From $750M in sales to starting over: Razor's founder on his new AI bet (Tushar Ahluwalia)

0:00 Each micro-business had their warehouse in the basement. We had consolidated three micro businesses, we had five hundred different suppliers and more than a hundred thousand shipments per week. Razor Group was very large operation at its peak. We were north of seven fifty million dollars in sales probably most brutal version of operational complexity.

0:20 The idea for General Mind developed while still at Razor group. Razer Group is consumer business. Now raised eighteen million dollars. What needs to happen? To reach a billion in ARR for us, to reach one hundred million which I think is possible now from a hundred million two a billion. Tusha you've been building in the roll-up space with Razor Group buying FBA businesses consolidating.

0:49 it's a billion dollar market that did not deliver on the hypothesis of, hey if we build roll-ups there it will be amazing. And now you're building in the AI category and are building agents for I'd say The German Mittelstands and then corporates on top of that? Of course Not only Germany. over time but once You decided Hey i'm Building something new and this AI thing is real.

1:16 how Did you approach Building the company. What was the foundation that you said after building several companies before? That's how I need to start My newest venture. today's episode is brought to you by leapsome and I want To start with a number. I find pretty striking. or out of five companies now expect AI skills from the people working in HR.

1:35 That's one of many findings from leapsomes new workforce trends report twenty twenty seven The report looks at how HRs changing right? Now, and above all And what the fastest growing companies are doing differently. How important do AI skills really become and what place does HR actually hold on the leadership structure of successful companies?

1:53 So, if you want a better sense while we were still at Razer Group. So large parts of the founding team and Razor helped to set up General Mind, We saw use cases... I mean, a Razer group was fairly large operation! At its peak-we're north of seven hundred fifty million dollars in sales.. We had five hundred different suppliers, probably four hundred five hundred plus employees.... We have one hundred million dollar plus even more purchase budget.

2:48 we had probably more than hundred thousand shipments in a week or in the couple of days that needed to be executed on. We have consolidated three hundred micro businesses, each micro business at their warehouse and their basements which you have to consolidate into ten or thirteen global warehouses. so operational complexity was something that we experienced firsthand probably the most brutal version of operational complexity in a trade business was an e-commerce business with essentially buying goods and selling good.

3:20 And that was, I'd say genesis... That experience of complexity was the genesis for us to see this new technology which came about really effectively. twenty twenty four where we saw oh, Chatchapiti Four Oh one you know there is. that's insanely I don't know if you guys remember, but chat GPT-FORO was like a dramatic improvement versus the previous models that Chatbot.

3:51 And we looked at initial sort of use cases while were still in Razer Group and thought this is Very very interesting thing. and historically the supply side of a business was not innovated on as much. As the distribution site, you know You had e-commerce which wasn't sorry Which is huge sort of innovation in the last fifty years thirty years?

4:15 We have online marketing result consumer facing And we thought everything that was pertaining to goods an order flow Sort of in the back end off the business although backend doesn't describe it. there was not much innovation beyond the very traditional, profitable large legacy players. The systems of records players that ERP players.

4:38 and we thought maybe that space in technology uniquely provides opportunity to look at opportunities in that space? We saw initial use cases while we were at Razor Group inspired us so much. this warrants another business. And obviously, side by side the dynamic at Razor Group was such that a new CEO is coming in and you know... The founding team were transitioning out.

5:10 we felt like they still have stuff to prove. We're still hungry It's clear for us then will start an other business. So I think thats how things came together How we got idea of our general mind. Really an idea was more a problem set that we saw while we were live operating at scale and We thought you know the new technology in that problems sets together warrants of business.

5:36 And then as we started general mind, and as we transitioned out of razor group I think. And we integrate every engage with many customers. some Of them are large customers not just German middle stand but i'd say german enterprise. So ten billion, twenty billion maybe even sometimes fifty billion plus revenue businesses.

5:57 We saw a lot of the things that we saw while we were razor group happen at that scale and at that professional scale. you know we call it human glue function. so alot of say repetitive white collar secretarial work happens usually its supply chain handover points which is-the problem sets. that was all within razer group.

6:19 from there on You know, things just solidified. And I would say what we do today is we want to own the unstructured layer outside of the ERP out side off systems of records and um... We think that you know this space will be very important and it'll transform businesses. You said, large parts of the founding team of Razor moved and built General Mind together.

6:58 Like what was important to you? Was it just like hey we know in general were functioning? somehow this will work for theater company as well. or did he take people who are not into actual founding teams? with that your like... Hey! Of course we're a founding team but need add on these people make best version founding team of General Mind.

7:21 And there was always science and art, it's not just a feeling in reset hole that started another business. we definitely set up the Founding Team with the idea to succeed big. so The question is what are the capabilities you need in the founding team? To execute fast. You know I think i'm always a fan of bigger founding teams.

7:47 I mean, their drawbacks and advantages. If you can hold together the founding team? And if you can mitigate conflict... ...and if you unlock productive exchange a bigger founding team should be faster! That has been my playbook in previous companies as well. It's one of reasons why Razer Group although it was underdog from a two thousand twenty-one perspective ended up being one of the bigger players.

8:14 now You know..you could question industry what happened to aggregators, etc., etc. Razor group I think while it's being consolidated right now... ...I think still does probably hundreds of millions of revenue Right? The way we thought about the founding team was. We had Shrestha Oliver and myself from Razor.

8:34 Shresta obviously Stanford Computer Science alumna Was first in all India secondary school examinations Indian Central Abhi which twenty three million people take, um so incredibly smart person and probably the only person that would have integrated ERP systems. uh three hundred times over given that we acquired all of these small businesses.

9:01 And then you had Olly who I know from high school in Dusseldorf and he is a sort best-of his class physicist. McKinsey Guyne was the COO at Razor Groups manage that operational complexity. He had deep supply chain understanding, built supply chain from zero to a large scale and integrated all of their businesses.

9:28 it was clear to us in addition to the subject matter know-how knowing how goods flow supply chains work experiencing problem set We wanted to have deep technical expertise in the founding team as well. Now Shrestha has a technical background, obviously she studied computer science but uh... In addition that you know we onboarded three more what I'd say founder level teammates one of whom is Lenny.

10:04 Lenny was my neighbor in Berlin while i was starting Razor Group. he previously at Coinbase and studied computer science at Columbia, we added Sergio. Sergio studied Computer Science at Tome in ETH and was at CDTM. And then we added Nishrit who I know from my time at Rocket Internet. He studied Computer Sciences at UIUC which is a top tier US institution worked at Google for a bit.

10:41 I think that technical firepower complemented, I'd say our deep understanding of the use cases. And I think together for me was a founding team made in heaven. so it's important we stitch together the right founders from get go because they really determined speed of execution. Speed of execution as i like to say sometimes you know if told before my view is not just function or working hard.

11:11 That's a commodity. Speed of execution, in my view is the function of working in parallel on different topics at the same time and trusting other person to come through without a lot of conflict. Constructive conflict is great. you should have constructive conflicts but like unconstructive conflict so can really be cohesive team with different skill sets?

11:36 Working many I'd say work streams and you know growing platform, growing deployments. Growing customer acquisition, uh...growing customers success, growing usage all at the same time in a short period of time? And can you build an incentive structure that works for everybody and the team is held together?".

11:58 I think that's some magical combination if you can sort-I think eighty percent off a company's success is just that! And um..and then these people also attract other peoples. so That's how we've thought about our founding team. The typical founding team, I'd say is two to three founders and even then over ten years often enough not all of them stick in the company.

12:23 so that's part one thing i want to talk about. And Part Two. Then at the same time you have a hundred percent shares always split through the number of co-founders. Typically, you end up with a larger chunk of shares would like two or three co-founders which also makes the incentive over the long run when your building company for some founders.

12:45 I know founders who've not have enough shares in their large company at some point anymore and said hey i'm leaving building something new because it's center structure doesn't work any more. so can you steal man? The argument from me to say Hey six co-founder Can make sense if Just need a really large outcome.

13:07 That's you, you need to subscribe to a larger outcome if You have two percent like hundred percent of zero is zero. very famous German entrepreneur used to say that Very often many off your viewers may know him. But you Know If you want I have already large outcome? I think it's very very important to have urgency and agency.

13:32 so You want to be fast and you won't have people that are deeply committed to the mission. And I think if, if people don't have um...you know they-they don't feel like hey i'm doing this for myself uh..and I have an allocation in the upside. it's very difficult to unlock that feeling right? Um....You can be as charismatic as possible Right, but it's very important that the in my view.

14:04 That the top um The first ten twenty thirty fifty hundred people Feel like they can make a life changing outcome. It needs to be live-changing. if there is no idea of life changing Why the hell would you undergo so much pain? And that's one thing. and the other thing Is I do think You know it is also how the equity is distributed.

14:29 So I think it would be different if everybody would have the same allocation. I mean, most prominently the. They have been very great outcomes that had large founding teams like cursor for example And i think it really drives speed of execution RSI. We are barely a year old and we've achieved quite A lot.

15:23 The way we're growing is one hundred percent a function Of our ability to do things in parallel. Which is this, which drives speed that drive speed and the other side as how do you keep everybody motivated right? And I think focused on the mission. That's probably more of a role over CEO type. founder uh...which i think um..I'm able to pull off an.

15:49 I think everybody's excited on our end in. I think it works for us may not work. four or many people who are walks for us and deliver speed up execution. You said only a really big outcome can make taking all the pain worth. Yeah, Can you elaborate a bit more on the pain? Oh well it's...you know that was there is pain and pleasures both right.

16:15 obviously you don't do It just because its painful. then you want to make a lot of money. I think That's one element. The other element is obviously Because you really enjoyed the subject matter And yeah, I think a large outcome. You know if the question then ultimately boils down to. do you want have A small share of a large out come or do you wanna?

16:43 Have a larger over Small Outcome? Or The best would be allowed here of a Large Outcome. but that is the entire economics of venture capital. So i mean That Is like Fair enough. i just Wanted To Like After Doing Several Companies. Yeah I think there are other people experiencing pretty much the same, but like they're people who haven't ever experienced it by one to.

17:07 And looking a bit into your mind and you head of what is the pain that you're experiencing through building a large company? Compiling and complying to venture capital rules to build at that big... Yeah! ...is helpful. What's the pain? The pain. You know this pressure to succeed and it's difficult. To succeed, but you succeed despite that right?

17:35 I need to make the right choice. It's like a lot of The Rings journey But on steroids. So sometimes there is an enemy That you need to overcome. Sometimes There Is Another Problems That You Need Overcome. Sometimes You Need To Work In Teams. It Can Be Quite I'd Say. Sometimes You Have Micro Failures That You Need To Overcome.

17:58 You know, you've started a business so you could probably relate to what I'm saying. Um...I think every entrepreneur would say that it's not hunky dory but there are moments of where we don't know if your going to succeed and you succeed despite it. um..there're moments when they will say oh no everything is lost uh....but you succeeded despite it.

18:18 umm...and i think having the tension between short term let's say micro wins micro failures versus the midterm, long-term vision where you're going. And when things are great not take it too personally and moving towards a larger goal because I think you can lose yourself in let's say day to day dopamine spikes but may loose.

18:52 You know, after six months if I look back what have you achieved? It is usually quite impressive. If i Look After Six Days... ...I just see like the daily pain and so that's What I'm referring to-to The pain. obviously they're great moments there They're not So Great Moments But as an entrepreneur.. ..I think you'll always Have To overcome these difficult moments to get to the great moments And if you look Back you need to enjoy both.

19:23 That's what I mean with pain. Talking about big outcomes, would you mind letting us in on the vision? What is the vision? knowing that you can't predict how future looks one or two years from now because as you mentioned GPT-IV was a game changer but we are at GPT‑VI and things. What's the vision? what are the rules or things that you can already define and one of the things at your life.

19:55 Hey, let's figure this out how it will be. so obviously we're Not a model company. We are sort of an enterprise AI company here in the app layer in the application there. Well You know we'll start with the problem sets And it feels like they're something big their Um, and you know let's see where this goes.

20:21 Uh we could say look after having built a couple of consumer businesses. uh it's evident to me that um You know beyond just the software Or the ERP systems or enterprise systems they're being used in an. around those enterprise systems there is a lot of repetitive work And currently I feel like There Is what i call The Unstructured Layer on top of the table layer, right?

20:51 The ERP systems are basically master data tables and transaction tables that essentially... And there's business logic associated with those. On top off that layer today they're human beings who still do sort help to enable these systems. I think certainly there is a short-term vision. There will be companies that will own the unstructured layer.

21:20 Some call it context-layer, some called workflow-layered and I think in a short time. there's tremendous opportunity for this to happen. And then you can break down that now also at quite high level into specific workflows An army of companies trying to solve workflows in the enterprise. It is also, I'd say an area that's quite anthropic proof In my view because you have go deep.

22:01 You need understand workflow. The work flow business Is solved very close To customer. People speak about forward deployment. You need to understand Workflows intimately. Not all of their workflows are judgment driven or need AI, or LLMs. Some of them are deterministic. some need human in the loop sort of review cycle.

22:26 so you know You need to build very close to the customer and I think The vision In the next one or two years is over time. Own more and more workflows To own the unstructured layer outside of the ERP And then Let's see where it, you know goes from there. You know maybe they'll be robots that will be connected to the layer directly and work in warehouses?

22:50 I think always The way world is moving. There are a lot of vision for this could go But i think we're very focused on next one or two years To solve the human glue layer Or system action layer That sits outside of the system of records. And thats what were going after. First workflow you're going after.

23:16 or the first client like how did he build. The first. well, the first problem we saw why were still at razor group which inspired us a lot to start in your business was we had fifty white collar workers sitting between Berlin and China back into day. these guys are basically tasked with coordinating with suppliers.

23:44 And they were usually used WeChat or email to send unstructured communication and coordination, so that would ask about hey when will you deliver? Because probably the most asked questions because people said I'm going to deliver on X-State but not delivered! They also wouldn't fill out the vendor portal And so somebody had to manually follow up two or three times.

24:09 I went will you deliver? they would negotiate price points? They wouldn't negotiate payment days The word withhold payments in case goods have not been delivered and there were Criticize overpayments under payments overdrive release under deliveries quality control rejects, and so all of that unstructured Coordination Would be captured by these fifty teammates and you sometimes with we chat, sometimes the telephone.

24:40 Sometimes with email And they would take that information structure it uh...and basically manually type into our oracle net suite which was are systems of records. in that what's basically the first time we saw these three layers which is layer one systems a record. layer two human glue. We called it so human beings gluing together, I'd say that supply chain handover points.

25:07 and then layer three was procurement. Team and supplier hand over point an. basically from the experience we thought. you know as we started to interact with initial customers while were a general mind. We thought doesn't make if really does not makes sense to try to replace your P systems Because earpiece systems are so deeply ingrained into the business, it's basically at the enterprise level.

25:37 It's impossible to go to a large company and say please replace your SAP. that's ridiculous nobody will do this. you know its too important of. software is working great but what we did see outside of earpiece system there many humans that essentially feed information into the ERP or extracted out. And so, that has been really the first use case that inspired us to venture and say hey!

26:12 That is an opportunity. we saw efficiencies while were still at Razer Group by deploying AI. We thought it must be something huge. That's what really inspired us, the procurement use case. Today obviously German does much more than procurement. you know we go across Procure to Pay in order to cash sort of journeys outside of sort of ERP systems and goods flow industries.

26:40 but that is how everything came together. How did you think about venture capital back then? Because at Razor a few requirements like Once you were able to raise money from banks, use credit lines and buy businesses. You had a certain amount of equity that is in balance with the credit line. so it's different game for business.

27:10 buy them through M&A and then with credit lines, but of course in the beginning it was also an equity play. You were like okay we need money to buy businesses and to finance the team. What was your approach? And is your approach a general mind To say hey We meet a very big and large outcome which probably means you're not bootstrapping.

27:31 But when was the right time? With who did you choose to work and how are you approaching all the money terms? Also Ford. Okay, I think i said it back then in our previous recording that we had two or three years ago Or one year a good one and half years ago While said razor group m&a is not a business It's a strategy.

27:53 So maybe there will be M&A as part of the strategy at general mind as well. And The type? The type of businesses also completely different. you know razor was a consumer business Was a roll-up into consumer business and General Minds let's say enterprise AI software business. And so the way. for us, it was clear that we want to build in that space.

28:23 We thought Enterprise AI is a quite defensible play-to-do. It is something where we know workflows. we've built businesses from inside and process from insides So you could probably add values very quickly And we onboarded capital from investors, many of whom were also investors at Razor Group that were eager to back the team again.

28:51 There was no room for debt in the business that we are building now. That doesn't mean there will not be a room for dead in the future. Not maybe I'm quite certain they'll be consolidation even In the workflow business, in the workflow software business because there are so many companies going after.

29:11 So many workflows. let's see when that will happen maybe in three years four years five years ten years from now where they would be individual experts on some kind of work flows. you know You could probably run an M&A strategy but I think for the moment just were. the industry is The workflow AI Workflow Industry.

29:32 i think Solve as many workflows in a platformized, abstracted way. As quickly as possible and solve and deliver as much efficiency to potential customers as quickly as possibly. so there is no need for debt right now in our business. So I think it's completely different business. And therefore because It's a completely different Business the capital requirements The capital structures too you know, accelerate that business are also different.

30:06 And with regards to of course we want to build a large business and offcourse that happens with external financing whether equity or debt hopefully do it in very efficient way. but I think if you wanna go for the outcome You need have coming through our previous discussion be okay with having a smaller share in a larger pie.

30:33 I think that has been the strategy for most large outcomes in the US and Europe, uh...in-in the last twenty years. Who have you've been choosing as partners to start? You mean on the shareholders side? We have um..a mix of different kinds investors. obviously we are very excited Lucid, Boom VC, Heliat and many others.

31:05 Many angels as well. on the German side we've onboarded some other shareholders including Defiant Joseph has been in the board of my former competitor at Celer X Um, and so glad he's choosing to deploy capital with us. Uh we also have Leo Capital is a Asia based investor um sort of from. you know. that connection occurred during the uh the razor group time And so we're very glad to have them as well.

31:39 Yeah, and I'd say these are this early stage kind of funds that have backed us. We've now raised eighteen million dollars in i'd say seed pre-seed kind of capital. And yeah, plan is to execute hard. I think we're growing very quickly and for sure at some point on board more capital To grow even faster.

32:05 you are around fifty people right now. You were six co-founders. Who where the first hires like? what departments did you build? The fastest in howl and why It might have changed or not. The advantage of having so many people starting the business with you and that kind of ownership is, Business started with the department.

32:41 So you would have sort of Lenny and Sergio, they work on platform. that's all about abstracting Sort-of our use cases making them extractable or making them reusable. so our forward deployment doesn't get stuck in one customer but we can learn And we can deploy very very fast In other companies. time to value is an extremely important KPI in Our business.

33:06 The second would be Let's say Shrestha, Shresta does product. So she has the most experience with European integrations among us and so she works very close to the customer to understand. you know what are idea synchrocies in this workflow? She looks at the same workflows across industries, across companies to see okay What can be productized?

33:29 What cannot be productised or features that we need in our products. then You have Oli uh, only in myself to our go-to market efforts. So we reach out too. um You know all of the different companies you know across various channels to build rapport and relationships. And then you have nishrit. nishrith essentially drives deployments so he essentially deploys on a tenant basis or on a customer basis Whatever the solution whatever the workflow is that with it?

34:02 That we automate? Um yeah I think for some people, the forward deployment part is not that clear yet because not everybody had to follow it. There's this term of a forward deployed engineers right now that are more and more hired in especially AI enterprise software. could you explain us a bit more? Especially on how your using so like In contrast maybe to the SAS World ten years ago, you know in a sort of enterprise AI.

34:41 In workflows You now the same workflow could be very different across not very different but different across companies. The use case Could Be Different Across Company Because The Process Could Be Difference Across Companies. So For Let's Say A Purchase Auto Confirmation Workflow At A Large Pharmaceuticals Business Versus a purchase order confirmation workflow and a large metals steel trading business could be different And The use cases within those companies could also be very different.

35:17 So therefore forward deployment is A new hot term where you say hey, I just deploy engineers with the company to figure out What it is that we need to solve. so we built for that use case? with forward deployment which is i'm a little bit critical about if you don't do it smart. You just build custom solutions you essentially become a consulting business.

35:45 to be come like the new age center, uh... witches are great business but its not a business we wanna built. so that real question as you know how do you marry. Forward deployment with repeatability? and where on this spectrum r u r u? Do you have one product that you just sell a thousand times? Uh, do you have very strong forward deployment motion and get stuck in that for the deployment motion.

36:14 And then what does it mean for scalability? right because if you have a lot of over-deployment That means you need a lot resources for dollar revenue depending on sort of how much your ACV is. so that math I think Is One Of The Most Crucial Maths That need that every company in let's say our court needs to crack.

36:36 you know what how much. What is for the deployment? What does custom, um...what is generalizable? or can i reuse another deployment? I think this is the key question uh..to build a large scalable business. and how can you learn from one deployment and move to the other. So how are you drawing a fine line between forward deployment, generalization or standardization?

37:09 That's very good question. I think we have probably key questions that you could've asked so congrats on asking this question. We split our founding team in such way as we had a deployment CTO which is Nishit and we have a platform generalization CTO, uh...which is Lenny. And both are at odds because the one guy wants to deploy quickly The One Guy wants to drive customer success and other guys says wait a minute how do you make this configurable?

37:48 How can learn from last three deployments where they've done same thing? maybe take couple of weeks more But then build in such a way that the next deployment is not I don't know two weeks, but today's or no two days. But to ours. And That Is i think how one of those solutions we've The organization is structured In A Way Where These Two Different?

38:14 Goal Sets Are Not at all odds With Each Other. and The other Thing Is you Know You You can only generalize once you're faced and confronted with the truth which is in The customer, you need to be very deep into the customers workflows. Very deep Into what? Customer needs to automate with AI or drive autopilot in To under and you need do that across customers to understand okay What of this?

38:48 You know it's repeating In different ways. How can I generalize this? what of this is industry specific and one of these actually company-specific because there's also companies specific stuff which you cannot generalized. But in general i think the asset sort off as you build a general mind will compound into directions, one is deep workflow knowledge so your understand what is generalizable was not generalizable for a specific workflows as you see the same work flows across companies and cross industries.

39:22 And the second vector where value will compound is within a company. We call that memory or let's say customer specific idiosyncrasies, That um... The system learns over time. You could think of it as like you know Know-how that sits in the operators head. For example for one of our customers They have a lifestyle business, lifestyle consumer business and they sell sort of let's say oat milk right?

40:02 And that oat milk cause some customers are boats because their coffees coffee shops our own boats. And that would be a typical, let's say idiosyncrasy to the flow and memory. So I think you build value in that space as workflow knowledge across companies. what can be abstracted? What cannot be abstract it?

40:35 Then within a company Can You Build Context Over Time? The context layer or memory layer have all of the tacit know-how formalized Two paths to go into the first time to set you. Go in two company There is no generalization yet because you or maybe be in your case Because of the razor background, he knew a few things already that you probably could have used too.

41:01 let's say Generalize. twenty thirty forty percent I don't know. so In my perception The First Customer would Be A Higher Percentage Forward Deployment and less generalization, the second one. And a third one each will be trying to have less forward deployment in more generalization right? One hundred percent.

41:25 there are massive amortization effects as you go from one customer. then massive learning effect that if The more you do More deployments with more customers and eventually You get better faster more accurate Over time Just wanted to make sure that everybody understands. hey Like it's an overtime process, like every time you try to automate a bit more.

41:48 So its getting easier to deploy. the thing that I have on mind is the question. one point. i would say sorry to trip but also don't think you can out money this and they cannot do you know? You cannot do like five hundred deployments at the same time And maybe he will be a little faster. But you need to correct This.

42:07 what you just described Is The core Thing That you Need To Correct. In my View that may be different from what everybody else thinks. But in my view, at General Mind we believe... What you just described is the core thing to crack which is to understand how can we deploy the same use case very fast? We call it multi-industry, multi workflow, multi ERP.

42:33 so How Can We Deploy Multiple Workflows That Have Different Earpiece Across different goods flow industries? so not service industry. So good's flow industries. But how can we do that in a very, very fast way? In a very fat time to value right. That I think for us is the key sort of challenge. if you solve that We believe you can build a huge business Because you can very quickly go into a company and they say hey i have say it's out of creation modification Say okay.

43:01 I'll have that ready for your tomorrow although You are a raw foods company that buys and sells fish. But I can do the same for a semiconductor business, because the workflow is generic and standardized enough. but what of it is idiosyncratic? What if it's not idiosyncratic? so understanding how you can generalize across industries and deploy quickly with different kinds of integrations into different legacy systems and across multiple workflows.

43:34 Because the key insight is that customers, once they're integrated with you. They don't want to integrate with fifty other workflow vendors. You are The Workflow Company, the work flow guy. so there will say hey I have this workflow Where i need an AI autopilot and in name of game as AI Autopilot right?

43:55 How can you take human glue what we discussed initially And fully move it? no touch, sort of execution with the AI. And one of the insights is that customers once you have demonstrated that you can successfully deliver AI autopilot on some of the workflows they will come to you and say hey I have these other workflows than i want.

44:22 do so their relationship right? Is what it's the asset in this space my view. You can then sell to all of the other workflows, but you need to do The same sort of motion that you just described. We have a messy invoice or three-way match four way match workflow. we Have to see That at five or ten companies abstract it and Then deploy It.

44:50 so we Can be very fast with Very fast time To value. i think that is key question to build A giant business in next one Or two years. your little intro A little bit, you know. go deeper there. Thank You. I think we can agree on the critical role that an FDE like forward deployed engineer has in that calculation.

45:13 Not saying the CTOs don't have a critical role. The role is even more critical to figure out over time Like how Can We Get More Towards This Generalization. But What Are The Things That UR Training the FDs on, because as a forward deployed engineer my goal is to deploy a workflow that solves every specificity this company needs.

45:42 At the same time MyCompany wants to generalize so it's like two things I have to balance. i think It could be probably faster just to deploy an individualized solution they want rather than trying Go through the whole process, but that's not in our company is interest. So how do I train? And yeah weaponize the front of the forward deployed engineer to actually be the most productive for a customer and the company while because he has two roles you just describe their reason why there will be immense winners in that space an like big failures in that space, and my view.

46:30 Because walking the narrow road is not easy. yet everybody will have to do it. some people don't even know they'll walk on their own roads. The answer for this question isn't a prioritization but an optionality. You just need so much demand. your go-to market team needs to produce so much choose in that initial phase, which customers you want to interact with.

47:06 And if you can do that then you pick and choose the set of customers you go with them. sort of both meet your revenue target, your customer success target and your generalization target Which are almost like pointing different directions? You do it over-and-over again. these things no longer point a different direction because end up having fully generalized stuff.

47:30 Once you have fully generalized a use case, speed customer success and happiness should usually point in the same direction. it's only at the beginning that they don't... That is the capex! You need to overcome this. In my view we do that by very strong customer relationships. so its go-to market relationship exercise AND a lot of optionality which are top of funnel acquisition exercises.

47:54 And then maintaining all those balls Uh, let's say all of the customers at the same time and then picking and choosing putting our engineers in a position where they can pick and choose. uh we're sort of Being super customer specific while generalizing. At the same Time is not that big off a pain as if you do like entirely different two entirely different use cases And then somebody needs to generalize.

48:20 That is I think difficult. so i think it As much of an optionality go-to market question, it is a discipline. You need to be very disciplined. what projects you choose which products do not choose in the beginning so eventually want all human glue use cases or workflows and messiness capture the entire unstructured layer horizontally across the enterprise on top of legacy ERP systems but walking that narrow road for your initial set of uses.

48:50 I think this isn't trivial. And I think as a consequence, what you'll find is. You will find an army of companies that would be stuck at one or two or three or five million dollars in ARR because they're forward deployed In a large fortune. ten company which it may be a huge ACV like five million and ten million dollar acv.

49:11 But then how do you get there to billion dollars off air are right? That can only happen if you capture many, many companies. So for a deployed engineer wasn't the term three years ago? Let's put it that way. of course you always had people going to customers and trying to figure out what is happening.

49:32 but What will you say? Is the right profile of person To become a forward-deployed engineer? or what are your looking at when hiring FDEs? It just You know. I think we have split the forward deployed engineering role into two sub roles. And often you find sort of engineering talent that can engage with customers, sometimes engineers are better at solving a task by themselves as individual contributors.

50:13 so it's very difficult to find profile to make that onboarding process or you know, the process for easier. For us we have what? We call a value delivery team and it deployments team And The Value Delivery Team. sometimes we would also on board Let's say commercial leaders that are technical Rather than Technical Leaders That Are Commercial.

50:40 Because Finding Technical Leaders that Are Commercial Sometimes It's Harder than finding commercial leaders that are a little bit technical. And so we always tag team, our customer from both teams. So you could somebody is there for the relationship? For the workflow knowledge oftentimes The technical people that we have.

51:00 they may also not understand workflows as well As somebody who previously worked at an integrator like an SAP integrator or something right. those Are the kind of profiles we'd hire in the value delivery team. And so those usually we try to tag team, to capture both customer know-how solving the problem for the customer.

51:20 But then also than passing on that knowledge uh two um let's say The platform the generalization team and Generalization doesn't happen fit the first deployment. it happens once you've done ten You know? It gets better with the eleventh twelve thirteenth in that use case In that over time you become cheaper better faster more accurate And that actually is a good sign because it means there's room for business where you'll have to... It basically adored, little bit adore by door game.

51:53 Therefore sort of platform businesses can just let us say launch product on self-serve basis or build your own agent. this my agents studio built their own agents. That doesn't work right? Because we had be deep. Both technical challenges to this, there are know-how challenges at the customer. There is workflow understanding, there's idiosyncrasies etc.

52:21 and so you know interaction with a customers. really that unearths alot of know how that you need actually be able solve these problem sets for the customer. So tag team Especially important to hear when you're thinking the first time about hey, I need to build a forward deployment team because we want to go into the market.

52:43 or We are in the market and even When you built something think Hey through AI. this needs to change like organizations changed. Even The businesses that our sustaining. either You have cases like a mural or an air table that do ARR and half cash on the bank but not bought for the like twenty thirty X ARR multiples that you saw at some point, but further two or three X multiples and cash on cash even less when you pull out to cash reserves.

53:15 But so I think either you have to transform a business if possible For some this won't be possible Or from any of the larger ones where re-think it from the ground up when you're building a new company, but this is one as You mentioned. One of the critical challenges that you currently have especially thinking enterprise AI software and agents What would you say are da?

53:45 Like where's the point in time at the moment for you? because we said It's a gradual process from like let's see to make it easy zero percent generalization. two As close as possible to a hundred percent generalization, which will probably not work but there for long time. But what would you say on that scale?

54:04 Where are after the one year of building? I think it's use case by use case and we've deployed with some. you know, we have a bouquet of use cases right now that we go after and some of them I'd say were fully generalized. And like no new deployment very rarely would surprise us because we've seen so many edge cases.

54:34 So we know how to generalize that specific workflows. Then other work flows. where starting...I think there are probably fifty workflows or hundreds of workflows depending on how many industries he goes after. We're not focusing only in goods flow industry that we can go after. so, We have now generalized maybe five to seven and another Five-to-seven are inbound.

54:59 And those five two seven could probably already produce a fifty or hundred million dollar peanut. right if you just look at how many companies Have the problem? I think over time You know i think there'll be a bouquet of workflows That you want to connect And that will form essentially the context layer on top of, let's say legacy systems or records.

55:29 That context layer we'll have know about how a workflow is done it will happen. It'll have all the memories, idiosyncrasies that are specific know-how to that organization and it would produce outcomes. for example following up on a purchase order or matching an invoice doing all of those white collar secretarial repetitive work That is being done in the organization, In The Enterprise.

56:03 And then I think that will be a very valuable layer to own and over time that layer expands as well because models get more intelligent. you can do more intelligent work You know? In twenty years we'll not just focus on white collar secretary repetitive work go into maybe more engineering focused work, you know it'll expand from there.

56:29 And so ultimately where will this be in ten twenty years? Who knows. but I think a lot of it would be captured by AI. the fact that hasn't been able to capture by AI right now is because every business different Every business as humans, there's politics. There is decision making. every human has or every enterprise has different workflows and systems that need to be understood.

56:56 so they're the needs of me. a lot off deployment. even the large labs have created these deployment companies I don't know if you've read it because to make their technology work in the last mile So it delivers are why and covers edge cases and works without hallucinations and produces outputs is a lot of plumbing.

57:19 It's a lot off sort of hand to hand combat, figuring out how you can productize that hand-to-hand combat. the is i think we're all the value sets. so one thing we didn't talk about yet this pricing like. We talked about standardized software that we all know. You have your typical ACV, it's pretty much the same for everybody.

57:38 or seat-based like oh you have ten users so you're paying ten times X and um...you've got twenty users then? Twenty times X! Oh do want to negotiate?! Yeah if you are trying pressure us as well can negotiate a bit on price but not too much. let's see based. So how were you approaching pricing and did it change since you started?

58:01 Yeah, we have iterated quite a bit on pricing. I think for us because we deploy AI autopilots and take over workflows end-to-end that are currently done by humans the natural way for us to sell is say okay how much do you... How much does payroll cost? Currently it's essentially focused on this boring repetitive but necessary work.

58:33 And so then we essentially say, you know and if done a couple of iterations I'm telling what do today? We said okay well we can drive an autopilot off let's in ninety percent that workflow end to end within the next X weeks... ...and then we doing ROI calculation on that basis.. ..and it has for us produced at day best win-win.

58:57 So you say your saving a hundred thousand in payroll. Somebody has twenty people full time equivalents working on a specific repetitive workflow email ping pong document validation writing information into earpiece systems and out of earpiece system. those could be the case an order management invoice ingestion logistics claims, workflows in those categories.

59:28 And then we'd say, okay well where are these guys situated? and ideally they're situated in Europe or Western countries? We'd assume a fifty to sixty K average salary for their sort of situated in shared service centers somewhere in Asia at which point the scale would be larger, but maybe average salary would be less.

1:00:00 And then we do an ROI math based on this and The idea is also not to sound super aggressive and blunt. But there's also no depending upon the company to replace these guys. It free up that time so people can devote their time On other more productive work streams. That I'd say layer one of efficiency.

1:00:23 That's like depending on the use case that is I'd say, The most basic sort of entry point into the ROI pricing discussion. Pricing decisions always an ROI discussion and don't we never see it costs five hundred bucks per seat because you know? The same workflow could be massive at a company And not so massive in another company.

1:00:45 the layer two which i think as more difficult to price Is let's say The mistakes that the AI does not do but humans have done previously. because there is no record of let's say a customer or an employee has sort of missed and email. Off let's see two are three million dollars worth off purchase budget, and you know it was no action.

1:01:11 so now thats right offer something. in other it difficult to comprehend what's the lets a top line mistakes? that humans have done versus the AI would not do. I often say, humans are also probabilistic technology just like AI because we will say oh AI hallucinates as well. human beings also hallucinate and I'd argue if you build your harness correctly That's essentially what we built.

1:01:41 We built a harness for these different workflows. then you'll probably hallucinate less than a human. Harness is typically something, for example like Codex or Cloudcode and can be individualized. Harness basically everything that you need to do. so your workflows are taken over by AI end-to-end in production with minimal mistakes.

1:02:08 if we would just let an AI lose it will not solve uh, task as well. Um, As it could because it doesn't have company specific context. It doesn't understand the workflow. It does understand different permissions that can and cannot do um, And so you know? That's sort of the entire engine. The justification for our existence.

1:02:33 right. what we buy tokens We build a harness and sell an outcome. Right. how abstractable that harness is Um, and how quickly we can deploy it is the discussion. We've had right now The question that I have still like let's say because Christ pressing is not tangible Like um And i don't want to Say for whatever you say Now It's Not Like Every Customer Will Be Same.

1:02:57 But Just To Get Understanding You're Freeing Up a Hundred Thousand in Resources. Can be Because Somebody Is Letting Somebody Go and Doesn't League Twenty People just Eighteen or Fifteen Or Ten or Whatever. Then it's probably more than a hundred thousand that you're freeing up, but or mistakes are reduced.

1:03:14 Or anything like that? What percentage of that Are you? Estimately pricing approximately Like off course. It's not always the same. But to give To get an understanding I think You know The customer can expect at least A ten X reduction. And in some cases where bucket two we've just focused on the payroll bucket, but Bucket Two.

1:03:41 In some cases is much more interesting could also be a hundred X reduction and costs. so for one of our customers they use legacy systems of records and We are on the invoicing side. They have a large fleet to have ten thousand cars and those cars basically Have true an operators need to interact with workshops because these cars need repair And the invoices that these workshops send are very small and they're like five hundred bucks, thousand bucks etc.

1:04:15 What we have uncovered through our systems is sometimes for different invoice numbers They would use the same tire number wheel number. We uncovered probably five percent of overbuilt APs to that companies. And so the amount we found in top-line mistakes far exceeded the payroll efficiency, and it's not only takeover work by AI but sometimes or often times the AI just also does like a hundred X better work.

1:04:54 So I think if you looked at the payroll piece In most companies would be a ten x improvement in terms of if you really wanted to free up those resources. And, oftentimes that's not the goal I have to explicitly say but often times In most companies we work with The goal is not to free-up It has to reallocate towards more judgement Higher level kind of free at time To spend their time on higher level activities like negotiation More strategic stuff.

1:05:31 But in some cases, it's far beyond the payroll efficiency that we drive. It is top-line efficiency particularly on... Some of our use cases could be immense and those are not captured by legacy validations. if then validations Those are captured like more AI driven validations Okay whoever wants to figure out their own pricing has a very specific case.

1:05:55 I'll link your LinkedIn profile because you won't. You won't give me what I want to hear in numbers like how do you actually make pricing tangible? and that's okay? It really depends on the negotiation dynamic, i must say. On a customer-by-customer basis right so it will be absolutely transparent. So if I have company where I have efficiencies of millions of dollars which is sometimes the case... ...I can take smaller percentage efficiency and pass on the majority of the efficiency to the customer.

1:06:30 And if sometimes, addressable efficiency is smaller I wanna have a larger share because there's a minimum cost involved for me to produce. so If i produced only hundred thousand dollars off-efficiency you know? I want to have a large share! If...I produce twenty million dollars off efficiency then im okay with the small amount right?

1:06:51 Because ultimately euros that hit my bank account are important, but in any case I want to produce it and i wanna reduce more cost than I create for my customer. But why wouldn't you just sticking to this topic? Why would then say hey its always twenty percent of what we free up on resources or like efficiencies etc.

1:07:17 because they will be scale with the tons efficiencies that you can. Right now, it is always an FTE-based logic so that's quite scalable okay? So its' always a FTE based logic. the question your asking me as of the efficiency that you generate what's exactly percentage and why i'm not directly responding to that question is because sometimes the efficiency bucket can be extremely large in which case we'll take a smaller share let's say only ten percent.

1:07:53 but if the efficiency budget that regenerate, just like why would you vary when you could say, hey we always take the twenty percent and then for some of course it's a bit smaller far worse. It's even bigger because let's say you could go with twenty percent. Why? Would You Go With Ten? Well I Could Go With ten if The Let's Say Addressable Efficiency Is Ten Million Dollars.

1:08:34 i'm Fine with A million Dollar ACV. Of course it's an amazing outcome, I understand. Just in Euro terms and i can pass on more efficiency like the larger work flow. obviously there are more economies to share There is more dollars to share And we can on a percentage basis pass-on more because we also have our minimum set up.

1:08:56 We are cost with off over deployment engineers, which is where the discussion that we had yet right now comes into play. write. if you don't abstract the cost of an forward deployed engineer's very high. If you cannot generalize and if he would not ever use case focus uh... and so how do you marry sort of your costs width the size of the efficiency that you deliver with the ACV, it's often not a straightforward answer.

1:09:25 Also each workflow is different in a company and so you need to figure out why does this company only have let say twenty thousand POs but fifty people working? And then there are other companies ten people working. And so FTE is usually a great indicator of the complexity and the quantum work, we take that as yardstick.

1:09:55 but to arrive at final pricing then there are approval flows within company needs to take it to board or review price. maybe they want renegotiate. It is not an e-commerce style. You get it for nine ninety nine, um you know do whatever you want kind of pricing. there's a customer motion involved and in that customer emotion we wanna optimize with the business case That the customers has.

1:10:25 We want to optimize For what we also Get. Of course we want To build A large Business Ourselves What Is The Efficiency That We Take On? And We Want To Optimize For Making Sure That The Efficency Lands. The last thing we want to do is build another this year next. DIY agent builder. now you can build your own agents.

1:10:46 You know just press the buttons and then whatever the outcome is. I don't care about me on the out come. i think that it's very fundamental, And so I think all of these different layers, they need to come into pricing. We're trying to standardize pricing as much as possible. we were previously thinking about let's just price by the microstep Of transactions that we take over.

1:11:19 Let's say purchase orders invoices sales orders whatever transfer orders freight orders Whatever. But we felt like that often does not always correlate with intensity of work that the customer has to do, and by company-by-company. The same process may be more complex or less complex depending on whether they are custom fields, custom specifications technical things that need to be validated uh... And so where we have come to in pricing is what we call okay how many people working?

1:11:47 We can deploy X percent autopilot in x weeks. That's the basis for the pricing discussion. but Usually whatever number comes out is a very nice number, I think with software margins. This gives exactly the insights that i wanted to have on pricing because it's like... That's like the actual discussion at everybody in a founding team or In the team in general needs to have on like how are we pricing for future.

1:12:19 So therefore I want it understand okay what's possible, whats not? What are the points that coming into discussion? and so therefor thank you. i think this helps a lot too. yeah And we're evolving. business is year old fair enough right. maybe uh ya know will have catalog at some point. buy I don't know logistics use case X for five ninety nine.

1:12:47 Probably probably not be that case, but i think right now pricing is there multiple pressures to pricing. what's the business case at the customer? What's the dynamics of the customers as a political element and are they any minimum thresholds? so if my price from ninety K do we get into an approval flow at the customers end where they have to go through the board or price ninety-five K, you know that person can just take the decision.

1:13:15 How complex is the workflow? That I'm automating? what's the share of autopilot i think i can deliver? What's the addressable efficiency? how quickly Can I deliver? Is this a core use case or is just let's say what we call an expand on r&d. Use cases as the new use case that we want to do, but We haven't fully platformized yet.

1:13:38 how much engineering resources are going to be deployed? How much for deployment? you know there multiple aspects and I think once we are hundred deployments down our two hundred deployants down. i think you know more repeatability will also come to our pricing thinking. but i think we've iterated quite a lot and I think the best win when were able to score with customers is if we do.

1:14:02 FTE based thinking that doesn't mean that fte's are then actually going to, whatever the company does what the efficiencies they're locked it on. the company put its usually at these people. can them focus on higher level judgment heavy negotiation heavy Worked. that usually only humans can do right now.

1:14:23 Thank you. last question for this session today, but What needs to happen? To You can choose a number either reach hundred million in ARR or billion and they are like depending on the points that you want to make For you. Like what are the hypothesis that you need to prove? What other things you need?

1:14:45 two figure out And whatever thing. so just need to keep doing. I think honestly a lot has already been figured out. There is massive demand. I think there's no constraint on demand. um, i could produce Uh ten twenty thirty fifty million dollar pipeline today. Um for different workflows. So there is no constraint of demand.

1:15:12 It really and also depends on the acquisition channel. I think certain acquisition channels work in Europe versus other work in the US, but i think right now? I Think there is ample amount of opportunity. You know we don't you know. for eight out of ten conversations We don't even see competitive pressures.

1:15:35 Let's say five out of Ten Conversation. The Other Five you Know it's like oh we are going to do it in house. then you convince them it's not possible. they come back. Maybe another one or two say, hey can we not do with the suite that already have? Another one to say I'm speaking through a company doing something is like there are so much in very basic workflows.

1:15:57 um... That i feel like.. You wouldn't assume because obviously as an adventure ecosystem everybody has been doing sort of. There're couple companies doing workflow automation. But if you speak to those companies, I think there is so much opportunity. So it's... There is no demand constraint for us to reach a hundred million which i think is possible in the upcoming quarters.

1:16:28 Very possible. we need more workflows that we cover. It's a question of more workflows. We need to have faster deployment, right? Despite multi-industry, multi workflow, multi ERP. So how do you deal with that fragmentation? we Need to make our forward deployment engineering extremely efficient. A forward deployed engineer cannot get stuck in a company.

1:16:57 It's the question of generalization. Um...we need to make You know everything less high touch. That is the question of scalability, it's not demand. It is about how do you capture that demand without getting stuck in a company? And without taking learnings and building near productized solution... ...that may just need configuration but not forward deployment.

1:17:25 Forward deployment to me sounds too much! If I get stuck on one deployment with single tenant For one company and build custom for them, I will not build a large business. It would create problems. there'll be quality gaps customers will not be happy. it'll require A lot of headcount to forward deploy And i don't think that math really maths.

1:17:49 Um...I Think To capture the demand You need to platformize you need to deploy quickly you Need to generalize quickly? You know, you need to get customers to essentially execute on those. So I don't think there is any major blocker on the path to a hundred million now from one hundred million two billion?

1:18:16 Yeah no i think we'll cross that bridge when uh... When we got their. but right now it's more use cases platformizing more quickly or customers for which they're as demand and the technicality of how to deliver high accuracy great features that being somewhat adaptable to the idiosyncrasies of the company via memory or like in a kind of self-serve way.

1:18:45 And then, you know deploying very quickly and making these seeming sort opposite vectors go into the same direction are they seeming is out of conflicting goals bringing them together. I think it's more often engineering execution deployment question, then it is a demand question. But I think if you crack that which i think we are moving towards... This our segment will have its legora moment.

1:19:17 Sounds like exciting times! Yeah thankyou for taking the time for the podcast and link your LinkedIn profile and also general mind below in the show notes. um.. And I hope It's not the last time. Let's find another point in time when things happened to pick up the conversation. Thank you so much. That was my conversation with Tushar from General Mind, three things I want you to take away.

1:19:44 first hard work doesn't make you fast. everyone works hard. what makes you fast is a team that works on many things in parallel and trusts each other to deliver. thats why general mind has six co-founders. Second, forward deployment only pays off if it turns into a product. If you build everything custom for one customer You end up as a consulting business stuck at the few million in AR.

1:20:11 The number that matters is how quickly the next customer gets value. Third sell an outcome not seeds. general mind prices based on How many full-time people work? On our workflow today to customer should save Many times what they pay. If you know someone who's building AI for Enterprise right now, send them this episode and tell about Unicom Bakery.

1:20:32 And if haven't already subscribe to the podcast so that.