up uh and host all of your data and let the model um run on that data and um you know as as the model trains itself it would evolve and it would um uh grow. The problem is if you had to do this all yourself, you would need hundreds of different uh hundreds of computers, really expensive setups.

[00:04:50]>> Uh what data centers really are is someone has pre-invested on this uh for you. companies like Amazon, Microsoft, uh Google, >> they have humongous amounts of millions of uh GPUs uh and computers >> that you can now utilize without you having to invest in it yourself >> uh to be able to build and train these models.

[00:05:15]>> Uh so think of them as just a repository of computers and hardware that is available for you to train your algorithms um and spread across the world. uh for you to utilize through the internet.

[00:05:29]>> Interesting. I was thinking about uh your answer. Uh you I see you've got 10 plus years of experience implementing enterprise uh SAS and AI and you've also uh gotten interested in AI when you were in your 10th grade. That's a long time.

[00:05:47]Uh I'm getting a bit off track, but tell me what what did you get wrong about AI back then and and what has changed?

[00:05:54]Yeah. So like I said um you know um not just me the entire industry was uh focused on knowledge representation. Um so think of it as let's take an simple example right so let's let's say that we want to make AI understand English >> and be able to talk coherently in English. Uh the idea back then was you would need to represent you would need to create a representation of grammar how sentences are structured uh which word comes before which word etc. >> And uh the other assumption was that you would need to build different AI systems for different use cases. Now the part that I was completely and not just me I mean if you talk to any uh any AI experts out there >> what surprised everyone with this new generation of AI is the intelligence has sort of evolved as more and more data was provided to these models. it now coherently talks in English and in any other language without necessarily uh having represented the the structure of an English uh sentence >> uh in advance to this model. So it has kind of learned it on its own >> uh based on a very simple rule um which is you know predict the next word in a sentence >> uh and just based on that one algorithm it is able to do all these things. The other very important thing is that um my assumption always was that you would have to build multiple different models for you to get the level of accuracy.

[00:07:36]And this was true until my last company where um you know machine learning models had to be trained on different sets of data for different types of problems. a world in which the same model can uh you know write legal contracts, can do marketing, can start coding um was sort of unimaginable. Um so that completely uh was a surprise for me frankly I would never have predicted that.

[00:08:05]>> Interesting. you've you've mentioned about uh generations of AI and and then you've you've emphasized a lot on predictability and and what will people realize let's time travel 10 years from now what will people realize they completely misunderstood about AI what's your best guess 10 10 years is a long time I mean the way things are moving right now >> every month is like a year uh models are changing all the time so any prediction I make 10 years from Let's say if you >> if you zoom out and and a meta view of things >> give me a very broad view because I know getting into the details and zooming in would would never be correct but >> just for the zoomed out view what do you think will will happen.

[00:08:51]>> Yeah. So think of it like this. Okay. So even anyone can imagine imagine uh a world where suddenly some aliens with much superior intellect and abilities uh suddenly lands on this planet right let's say there's a there are a million such aliens that have landed which have capabilities of you know or intelligence of an Einstein where physics is concerned >> uh is as good as the best neuroscientist in the world >> is the best lawyer and the best philosopher uh and uh the best creative writer >> all in one and imagine the impact of that on economy if you had a million such u beings suddenly land on the planet right so that's what we are we are heading towards we are heading towards these entities that are going to be so superior to who we are that um it's mindboggling what's possible so what what I imagine is almost Every problem that today we are grappling with large problems like in physics uh there has been work around for the last 30 to 40 years in trying to find a unified uh theory uh which unifies both uh what Newton says and what quantum mechanics says >> uh if you can solve these problems so these are going to be beings that will resolve all of the issues that we've had things like weather will get resolved uh consciousness which is another big question uh will perhaps get figured out. What is consciousness?

[00:10:27]>> So you're going to have these uh extraordinary beings to completely transform economies, right? So that's that's the impact. U >> you you talk about economies, you know, you just uh like you're getting me thinking into second order and third order consequences.

[00:10:44]Uh whenever and I'm talking about because we're we're in a zoomed out view right now. I'm talking about the human race and all sorts of um evolution and uh um new technologies have come up and and and had a major reset on on the existing human establishments. What has happened is and for example the industrial revolution every time something new comes uh the destination has always been good.

[00:11:19]You see for example here's a a sand clock now right now with AI we here and the clock is turning and eventually we'll be settled when we reach here when we reach here it will be a it will be a great time and and the general public and everybody will have a better life that has always been the case >> uh be it about industrial revolution or or any other um when with with the invent invention of computers and whatnot. What happened during the journey of that evolution has made the humankind make a lot of mistakes but at the end of it things have settled down and and the overall life of even the general public and the and the workers and everybody has evolved and and you know they have they had a better life as we raise towards AI what do you think how what can what can the common people or the CXOs enterprises what can they do so that they don't mistake mistake a thought and and don't make a lot of bad decisions while they're on this journey of going from nent AI to a very advanced uh stage.

[00:12:32]>> Yeah. So, you know, there's there's a lot of things uh packed in that question of yours. So, let me slowly try to break it down.

[00:12:39]>> You you got me thinking and uh yeah.

[00:12:42]>> Uh no, so it's the journey is always going to be tough, right? So um we are in this transition period right now and this transition is nothing like the industrial revolution. It's much much much more um everyone that we know is going to get impacted and everyone needs to review and renew the way they do things.

[00:13:03]>> Uh my son uh right now is doing engineering. Uh he's he's in he's in his second year and all kids today are wondering you know what their life is going to be like when they graduate.

[00:13:14]This was not the case. uh you know back when we were doing engineering. It was pretty obvious what's what your life is going to look like.

[00:13:20]>> Yeah.

[00:13:21]>> Yeah. No, you were saying something.

[00:13:23]>> I I mean I don't even know what what how are they deciding what to teach at colleges today.

[00:13:29]>> It is like almost almost I I don't know it feels like fooling around.

[00:13:35]>> So what I tell him and what I tell this the I I think I tell the same thing to organizations as well. You need to get really good at learning how to learn >> as opposed to relying on what you've already learned.

[00:13:48]>> Uh the world is changing so rapidly that there is only one thing wrong you can do and which is stay stagnant. Right? If you do not move and hope that the world will just uh ignore you uh and what's worked for you until now will continue working, >> you are in trouble.

[00:14:09]>> Okay. So um and now where where does that impact that impact is in multiple places. Uh the first thing you need to do is change this mindset that you can just sit there and wait for things to happen. Um once you change that mindset once you've got off that that comfort >> that's when things start happening and at that stage uh everyone should be reevaluating their entire organization.

[00:14:35]This is not about a particular technology impacting one particular aspect of the business. Everything is going to be reinvented. The shape of organizations are going to change. Um start you may need to start thinking about AI as actual employees in your company. You need to re uh rethink what uh roles um that were played earlier by humans can perhaps now be played way better by by uh AI systems and completely evolve the roles that humans play >> and let me just give you an example right so so there's a there's a small simple template that almost everyone can use I think >> and um because I I keep thinking about this question myself right all the time Um, AI is great at execution and I'm I'm using AI right now in multiple things myself, right? And including the the platform that we built. AI is great at execution. What AI is not good at is judgment. It does not know when to do what.

[00:15:37]>> Uh whether this is the right thing to do.

[00:15:40]>> Um whether it is politically correct to to do this right now. M >> um is it more important to build a relationship or um uh increase revenue?

[00:15:52]I mean these are these are some very fundamentally difficult questions that only humans can answer uh and redirect uh the AI. For example, there's a lot of talk today around coding being completely done by AI, right? So what what do software engineers do?

[00:16:10]>> Completely untrue. If you talk to anyone that has done software engineering, >> they will tell you that coding was always 5 to 10% of the job. The job is not coding. Coding is very easy to do once you know what to be what is to be coded.

[00:16:24]>> The architecting the the designing the entire system, deciding uh you know what kind of uh production system you require today >> uh based on how fast you expect to grow.

[00:16:36]Those decisions are always going to be human decisions. And so organizations need to understand which roles to give to AI and which roles human should be playing. Um so that they can they can uh you know with the same number of people with the same um org structure >> they can start growing much much faster.

[00:16:57]Um and anyone that doesn't do all of this frankly uh is in trouble right now.

[00:17:02]Yeah.

[00:17:04]you've talked about judgment uh which which leads me to decision- making and before I move on to my next question I'd want to take a step back you I've I I've almost believed that the more there will be AI the more there will be trust issues and and when you pointed that AI is nothing but uh good at um automating things and and and executing and implementing but it's very bad at judgment I will tell you I've studied uh a line or two about the human brain and neuroscience.

[00:17:40]Anything that we do a as humans is a result of a conversation between three parts of the human brain. The prefrontal cortex which let's call is the the CEO of the brain or the thinking brain and then the lyic system which is the emotional brain and the third brain is the habit habit brain. It's called strriatum. Now the job of the the prefrontal cortex, the thinking brain is to do what's good for us in the long run and it's the most most young part of of our brain. The lyic system is is more uh focused on keeping you alive. It has been you know systemized to keep you uh safe and and for survival. The strriatum that is the habit brain it's there's only one job that it has to do. It has to just do what it has done before n number of times. It doesn't know if it's good or bad. And I think AI as you say as you talk about judgment I think AI is just the the habit brain the the stratum uh as we call it technically and and and stratum has again two parts the dorsal stratum and the nucleus accumans. One is is uh is hungry on dopamine and other one it's just doing something that you've done before. So AI is not different than what what humans do and have always been doing. It's like why people smoke because they've done it so so many times that it's it's automated down to stratum and they're now just executing what they've done before regardless of good or bad. Uh I will ask a question to you on behalf of CFOs because I talked to a lot of CFOs um as part of my profession >> and and because you've spent 10 plus years inside enterprise systems. Tell me what's that one obvious things that most CFOs are not seeing today? um but it's very clear and visible to you.

[00:19:28]>> I mean um why why don't I just talk about what all caps um has found right this this is a question that I have grappled with as well.

[00:19:37]>> Mhm.

[00:19:37]>> Um the the issue that I see is there's a huge disconnect CFOs um are and the disconnect is primarily because of the way things are measured. So if your measurement is based on let's take my example of all caps >> um vendor contracts negotiate uh things like rights, discounts, rebates you know and and millions of dollars of such uh negotiation happens and it goes back and forth multiple times.

[00:20:06]>> Now the CFO assumes that once negotiated the money is in the bag.

[00:20:13]Okay. Uh the problem is and and they they're making their decisions based on that um that once once these discounts, rebates etc have been negotiated they are bound to come to us. M >> that is the biggest assumption and this is just one example right there are several such cases where you negotiate things uh and you assume that this money has been um uh you know recovered or gained and can now be utilized for other things and you may be making uh judgments and budgeting things based on this assumption >> and that can completely change the way uh you know things pan out because a lot of these things cannot just be recovered they're not autom automatic recovery, >> right? Like in this case, someone has to actually do the effort of realizing when a claim gets triggered, raise the claim request on time, make sure the claim gets um gets back to you know to the company and so on. So such things I mean there are other areas where assumptions are made on um you know based on planning uh and operations I would say right so you you know that you've planned it >> uh doesn't mean that it's actually going to pan out >> uh and uh so I I think that that's a blind spot I have seen and you know when I talk to S I don't know what your experience has been but >> I'd love to hear that as well >> and I think you're talking about uh the established systems of of negotiations and and general operations that there has been a way of doing business and they're just relying that since it has been working it it it still works and that's the only way to go about it. uh and I think the most danger you know I was talking to a CFO of a company that's very profitable very well on very green on balance sheet and and they said that uh we do not have a problem and I said because you're very profitable and you're in an industry that's trending and booming chances are high that you've got the some of the biggest problem sitting there but you don't notice because there's no um reason to to worry right and at times >> it doesn't bubble up >> yeah and and at times it explodes once and then uh there's a massive uh wave of uh bad things and negative news. You see um you've highlighted it very correctly uh Diva but and I'll again link it to an human uh human behavior insight that I have.

[00:22:40]>> Mhm.

[00:22:41]>> Humans in general are very resistance to to systems that uncover uncomfortable truth. And this is what you're doing at all caps.

[00:22:54]>> Mhm. Um and it's it's generally very harsh uh to to you know revoke systems that has been always the way they are.

[00:23:04]What's your advice to CFOs?

[00:23:07]Um see advice to CFOs right now is just review these systems because a lot of the times when uh especially in the in the age of AI where everything is going to get disrupted even if you are highly profitable today >> uh many of those flaws that you said that are currently hidden under the carpet >> uh will start to hurt. And so my advice is and forget about what all caps does in general. The advice is please review all of these processes and uh the time has come simply because AI is such a unique technology >> that you can now audit and uh repair many of those workflows and identify the gaps where let's say nobody is currently taking responsibility where some of these things are leaking >> uh can now be plugged. So my advice is do a thorough audit and review of every process that you have because each of those are are potentially uh ready for disruption.

[00:24:10]>> Uh you've you've you've talked about identifying gaps again uh makes me thinking when you say that uh okay forget it think that I am somebody who's never worked in a large enterprise. Explain to me where does this this leakage you you mentioned about leakage. Where does this leakage happen in in these systems that that people don't notice or it goes unnoticed. What are the top three uh areas?

[00:24:39]>> Yeah.

[00:24:40]If if you've never worked in an organization, let me take a completely different example right from outside of something that everyone is aware of.

[00:24:47]Let's say think about a relay race. Uh a relay race is a great example because um in you every athlete in that relay >> uh trains uh tries to do everything they can do to run fast >> and to pass on the bat to the next u runner >> uh so that his job gets completed.

[00:25:09]>> Yeah.

[00:25:10]>> Their job is to run as fast as they can and pass on the baton. Right. That's that's all there is. Eventually though the four guys have to win the race together. M >> uh while you're passing the baton, two things can happen. Either you drop the bat >> and everything goes haywire. You did everything right, but the baton got dropped >> or it took a slightly more time to to pass on the baton than uh you had earlier anticipated and you lost the race, >> right? So the what I see I mean the point I'm making is the loss happens in the gaps. Everyone knows their role in an organization reasonably well, >> but there are parts which nobody owns.

[00:25:53]Uh, and if they're not thinking about the whole picture, >> they're going to lose out. So, in in our case, just coming back to the the case that all caps has, >> um, negotiation is done really really well because procurement and legal believe negotiating these rebates is is part of their job. M >> but then there are thousands of contracts that get negotiated and and are archived.

[00:26:18]>> Um finance is really good at sending invoices, making sure purchase order, inventory, invoice, etc. match and so on. There's nobody really looking at these clauses and trying to make sure that um the the claims are actually happening because it's not really part of anyone's job. I mean there there's no clear definition of you know this has to be done in order for me to do well in this organization and so it falls in that gap uh and and so much money is lost and there are several such examples throughout the organization right where >> um the the part where you're passing the baton >> to the next guy is where the gap is.

[00:27:02]>> So I'm just uh zooming out again just to understand uh the meta view for our viewers.

[00:27:08]You're saying that the moment the baton gets on uh to the next owner there are uh gaps and because everybody is is thinking of their own baton uh >> or the time when they have the baton >> on their own performance >> on their own performance um they that's where most of the leakages happen.

[00:27:30]>> Mhm.

[00:27:32]>> Basically that's that's exactly what's happening right? So if you look at a company you are measured on let's take a very specific example now inside an enterprise.

[00:27:40]>> Yes. Give me an example.

[00:27:41]>> So let's say that the um I talked to all of these procurement leaders >> and um what are they being measured on?

[00:27:50]They're being measured on um let's say a company has a budget identified. They have to now procure a bunch of things. M >> u they have to take uh multiple um RFPs or proposals from various people make sure that they got the best possible deal.

[00:28:07]>> They invest a lot of time negotiating with those vendors based on their past understanding etc. and and they try to capture as much value as they can for the organization in those contracts.

[00:28:19]>> They are measured on the value captured um in the contract.

[00:28:25]Now beyond that they have no time to now go back and keep checking. I mean large organizations do it right but they do it on large contracts. They would let's say some very significant contract may have a dedicated person that's constantly looking at whether these claims are being made or not or whether those things are happening or not.

[00:28:46]>> But by and large there are many contracts and especially the tail spent >> that just gets archived and kept in in folders.

[00:28:55]never to be seen again, right? Because everyone thinks the job is done and nobody's measuring the procurement head >> on whether or not those those uh those contracts are being um uh looked at all the time or not.

[00:29:09]He's only being measured on whether the procurement happened properly or not and is it according to budget or not. So those are the things I mean metrics drive people in companies. Okay? So if you've got the wrong metrics, u you're going to have the wrong drivers, right? It's pretty >> there's a book by John Der that I recommend to people who who are worried about these areas. Measure what matters.

[00:29:32]Um it's not a sponsored recommendation, but it's just a good book. Uh and also it's also very tempting for me to ask Dyb um explain all caps to me in in in like a couple of lines because uh you've been talking about negotiations, procurement, metrics, savings, but you've not really told what all what all caps does. And before you answer >> before you answer, answer if as if you're explaining it to a to a 10-year-old.

[00:30:05]Yeah, I'm I'm uh I haven't I haven't said anything because you haven't asked yet. Services. Uh now um so all caps is um uh think about vendor contracts uh that have millions of dollars negotiated into them in terms of credits, rebates, discounts.

[00:30:25]U but post that negotiation, nobody ever invests any time in trying to recover that money. So all caps has built an automated system that tracks each and every contract that you have uh today.

[00:30:41]>> It looks at every clause in that contract and it continuously monitors those clauses for your transactions and if any of those transactions and clauses get triggered, a claim request is automatically generated, sent out to the vendor and the money is recovered. So in short, we are recovering the money that you have already negotiated into your contracts for you.

[00:31:05]>> I I will reframe this question for you Diva. If if I'm a CFO who's only uh focused on measurable outcomes, I don't I don't want to know what technology, AI or any buzzwords. Uh tell me what can all caps do for me?

[00:31:20]So if you've spent $100 million uh let's say if you're spending $100 million on vendors >> uh I can recover $6 million for you which is available ready but you're not currently doing anything about >> tell me tell me something that you've done in real world have you uh helped somebody or any use case that you have or any >> there are sure there are multiple use cases but let me take a very interesting uh sort example because um you know volumes whenever someone thinks about discounts, credits, rebates uh you tend to think about simple things like volume uh discounts or SLA credits.

[00:32:01]>> There's a very interesting case in in a large fortune 500 company I'm not going to take the name um where it's a manufacturing company >> and whenever you're manufacturing a new product usually in your contract uh you negotiate two things. one is the the tool that will be used to build that product or that that part.

[00:32:24]>> Uh because usually if it's a new new uh part uh the tool also has to be designed.

[00:32:30]>> M >> so let's say the tool costs a million dollars and each part costs $1. Now usually in in contracts what gets negotiated is I'm not going to pay this million dollars up front. I will amortize it over the first million parts that get produced by the tool. Right?

[00:32:48]So, ideally what should happen is you should be paying $2 per part for the first million parts and then it should taper down to $1 per part because the amotization has been completed. M >> I have seen in this organization and we were able to discover that this organizations ended up spending millions of dollars because that amotization schedule did not stop. The vendor continued to charge $2 per unit >> and millions of dollars of extra money was paid to the vendor before realizing that uh that amotization has completed long back and it should have stopped.

[00:33:27]And why does this happen? because it's not you know usually there are systems in place ERPs are pretty good there are uh CLMs in place there are several systems out there so I'm you know suddenly you know here comes all caps and you know we we are claiming that we can do things that others have never done before >> you know we are better than SAP Oracle Koopa etc >> the reason is that all these systems are designed for fixedterm type of notifications right so if this same amotization was a 5-year term, it would have definitely stopped. The system would have stopped it. The problem here is that when that million part completes is not a known uh time.

[00:34:11]>> So basically it could happen very soon if you produce lots of parts quickly or it could happen over years. Why do you think why do you think uh somebody like Kooper or or Oracle or SAP or the big force they have not made something like this before?

[00:34:28]>> See it's this question is is you know let me ask answer it sort of at a higher level and then come zoom back in. See big companies are focused on protecting what they have. uh startups don't have that luxury and so uh many of these uh systems are not designed >> they I mean in fact they're doing what they're doing pretty pretty pretty well and they're getting paid billions of dollars for it >> uh and so these gaps are things that have not yet been uh on their radar I'm not saying they cannot do it the second is some of the things that we are doing today uh the technology simply didn't exist even say 12 months back. Uh why?

[00:35:10]Because my previous business was all about structuring data as well, right?

[00:35:17]Uh what I have noticed is it was almost impossible before this to be able to structurally extract clauses from contracts in a way that can actually be utilized by models um and uh be used to make decisions um accurately, right? and reliably that was just not possible before this >> and so these systems didn't exist. The second thing is I I'm structuring all caps more as a service provider. Right?

[00:35:50]I'm saying and this has been generally my belief in in in building any company.

[00:35:54]I'm not here to sell you a product. I'm not here to sell you a license. I'm here to take your contracts, identify money that is sitting there >> and you're not recovering and recover it for you.

[00:36:07]Okay. So the the outcome is the value recovered not the product or the methodology or or whatever. Uh I don't expect SAP, Oracle and Koopa to completely change their business models uh to start doing this. Right. So it's I think most likely that's the reason I would say. Yeah.

[00:36:24]>> Why should uh why should a CFO trust what you're saying in 2026?

[00:36:30]>> They should not um is my answer. what they should do is and I don't expect I'm a new company I've never done this before you don't know who I am >> um I talk well maybe a little bit uh you know etc and and seem like someone that you should trust but who knows so the way I I I approach any customer is don't trust me uh hand over five or six of your contracts okay let's start small let me show you what value is hidden in it and let me actually open your eyes to what is possible and what is recoverable.

[00:37:04]>> How much do you charge for it? There's an investment that what's the risk levels >> investment is minimal, right? So, basically u we would charge something like anywhere from $2,000 to $5,000 >> and you would have a very detailed report of what's hidden in each of those uh contracts. Uh for example, let me take an example. We just completed um something for an educational institute in Texas.

[00:37:33]>> It was $11 million contract >> for a SAS company.

[00:37:37]>> We were able to find and uh you know with with justification $800,000 that they were overpaying >> uh that they could potentially recover.

[00:37:48]So imagine 8% from just one contract could be recovered. Uh so this is what I say. Don't trust me. just let me show you.

[00:37:58]>> Is there something that a CFO can just plug in and and without getting any massive approvals or or into a sort of a paid pilot or something with you and they can instantly start seeing results or some impact or some evidence or some early signals.

[00:38:12]>> Yeah, we could we could look at uh a similar like a shortened version of the same thing. Let's say smaller contracts, let's say five.

[00:38:20]>> Yeah. So, that's a possibility. Yes. And eventually again uh my method is always you have to build trust. You cannot assume trust and you have to sustain trust. Trust is not something that uh comes automatically. So my my goal always is let's start small. Let me show you something.

[00:38:40]>> Uh uh hopefully I met the promise I made and and let's build on that. That's that's my approach.

[00:38:47]>> You said you said uh now you don't know me. Who's Dyat? I I ask you I know you though but for the viewers who's the viewer?

[00:38:56]>> Oh that's a very difficult question.

[00:38:58]Okay so um the way I like to be perceived let's say that right because I don't know how I'm being perceived uh like everyone else uh it's very hard to know what others think of me. M >> my uh what I like to emphasize most is don't you know value me for my honesty uh my resilience um I never give up that's basically who I am right I I like to I like solving interesting problems >> uh when I see a problem I see I see a potential uh solution it you know it's it's never know there are different kinds of people out there some people will find every possible problem in something >> and u you know make sure that you uh you never sort of um uh you know they they they are great at identifying problems let's say for me everything seems like an opportunity so I like to look at things as opportunities and I like to work on solving things and uh I I would rather be judged on my resilience and my ability to continuously work um I value faith failure >> uh that's very different from uh a lot of people. I in fact encourage failure in my organizations.

[00:40:16]>> Uh one of one of the very important um uh statements in our vision is if you cannot fail you're you're basically uh not really doing anything uh that significantly changes the status quo. So >> I like I I enjoy failure. I'm comfortable now, right? I've done this now so many times that I don't like being in a comfort zone. I I like change.

[00:40:45]>> Interesting.

[00:40:45]>> And I like failure because that's where value comes from.

[00:40:48]>> I have two questions, Da. You've you've mentioned about um mistakes and your focus on opportunity. First off, you've you've had a very successful exit from your last startup. What were you building and uh what was it? The second question I'll just leave it with you is how do you identify opportunities and you know these are opportunities to build a company around.

[00:41:12]>> Yes. So success success was very hard by the way hardearned.

[00:41:16]>> Start start by uh telling me about the the first startup your previous startup and what went right.

[00:41:21]>> Yeah. So the previous startup was basically the company is called dataax.ai.

[00:41:26]Um it started very differently. uh my thought when I started in back in 2011 uh if you remember there was a lot of talk around big data analytics back then so people were getting very excited they thought that now suddenly they have access to so much data that now all it requires is putting some sort of AI on top and the world will change >> uh that didn't really happen and my company was called crowd analytics back then so I I was I was also of the same opinion that you know there's so much data that all I need is to crowdsource data scientists that will analyze the data >> uh worked okay but what you what I realized quickly while doing that was uh it's the data that is really the problem right the data had to be structured in order for the AI to work >> so we completely pivoted to what is now called dataax.ai AI where we started structuring product cataloges for very large companies. Walmart was one of our first customers >> uh large customers >> and um we turned that into the business.

[00:42:31]So business was very simple. You have paper cataloges uh you have uh data sheets >> uh you have different types of just PDFs with information. But when you're buying online, you need to structure that data in a way that online searches can be enabled. So if you're buying a blue striped shirt, >> the engine needs to know blue is a color, striped is a pattern, shirt is a product type.

[00:42:57]>> Yeah.

[00:42:57]>> So we created an engine to create that data and now we've turned that into a business. Every B2B distributor uh uses data X.

[00:43:06]>> We sold the company uh to a Japanese distributor in fact which is a semiconductor distributor, third largest in the world.

[00:43:12]>> Wow. Uh and yeah it was a successful exit but it it was an interesting uh ride uh along the way. I mean there were times when when you feel nothing is working and uh eventually you do exit and it's been it's been a fun ride.

[00:43:26]Yeah.

[00:43:28]>> Perseverance is is always greater than perseverance and consistency has always beaten talent. That's what I know. Uh >> absolutely.

[00:43:37]>> Anyways coming back to the second question. How do you as a founder identify that this is an opportunity and a company can be built around it?

[00:43:44]>> So I I'll stick to enterprise uh because there are two different B2B and B2C businesses are very different. I'm not good at B2C at all. I I have no idea.

[00:43:52]Those are those are just basically uh gambling in some ways. You know some things succeed some don't. In enterprise there is a methodology and the methodology is I connect very early with potentially what I call design partners.

[00:44:07]So uh I will go and talk to people about generally about where they're finding bottlenecks right and they don't they won't tell you but you need to just talk to them about their job you need to ask them where they get stuck um have various conversations so I did that for for a period of time before arriving at this problem uh even then um from outside and this is the mistake a lot of entrepreneurs make is externally you know unless you actually go deep into a problem and actually talk to people. You may think there is an there's there's a problem to be solved here but organizations may be willing to live with that problem. You you need uh you know I think someone put it really well is you know are you looking for a vitamin or a pill is very very important right vitamins are not going to build businesses.

[00:44:56]>> It's pills that build businesses and you need to the only way you will know that is actually having people use your product provide feedback. So you need to have the biggest hurdle especially in enterprise for startups is that the cycle time that it takes for most companies to be able to get that feedback from um you know on on your ideas and experiments >> is very long. If if it takes you seven eight months to get the feedback you are never going to build a business.

[00:45:27]>> So you need to find five or six folks that are willing to work with you.

[00:45:31]They're going to give you feedback quickly and which is what we did before we built all caps >> and then you realize no matter who you talk to everyone starts resonating the same I mean they they quickly agree with you that hey yes we've negotiated credits discounts rebates but we don't have the time to recover it >> how much >> everyone agreed to that statement at least at at a high level >> how much time did you spend on uh this phase >> discovery was at least so I I wanted to go slow this time Right. I mean because I I wasn't under the pressure. So first business uh your life depends on it and you know everything uh you're constantly running like a headless chicken.

[00:46:10]>> Yeah.

[00:46:11]>> Finding whoever the hell can give you business.

[00:46:13]>> That was not the case now.

[00:46:15]>> Yes.

[00:46:16]>> Yeah.

[00:46:16]>> That was not the case here. I was very slow and determined. And you know the one gain that you get uh once you built a large business yourself is you get a huge enough network. You get a bunch of people that you can actually approach.

[00:46:30]>> You unlock.

[00:46:32]>> Yeah, I liked that in the in the first place, right? So, you get to talk to a lot of people. So, I spent almost 12 months just trying to think of various problems. And remember one thing, this is a very different world we are in.

[00:46:47]Okay, this AI is changing every month.

[00:46:50]Um unlike previous generations of identifying a problem, >> you are also fighting against time because you don't want to build a business that the foundation model will automatically do for you >> uh and then disrupt you instantly, right? So you need to find something interesting enough, real problem that will last, you know, and be still valid 3 years from now even if AI has evolved to that point.

[00:47:15]>> Very very well put though. So the the takeaway is that the the pain of the problem should be 10x more than the cost of the solution and that's when you know you have a business in place.

[00:47:30]>> Absolutely. And uh one one very one other very important thing because it's very hard to change behavior >> in large enterprises. It takes a long time and it the time it takes can be uh sufficient to destroy a startup because uh you know there have been cases where you know if you cannot change them you're done. So you need to look at um don't try to build a market >> uh try to find something that they're already solving or at least the bigger companies are solving. So for example very large pharmaceutical companies they are actually outsourcing this work what all ccaps does today to BPOS uh in India and Vietnam it's like a 7 billion industry where procurement services is hey here are my contracts and there are actual human beings going through every contract and and doing the same thing that we are saying we will do. So at least you know market exists. Now if you are able to create a solution that is way better 10 as you said 10x better 10x faster and 10x cheaper you have something that will definitely succeed eventually right it will still take time because very few people you know especially in enterprise the trust that you talked about it takes time to establish that trust >> but if you know the problem exists you will eventually succeed u that's >> so It's sort of the first time when people start up, they they have a machine gun and they kind of shoot everything that they have to. But then >> if you if you're able to survive that storm and and be successful, then you can pick up a sniper and and and lock your target and shoot. Fair enough.

[00:49:11]>> You followed the systems that you've described. You've done everything right.

[00:49:16]Um you've you've validated. You've asked the right people in the right time.

[00:49:22]I'll talk about you.

[00:49:25]Give me a real reason that you think your startup all caps AI can fail today.

[00:49:32]>> Don't kill me for asking this, but >> that's a good question. See, and failure is, as I said, failure is always at the back of my mind because that drives me.

[00:49:44]Okay? Because it keeps you you keep on your toes at all times. uh when you're thinking about failure and failure is possible for everyone. Uh so the way I the things the two or three things that I really worry about right all the time one is obviously this is something any AI company should worry about is in 3 years time think about it this way in 3 years time are you going to still be relevant uh with the pace of change of AI if not then you're you're in the wrong business get out of it uh and I'll tell you why all caps isn't uh in a bit right so that's one thing I I always think about the second is um this technology is such that um most of the magic that you're seeing out there is driven by the underlying foundation model and so the the by foundation model I mean the models from anthropic from open AAI from you know Google and so on Gemini claude etc >> that magic um now many companies may think that hey why don't I just build on top of it myself why do I need all caps >> you know the problem is known we know what the problem is. Why should all caps be doing this for us? Why can't we do it?

[00:50:57]>> So that's the other thing I worry about.

[00:50:58]And third of course is execution is where startups mostly fail.

[00:51:04]Okay. So you have a great idea, you have made a great promise, it sounds great on paper. What if I can't recover any money? Right? I do everything and I can't recover the money.

[00:51:16]>> Yeah.

[00:51:16]>> Then why the hell should you trust me or or value me?

[00:51:19]>> Correct. Um so I have thought about this a lot and that's why I picked all caps.

[00:51:25]So the first one I think I would say is um in our case I mean AI as we discussed a little earlier as well AI is great at automation not judgment and this entire workflow that I have thought of >> this is incomplete without judgment. You cannot give the end value without judgment. Let let's take a simple example, right? Let's say SLA credits.

[00:51:47]>> The AI says $3 million of SLA credits is due because this vendor has not met those SLAs.

[00:51:56]>> But you know that it was a tough problem.

[00:51:59]>> And these are things that you cannot quantify uh or or put on paper. It was a tough problem. They did their best. We want the relationship. You know what?

[00:52:08]I'm not going to I'm going to ignore that credit because I prefer the relationship over the money. M >> now that decision uh cannot be ever made and that's why I like this entire workflow. At every stage in this in the in the process a human is required to judge and validate and I don't think that's going away. The only thing that will improve over time and which which is beneficial to me is the accuracy of the AI will keep going up. So it's actually it's it's good for me because today it hallucinates a lot. I have to spend a lot of time validating verifying information. that validation verification time will go down um as AI improves but it's not the business that will go down that's that's one >> secondly um no it's it's fair everyone should try to build in house but we all know 80% of pilots fail and you know why they fail because uh it it sounds easy initially >> but setting things up takes time right so >> even failing is a blessing I believe if 80% or so doesn't reach completion you fail when you complete treat and then you'll find out that oh it it it doesn't work but I think even a blessing so a a very high percentage they don't even reach completion wherein they know if they failed or they didn't right so because in straight down so um that's how I'd like to think of it uh v you you've given a lot of your time uh and I'm sure uh the members would benefit from your insights and in your journey in in person um the strategies that you've shared for people to manage their leakages is at large enterprises will will really resonate is what I what I know not believe um and and I think you will also inspire a lot of people who are sort of looking to start up and and I'm sure if they watch this they will avoid mistakes that are very costly not just around money but also around the time that they invest in in building platforms and startups. So thank you very much Dyab for all that you've done today. Uh I am sure I'm tempted to bring a round two of this conversation with you wherein we'll dive deep into how this can be more systemized for CFOs so that we're able to give them an actionable prescription and and and I generally believe that that people will benefit in working with you. So thank you so much.

[00:54:22]>> Sure. No, thank you so much for having me Raz and you know as uh I'm always available to entrepreneurs. I did not get that uh access when I was starting up and I I genuinely want to help. Uh so yeah, I hope people get inspired and and they do come to me uh and uh happy to help.

[00:54:43]>> Thank you Dy. Thank you for being on the mastermind.

[00:54:46]>> Thank you so much. Thank you.