Varsha Uday Ubanu (3:23)

So exciting. Thanks cj. Thanks for having me on.

CJ (3:26)

I'm hoping you can illuminate this data labeling space because everybody's talking about it. It's all the rage. But me and other CFOs are trying to figure out what's what. Maybe you can take a minute to tell us about your company and what it does.

Varsha Uday Ubanu (3:39)

Yeah, for sure. So I work at Invisible and you know, Invisible, like a lot of the other people in the industry started doing what people broadly call data labeling, which is in Silicon Valley parlance, hot dog, not hot dog. But even then, what they were really doing was effectively using reinforced learning with human feedback. They were providing human feedback every time a model said this is hot dog, this is not hot dog. And I think that is what it started with. But now if you think about where we as invisible are today, we are in like what we would call the enterprise AI solution market and we sort of like span across. We work at the core of the AI tech ecosystem where we're helping every model builder think about what it takes to train their model, what data it takes for them to train their model for like the next frontier, so to speak. And then we take that theory and then we move it to application and work with enterprises to try and figure out how we deploy AI for like real production workflows. Data labeling is at the heart of it. But if you were to think about like where the world is really using AI today and you've seen a lot of research, like, you know, 95% of like AI projects fail and you look at what consumer and household adoption of AI is, it's at like 40, 60%. And that dissonance Is really explained by like what we see when we work with like enterprises because the enterprise adoption has not really kept up like for your day to day use cases, your and my use cases where we're like, so who is the President of the United States at this point? Models. Great, they'll give you a data point. Take that and think about it in the case of like all of the enterprises and it is, hey, I want to do my book. Reconciliation at the end of the month fails dramatically. And that's really where invisible comes and operates. We're like, we take all of that incredible model capability and then we move that into like enterprises and give them a system that they can actually trust, validate and run at scale.

CJ (5:34)

Cause what I struggle with is I'll go to a website like ServiceNow and you go to, you go to their site and it uses all these buzzwords about workflows and AI and enterprise and you're like, I don't know what you actually do. What I'm trying to figure out is like there's cool companies out there like invisible. And what it seems like to me is you have like almost these modules that you can solve all these complicated problems but you can almost back into what they're trying to solve for what workflow they're trying to optimize. Do I have that right?

Varsha Uday Ubanu (6:00)

I think you're absolutely right. Within enterprises, their use cases, like take a law firm for instance, their use case is very specific. The interesting thing I've realized with a lot of like gen AI use cases and gen AI adoption is that it's very often an end of one. What a company does is very, very specific to that company. And that will. A book close process for you is different from a book closed process for me is different from a book closed process for someone else. It is the same process. But I cannot bring one technology and say hey go, this is going to like out of the box work for you. There's a lot of nuances that you need to understand about a certain process. That's our sweet spot. We like the n of 1. Call it no, maybe not n of 1. As a finance person, I like n of 5, but at least the n of 1 use case where I'm like I will understand your specific process and I have all of these modules which we could potentially use and all of this expertise where we effectively trained a model to get better at this. Most decisions come down to a multi step reasoning process. It's like understanding what a human does and a human doesn't make One decision. It doesn't say cat, dog. A human says, I'm trying to figure out my spring break schedule. That's a multi step reasoning process. Oh, where do I want to go? Where is the weather? Great. What are the flight options? Oh, Japan. Great flight options. Too expensive. Backup, backup, backup. I think it's this step process which we initially had the foundational models do and now even in enterprise use cases that is the multiple decisioning. First it's about articulate what those decisions are and then helping build that decision at every layer. What are my variables, what are the nuances I need to be aware of, what is the context I need to be aware of and what is my decision making framework?

CJ (7:38)

Can you explain what a forward deployed engineer is? Because I understand that this is significant to your business today. And Varsha, you're speaking to these n of 1 scenarios. I imagine that the context is king and you have to understand what they currently have there in order to build something.

Varsha Uday Ubanu (7:53)

Forward deployed engineers are on like everybody's mind today. For what it's worth, I've seen a lot of commentary out there on how it's just professional services. I think the thing that is different about like Genai adoption and sort of going back to what I was originally saying in a lot of Gen AI adoption it is the same workflow that you're doing but you're actually adding the nuance of the very specific nuance of where you are, what you're doing and what your different parameters are. And that is in an out of box solution I can never understand. So and every enterprise in all honesty is super complex and it's a very bespoke process. Like I'll take the same book close process. It's an extremely bespoke process. They have five different systems they're pulling data from and those five different systems don't talk to each other. They have to make like some reconciliations. You need somebody to go sit, understand, explain explicitly, understand, build a solution for them using the modules that you have available that answers their specific use case in that and that is what the forward deploy engineering motion is. Let's take a use case. We work with all these law firms and helping them, helping them figure out how to mark up a document. The initial step is FDA's understanding when you mark up a document, what are the things that you care for? What does the model need to be aware of? What is the legal context? Where do you get the legal context from? Where do you get the sort of like the prior from if you will and Based off of that, I will fine tune your model and I will deploy it for you and I will continue to test it. How law firm A marks up a document is a little different from how law firm B marks up a document, especially on like certain circumstances. Like, and that circumstances is what the FD is actually going in there and doing any use case. And that's the FD's job. You have a team of FD's that go in there and really understand what that process is and how that workflow works.

CJ (9:43)

I've come from the software world where you have an account executive who's doing the selling of a systems engineer who's figuring out the technical specifications and helping with the technical part of the sale. And then you have a BDR who's helping to generate the pipeline. And then you have all these other resources. It kind of feels like all these roles are melding into one and how companies like yourself go to market. And this Ford Deployed Engineer is, is somewhat the output of it.

Varsha Uday Ubanu (10:09)

It is an output of it. It is definitely an aspect of what you had before across sort of like an AE and a systems engineer and a professional services. But their job is really to go understand and build a system that is durable and sustainable and delivers the impact that they need. The thing that is very different about this is you're not effectively selling them on a capability, you're selling them on an outcome. And the FD job is to deliver the outcome.

CJ (10:36)

And you're delivering the outcome before you even have a signed contract with them.

Varsha Uday Ubanu (10:40)

Not always, but you do it. So I'll take a specific example. Yes, prints. They're a very big way about how we go to market. It's something that I think Palantir has done that in the past. We do that and we do that pretty aggressively. And very often when we walk into enterprises, the thing that we find is business owners have sat with a problem or have tried to have their team solve a problem for many months at a time, sometimes 12 months, sometimes 18 months. And they've been trying to solve that problem in many different ways. They spent a lot of dollars on it and they haven't yet seen results. And, you know, the Genai promise has been around for a while. So it's not like people haven't tried and tested and piloted. That is what we encounter from business owners. And what we tell them is don't pay us anything until we can show you that what we're going to do works. We work with a company that sort of is like, has a Broad US presence. So when we say we do a solution Sprint, we'll say let's show you that in one market, in one market, what we're saying works. And then if you're happy with it, we'll deploy it across every single market. It could be trying to run some demand, planning for a market and showing them that our models will give you the output that works and you're happy with. And that generally takes us anywhere between six to eight weeks. Fundamentally what we have is we have all of these five modules that we've spent all of 25 building and we feel good about our ability to take what data you have, like bring all data together, which is our neuron platform, build an agent if required, which is our agent platform, or use human marketplace to effectively give them more signal marketplace. And like data labeling is such a overused word. But like humans putting humans in the loop. There is nothing like putting humans in the loop. They just make a process a lot better. They understand the context, they're able to train the model so much better. And then delivering a solution that delivers the outcome. The outcome you needed was I want to know what demand I'm going to have in this market for these different SKUs and how close are we?

CJ (12:34)

It's so fascinating and I love this new world we're in. Hey, thanks for listening. We'll be right back after a word from our sponsors. Well, well, well. Here's what nobody tells you about being a CFO. You'll spend 50% of your time on stuff that is killing your momentum. The best CFOs I know are business leaders who know how to drive growth in heroic family fashion. But most of us end up spending our days buried in manual work. I'm talking about collecting receipts, reviewing expenses and manually reconciling spend. It's painful. That's why CFOs need Brex. Brex built an intelligent finance platform that pairs corporate cards with built in expense management. Plus a team of AI agents to handle the manual finance tasks for you. That way CFOs have more time for the high impact project set, drive growth, you know, the shit actually worthy of your CFO time. Bottom line, Brex is automating hundreds of thousands of hours of manual finance work every month across 35,000 companies like Anthropic, Coinbase and Doordash. Ready to spend less time buried in expenses and more time driving results. Check out Brex app Brex.com metrics. It is Brex.com metrics please. Guys, how the hell do I have three kids in daycare Brex.com metrics you just launched your new AI product. The new pricing page looks great. I'm talking crisp but behind it, last minute glued code, messy spreadsheets and running ad hoc queries to figure out what to bill customers get invoices they can't understand. Engineers are chasing billing bugs. Finance can't close the books. Well, with Metronome, you hand it all off to the real time billing infrastructure that just works. Reliable, flexible and built to grow with you. They turn raw usage events into accurate invoices, give customers bills they actually understand and keep every team in sync in real time. Whether you're launching usage based pricing, managing enterprise contracts, or rolling out new AI services, Metronome does the heavy lifting so you can focus on your product, not your billing. That's why some of the fastest growing companies in the world like OpenAI and Anthropic, run their billing on metronome. Visit metronome.com to learn more. That's metronome.com Here's a growth tax nobody talks about Every new monetization model you ship creates a nightmare for your finance team. Ad usage based pricing. Now you're tracking consumption against commitments. Launch product bundles. That's multiple performance obligations per contract. Offer mid cycle upgrades. Good luck reallocating revenue manually. But that's exactly where Right Rev shines. Right Rev is the revenue recognition engine built for companies that can't afford to let accounting slow down growth. When your Rev rack is automated, your product team can ship new pricing without asking finance for permission. And your sales team can close creative deals without worrying about downstream chaos. To get up on my CFO soapbox for a sec, I love talking about creative pricing models, hybrid pricing credits, tiered usage, but I've seen too many companies where the sales team is celebrating a huge quarter while finance is still trying to figure out how to recognize half of it. In a world where your pricing model might change three times next year, that flexibility is everything. If you want to scale your monetization without breaking your books, visit right rev.com that's right rev.com where modern monetization meets Bulletproof Accounting what breaks my traditional CFO brain is I'm having this outlay of resources ahead of time, right? So you have technical resources that are now showing up in my go to market costs which which before wouldn't be in my CAC payback period. You assume that the close rate's probably going to be a lot higher because you're already showing them like wtf? Your product does exactly what you told me it was going to do. And then the retention to go and expand within the account to other problems that they probably have is also higher. So I guess what I'm trying to square is like it seems riskier but the upside seems so much higher.

Varsha Uday Ubanu (16:31)

The upside is absolutely. And you know you're bang on right. The minute you do this and you can prove that I can actually deliver the outcome that you were looking for, you're in use case one. The thing that we're selling is we're selling trust. I run the deal desk. I think a lot about like value sold value delivered, how long, you know, how much can we keep? It's very top of mind for me. But like the reality of enterprise deals when we're doing a deal, one, it's multi millions of dollars, they have so many different workflows that they have been trying to improve over a very, very long period of time. So first of all, the opportunity size within most of these enterprises is massive. And two, they are very bespoke. So most things I will tell you about like all of this will feel very generic. But ultimately what we it ends up being is that we find a business owner and we know their problem statement and we deliver the result that is really going to help them. Think about it as like this is something I've been trying to solve for so long. I have $5 million of cost, I have sunk on it and I am yet to deliver results on it. And that is the problem statement of most business owners we work with. And if we can convince them that we will deliver, we will be that partner that is going to help you deliver the outcome for yourself, help you sell us to the broader buying committee, whatever the buying committee is and like help you talk about it to five other people in the company who also have problem statements. We know we're set for like multiple years. So that initial eight weeks of investment that we do feels so small, the

CJ (18:03)

initial investment, it pales in comparison to the potential upside that you can get with these trusted customers.

Varsha Uday Ubanu (18:09)

That's the beauty of the enterprises. I think you will see most enterprises see that once we put AI into production, they try it for a couple months and these are long cycles. We're playing the long game. You're not playing the game for like I'm going to put in some money this quarter, I'm going to get X money next quarter. No, the payback periods are very different. Even when we actually deploy AI solutions for a client, they try for six months, 12 months, they're very happy. They're like, oh, here are five other use cases. Could you help me get into these five other use cases?

CJ (18:38)

Usually deal desks have SKUs and it comes across and you say, I approve this at this amount. You know that discount looks too big. Do you even have SKUs at your company? Because you're solving different problems.

Varsha Uday Ubanu (18:50)

That's one of the reasons why the deal desk runs the way it does. I think I do think of every single deal as a bespoke deal. What is the opportunity size we're going after? What is the value they have in tables? So for example, there is every AI problem is first a data problem. With enterprises, the biggest problem they have is a data problem before they actually have the AI problem or they have an AI problem, but the solution is in the data. What it takes to actually bring all of your data in one place might be very different from what it causes for the other and how much it would cost for them or how valuable it is for them versus us. So it's web very value based pricing. So it's one of those interesting things I say a lot about, like yeah, we do value based pricing, etc. But it works. It works very well in theory and it works selectively in practice. You have to actually know where it, where it works, where it doesn't. And at the end of the day the main thing you have to remember is enterprises want value. They, you know, they want value, but what they're buying is predictability. As long as you're giving them a pricing model that is predictable, repeatable, they feel good about being able to pay that much money to you all the time and still feeling like, hey, I used to spend $10 million on this before. They're going to give it to me for eight as much as like outcome linked and all of that. But I think at the end of the day what they're looking for is value. But buying predictability is how I think about it.

CJ (20:07)

Because the way that my simple mind works is traditionally there are three different ways to price something. Value based pricing is the hardest because you have to figure out what it's worth to somebody and that's difficult to get to. The other way is to do cost plus and just mark it up. And then the other way is just to take whatever your competitor's charging and you know, make it something similar, maybe a little bit lower than that. Can you just talk to how you go about value based pricing because you're solving different problems, which means there has to be a different value Metric depending on who the customer is.

Varsha Uday Ubanu (20:37)

Yeah, absolutely. So you know, value based pricing, it makes total sense we should be doing value based pricing. But the challenge in a lot of enterprise AI use cases is that the value is rarely immediate or cleanly attributable. You can't effectively and you know, the biggest upside isn't always going to be, hey, I'm going to give you X amount of incremental revenue. It's very hard to meter price. Sometimes it's like risk avoided, sometimes it's failures prevented, sometimes it's time saved. So I think the, the what we want to effectively be selling is the kind of value that compounds with time. Like it's very hard to meter and with like, you know, at that moment of purchase you can't. But it is value that adds with time. That is why I talk about multiple use cases, about multiple different problem statements that they have for years tried to solve and not been able to get to. When we think about like when we sell and like when we are actually prioritizing work, we effectively understand one minute use case. What does it take for you to see value from this one use case that we're helping with this really minute, Let me scope it to be really tiny, however tiny it is, and let's ensure that we are effectively coming up with a pricing structure that's very transparent to you and defensible. We're telling you, hey, it used to cost you 6 million before we're going to build this, deploy this and run this for you. At 3 million, you have 50% ROI out of the gate year one. I do expect most of them to look like a flat fee. So you either, you just replace it with a slightly different cost base and then you say I'll show you value and as we go along we will be able to meter it better. We will be able to actually measure what is the upside you're seeing or what is the downside you're preventing. And like, as you improve metering, you can actually tie a lot of this fees into an outcome.

CJ (22:18)

Stupid question. Is there a difference between value sold versus value delivered?

Varsha Uday Ubanu (22:24)

There is and there isn't. I think about like, you know, we sell on the art of the possible, we deliver on the art of reality. And then there's a little bit of like bridging one has to do to effectively see, hey, this is what we saved on like value sold and this is value delivered. Which is what I mean by like year one is really our initial year to identify what that value is. Everybody has a thesis on what something is going to do. Nobody's done this before. Remember if they had already done this, we aren't coming in here and saying let me replace player A who used to do this for you, and I as player B will deliver this to you. Well, it's very clear what is possible here. We are often discovering what is the art of the possible, what is the actual incrementality. You will see what is the value upside, what is the additional revenue you do unlock. We work with like a very large consumer brand and they have like all of these menus across the country and very often their brand is misrepresented. Their brand being misrepresented means revenue being left on the table. We don't have a counterfactual, we don't know what is the revenue upside. You're going to see if the brand was rightfully the rightly represented. As an example, how do you do value based pricing? What is value delivered? What is value soldier?

CJ (23:31)

Varsha, you've spoken about multi threading opening up more opportunities inside an account so that like eventually one of them sticks. How do you balance short term pipeline activity that might not show up in revenue yet with, with the long term bets that you're making?

Varsha Uday Ubanu (23:46)

I honestly think of enterprise revenue more as a portfolio of bets versus a funnel. Like I think we've traditionally thought of like sales very much as a funnel. But I think enterprise revenue, it's like a bunch of bets that you make making a bet by saying most cases. Enterprises explore multiple AI use cases in parallel. They don't always have one use case they're sitting with. They're like here are the five places I think AI can be helpful. Most won't scale, a few will matter immensely. So I think a lot of our initial exploration is really getting learnings to say this use case, this is the actual value they can deliver. This is the use case they can. And I would say that like multi threading is one helpful because it helps us create that optionality and that optionality is heavily powerful. But on the flip side an unmanaged optionality also just quietly kills the business because that's just like optionality with like no discipline is just burn. Like it's, it's another word for burn. So there's a lot of like closing the loop very quickly. Eight week solution. Sprint, beautiful example. Again making a very time bound investment to say we, if it works you pay us, you give us a year long contract. If it doesn't, don't. It also means that I have all of Our cost tied in that eight week like very well defined period.

CJ (25:00)

Any other advice for finance leaders out there who want to use this multi threading strategy? Because it seems amazing to go out and find all these different opportunities. Hey, the first one may not be the home run, but once we're in the account it's going to compound in terms of how much we know about them. How should they size up the costing? I know you said you put it into almost like this eight week block. What else do you think about when you're proposing going into a company?

Varsha Uday Ubanu (25:25)

I think there's a little bit of like how big the use cases and how achievable some of these use cases are. So I'll take the example of like we worked with a construction company and then as soon as we walked into the construction company, they had so many use cases, so many things that like, you know, as you can imagine in any large construction company, their systems are ancient. You know, the world is massive there. And then you basically start like you, we walk in, they sit down, they're like, Here are the 10 different things I've been trying to solve. Go ahead, what do you want to pick? So I think it's a lot of knowing how big your opportunity is is the biggest important is the first thing you need to do before you even start. Which is why I take the construction company as an example because it's like we can work with them for the next decade in just like improving all of their process. There is that much work and that much potential in that. Take a large tech company. Tech companies have like smaller problem statements. It's also a little bit more on what the company is. There's an internal stat that I hear from a lot of our sellers when they go into a company. They're like, you know, we always like the challenger brands a little bit because they don't truly believe they have the appetite to go and do a lot more because they really want to win, they want to invest, they want to get to number one. And that just like changes how many bets they themselves are ready to make and what amount they are ready to like put out there in order to effectively win. So I think it's a little bit that is a, that's not as much like what we do, but as much as understanding the market context and what is the size of the opportunity there, which is very important. Before you go in, I'd say most enterprises that are in the Fortune 100, Fortune 500 have enough and more that you could potentially like work on. For next two to three years.

CJ (27:06)

Hey, thanks for listening. We'll be right back after a word from our sponsors. Being a cfo, you know how much I love tools that actually make the lives of accounting and finance folks easier. One of my favorite tools right now is Rillet the AI native ERP Going head to head with netsuite yes, someone is finally doing it. I met Rillett two years ago when they were still in stealth. Since then they've absolutely taken the finance world by storm. Their mission is to make the zero day close a reality. And they're actually doing it. Customers are literally closing their books at 1:35pm on the first day of the month. They've got everything you need to scale your business. Complex revenue recognition, native integrations, custom reporting, multi entity close management, and much more. They're only a few years in and are already supporting NASDAQ Publicly listed companies. Yes, seriously, if you want to scale your business on an ERP that wasn't built in the 90s, you need to check out Rillit. Book a demo@rillet.com CJ oh cool, that's me. That's R I L-L-E-T.com CJ R I L L E T Tell them I sent you there. Ali and the team at Tabs are building something that directly addresses one of the biggest headaches I see finance teams deal with today. Pulling data from your erp, your CRM, your FPA tools and your usage systems and actually making it all work together. If you're running usage based pricing or complex contracts, you already know how this usually plays out. Contracts Level 1 system usage data lives somewhere else. Billing happens in another tool. Revreck gets layered on top at month end and finance ends up just reconciling everything by hand to close the books. This is exactly the problem Tabs was built to solve. Tabs is an AI native revenue platform built for controllers and CFOs. It brings together data from your ERP, CRM and real product usage into a single system of record. From there it automates billing, collections and revenue recognition without Finance having to duct tape workflows together every month. If usage based revenue is core to your business, this is the future of how billing and Revrect get done. Go check them out@tabs.com run that's tabs.com run we've all been burned. We've all bought that enterprise planning tool that promised the world, only to realize six months later that we've basically taken on a second full time job just to Keep the software running. I want to talk about a company that was built specifically to kill that cycle, abacum. I actually remember my very first conversation with their founder and CEO Julio Martinez over three years ago. Back then they were just starting out in Spain. Fast forward to today. Julio moved the home base to NYC and they're the engine behind finance teams at Strava, Replit and JG Wentworth. Abacome doesn't turn you into a software admin. The integrations are actually self service. You don't need a $300 an hour consultant to plug in your ERP or HRS. And they're doing AI in a way that actually impacts the things FP and A teams are doing every day. Things like creating variance summaries, building formulas and modeling scenarios. If you're scaling fast and you're trying to avoid that legacy platform trap, ABACOM is the move. They're building the future of our tech stack and they're doing it with a CFO's perspective. Go to Abacum AI to see it for yourself. That is Abacom AI. Varsha, you'd mentioned earlier that having a human in the loop, especially a human with rich business context from being a practitioner just makes the product so much better when you pair it with AI. I assume you have to pay these people. How do you think about incentivizing them,

Varsha Uday Ubanu (30:46)

compensating them based on what the context is, based on what the expertise is? You have to like pay them fairly. I think at the end of the day it all comes down to what we have learned by working on this for like four or five years is if I were to find a lawyer for $130 an hour versus a lawyer for a $150 an hour, this is what the 150 lawyer is able to do, this is what a 130 dol dollar lawyer is able to do and this is what a $300 lawyer is able to do. What you can get from people at different price points is what we have learned over a period of time. So I think the, the fees are very based on one their contribution and their expertise, skill set and experience. And you just like, like it's a market and that's the clearing price. You have to pay that to get what you need. But the interesting thing about where you use that expertise and how that expertise shows up, whether it's in the enterprise AI use case or like especially within enterprises, tech, healthcare, finance, legal policy, you're not training those models and you're not getting human in the loop with like generic Data and just hoping for the best. You want domain experts who understand every single consequence of a decision going wrong. Because a wrong answer in that case is not just inaccurate, it's actually very expensive. Like it can be very costly or risky. Which is why we think that the closer AI gets to a real decision, the more valuable the human expertise becomes.

CJ (32:04)

I like what you said about the market clearing price. If you needed advice from a heart surgeon and they were going to charge you $50 an hour, you're probably going to get what you pay for.

Varsha Uday Ubanu (32:12)

For your and my own health, we wouldn't call a person who said, For $50 an hour, I will give you a diagnosis.

CJ (32:17)

Now, how do you staff an organization like this that relies upon third party insights and third party contractors? Is there a team that just goes out and gets these people or is it there's a liquid marketplace to find them.

Varsha Uday Ubanu (32:30)

Today it is a pretty liquid marketplace. Whether you're a foundational model that is trying to like get better at the next frontier or whether you're an enterprise that expertise changes. So it is a heavily liquid market. And it's a lot of the muscle. The organizational muscle we have built over the last three to five years has been how do we find the right kind of people with the right kind of expertise in 24 hours? Because that is the need of the job. The need of the job is I want 500 financial experts who can tell me everything about M and a negotiation as of tomorrow. How do you find, where do you have to pay, what is the market clearing price to actually be able to get that? That is the organizational sort of muscle we have built over the last couple years.

CJ (33:13)

And then does AI interview them to get these insights out? How do you extract them? Or do you have a staff of people who are trained to ask these questions?

Varsha Uday Ubanu (33:20)

We've built systems that help us figure out assessments based on expertise so we actually have some form of certification. This doesn't scale if I have to staff humans to run every single interview, because every time a single interview has to be done. And there is also, there are some places where we have to heavily invest in a domain expertise. Take audio for instance. Like, everybody wanted to train their like multilingual language. We need audio experts there. We need like linguists there who can effectively understand as a system. Can I tell the difference between. And I will take the India example because we get a lot of like audio requests for different dialects in India. There are only few people who understand that like, you know, no test is going to say, okay, is this person's Hindi from Bihar and they care about it because it's very different. For the person who actually can hear it, it's very different. But people who don't hear it, it sounds the same.

CJ (34:10)

I want to transition to talk about metrics a little bit because historically software or service based companies would lean on metrics like ARR and NRR. In the context of a high growth AI labeling company, what do you think of the right KPIs to track data labeling space?

Varsha Uday Ubanu (34:25)

Isn't ARR. We've never marketed that business as ARR.

CJ (34:28)

I'm glad you said that. And that's okay. It's totally okay. But there are some companies out there who are calling data labeling ARR.

Varsha Uday Ubanu (34:36)

Yeah. And I don't particularly like that because I think it just paints us all with like a different brush which I don't particularly appreciate. But like on the other hand, the enterprise platform business that we have is within an enterprise. We have five modules, we have forward deployed engineers that go in sec up a working solution on top of that platform. And when we get commitments, that's like year two, year three year commitments for like, hey, this is how much we're going to give you. And when we talk about revenue numbers, we are very hyper clear about, okay, this part of our business that is ARR. In our labeling side of the business, it's revenue. It is like I, I can bet on the fact that it could potentially be reoccurring as long as we get a certain set of things right, but it isn't arranged. And you know, also again, at the end of the day, AI that only tells you what's happened. It doesn't tell you whether your business is working, whether it's durable, whether it's actually going to keep growing, et cetera. And that's really where I think, I have been thinking a lot more internally when we think about metrics that really define where the business is headed. I've been very focused on sort of like momentum based metrics. Like, and you know, when it comes to momentum based metrics and specifically around like, is the customer happy with the current use case we're doing? Are they happy enough to actually give us more use cases to work on? Are they actually going to go push us towards other people? I think we have one use case up and running and the only thing we're focused on is easy driving value for the client. Because if it is driving value for the client, we're golden. We have more opportunities we will be given.

CJ (36:01)

I love the concept of momentum the hard part is it's difficult to compare from company to company and you probably have to come up with your own definitions with internally based on what your product is. It's really cool.

Varsha Uday Ubanu (36:12)

You know, the way I think about momentum is that when effort sort of like converts faster than it's been used when, than it's used to. When I've gone into sell, it's taken us three months now it's taking us two months because we have something going, there's like momentum going. I think it comes back to like the point which I truly hold, like we will hit scale when like the marginal cost of our business grows slower than the marginal value we're driving. So it's like very clear that I want to get to a point where we want to continue to deliver value to the client because as long as we're driving more value than the marginal cost, we're good, we're set.

CJ (36:48)

And you've worked at a bunch of other tech companies before. I mean, are people, myself included, just making it too complicated? Like at the end of the day, the whole goal of business is to make more money than you spend.

Varsha Uday Ubanu (36:59)

As SaaS picked up, we got so used to payback periods and you know, we got used to seeing a certain set of metrics which we were expecting to be the certain like exactly the same way. And I think the thing that Genai has done is it's basically changing the market a little bit. And I just feel like we haven't found those what I would call my back of envelope numbers, which would tell me, okay, this business great. This business is not so great. I think we as an industry haven't found it. And that's really what you see like people trying to find metrics that work very well in a different setup, like force onto this. Yes, they're great metrics and they potentially give you a way to actually understand a business well. But that doesn't give me a way to actually run a business well. And when I am running the business, I care for metrics. That is going to say like, am I going to hit my numbers? And how do you run that game? ARR isn't going to tell me that.

CJ (37:51)

No. And actually never has for any other company that that's an output metric. It's the output of all these other, of a million other decisions that went in. It's better to focus on what I'll call momentum, but it's a series of input metrics to produce it.

Varsha Uday Ubanu (38:06)

Absolutely. And I like the input metrics. I like being able to see it before something shows up in like churn or margin compression. I'd rather know. I want to know if the system is compounding or the system is decaying and where it's compounding where it's decaying. So we can like cut the losses before it starts getting too big. So I think yeah, momentum's great. Like for what it's worth, I try very hard to like be able to define moment. For us, momentum is very much use case expansion. At what rate are we able to land use cases and at what rate are we really like expanding on use cases? And how long does it take for us to land the first client? We really started going to the enterprises with this motion sometime like March last year. It took us a full six months before we were able to like really convince a client that yes, we know what we're talking about and let us actually come in and solve it. These days, two months we've gone in, we tell them we understand their problem statement. We tell them, okay, here is a problem statement. We will pick and we'll help you solve this. Give us eight weeks, we'll tell you, we'll solve this for you.

CJ (39:00)

And you've looked at a lot of different PNLs from all the different tech companies you've been at or advised. Do the P Ls of these new gen AI companies or companies that are involved in data labeling, do they fundamentally look different? Do you have to take a different lens and you evaluate them? I'll give you an example. You can't look at gross margin the same anymore. That's just not going to look like what you're used to and that may not be a bad thing.

Varsha Uday Ubanu (39:23)

I was saying I think all of the investor like metrics that we used for businesses are getting reinvented right now because I need a high gross margin business with like high NRR and like, you know, I'm sad that's the kind of business I want to go after. I don't think some of those things truly have played out yet. So yes, you do have to look at it with a different sort of lens. I don't have clean metrics today to say this is the lens I would look at. The things I focus on is at steady state. Like take all our enterprise deployments. One of the things I think about is once we have gone in and solved a problem client's problem statement and deployed something, what margin can I run that at? So there's a little bit of like what is my steady state gross margin that I Care about it's use case by use case. Because if I'm doing like every three months I open up a new use case. I'm always in the build phase but on a different use case. So I think that's something that we focus on a lot. How much investment do we have to make before our first revenue dollar comes about? Is something I think about a lot. What's our ARR to fd? Is something I think about a lot because I'm like, okay, what is the productivity of the organization? Like, how many FDEs do I truly need? And is that metric growing, reducing? Like I do think about that.

CJ (40:31)

Varsha, when you think about the atomic unit that you use for headcount and resource planning within the org, you probably just went through another annual planning cycle. What is it? Because the companies I've worked that it's usually the account executive trying to figure out how much quota they can have. This is a much more technical sale.

Varsha Uday Ubanu (40:47)

Yeah, this is a very technical sale. So that I was talking about the ARR to FD as an example. I mean, at the end of the day it's a very technical sale. And sales planning is one of the harder things I've had to like really rethink because you don't do annual planning cycles where you're like I have 10 seller, I have 10 AES and each one can carry a 2 million quota. It's also funny in enterprise AI, this year's revenue, when I think about 2026 revenue, for instance, it's very little of what my AE is going to do.

CJ (41:13)

You already found the client?

Varsha Uday Ubanu (41:15)

I already found the client. It's very much expansion versus like new. New is great, but new is great for latter half of 26 out of 27. Like it's, it's what is needed to like set my 2027 up. The timelines have also shifted. It's become how many clients are you at and what is your expansion potential within each one of them? That is the most important atomic unit. And then like in order to address that, how many of these do you need? How many of these do you have?

CJ (41:39)

I empathize with you because if I was to try to come up with a typical, you know, six box or four box splash slide, which metrics that I put on there to help convey that we have momentum. Some of those would be the output metrics, but ideally they're things that allude to your pipeline and the durability of the accounts you already have on board.

Varsha Uday Ubanu (41:57)

The pipeline is, is definitely something we stare at all the time.

CJ (42:01)

It's a brave new world. I do want to transition though, to talk about a world that you were in before and that's ad tech, which I find immensely fascinating. And given your background in ad tech, mobile advertising, I'm curious how that experience informs how you think about marketing and growth from, for now, B2B, AI data companies.

Varsha Uday Ubanu (42:19)

It's very interesting because I think one of the things that I took away from like ad tech is like how much distribution truly matters. It's one of the things that like was very, very important for me as I came in and it also taught me how quickly markets can commoditize. Like ad tech commoditized overnight. It felt like, you know, 2012, 2013, it was like, oh, the hot thing. 2016 is a commoditized thing. In AI and AI data, there's a little bit more that is at stake here because your failure is a customer's model failure here. I want to say that when I think about how we are marketing and what we are focusing on, that's where the role of like brand marketing for us has like gotten a little bit more important. I mean, one enterprises aren't like buying hype, they're buying confidence. And very often confidence here in a startup is confidence in the CEO and what the CEO knows. When you think about a startup, I think we start thinking about how the company and the CEO founder, sort of like the brands overlap. So how do you ensure that they can trust? They all sort of like intertwined. We're not running B2B. Okay, here is outbound marketing. Here is X number of dollars we put on outbound marketing. So it's like very different. Our marketing motion in itself is 100% different.

CJ (43:32)

I'm so glad you brought up brand marketing because our mutual friend Angela, who you work with, she's been a guest of the podcast and she was kind enough to introduce us. She said Varsha is remarkably refreshing as a finance leader to work with as her marketing counterpart because she understands the brand halo. Can you just say more? Because I think the importance of brand in the age of AI is even more important than it was before.

Varsha Uday Ubanu (43:57)

Yeah, no, it's actually very interesting. I think the brand today matters so much more, especially when we go into like all the enterprise, into the enterprise segment, the enterprise buyers are inherently risk averse. They want to know that you'll be there and you will take reliability seriously. That is the most important thing that they care for. If your brand did not exist, you can't as a seller show up in that company and then try and convince them, like it's such a uphill battle to actually convince them that you will survive, you will actually deliver the outcome for us. Brand actually reduces friction across sales, across procurement, across expansion. And from like a finance perspective, honestly, this is. It's loading my cac ton, it's loading my risk. So it's very easy. These aren't soft metrics. This is just pure economic metrics. For me, when we start thinking about, oh, can we afford, like, you know, a general question is, can we afford brand marketing? Are we early enough to brand marketing? I think the actual question for us is, can we afford not to? Because that is the only thing that helps us go land enterprise AI today. Yes, me understanding marketing is an important part of it, but it is also. That's the world that we're, that we are in right now. The only thing that truly matters is

CJ (45:09)

brand marketing, because a lot of finance leaders are listening to this podcast and they're saying, okay, I'm not of a revenue scale yet or even close enough to profitability to be thinking about brand marketing as if it's like somewhat further down the path. Listening to you reflecting on it, it sounds like it's a rising tide that lifts all ships. So the attribution may not always be there, but you're greasing the skids for other things.

Varsha Uday Ubanu (45:29)

We want to effectively start prioritizing things where you talk about the brand and the work that you do. And so it's sort of like gone hand to hand. Like Angela will tell you many things around, you know, brand as a voice is dying. You need to think about, like, how you build a brand, et cetera, like, think. At the end of the day, it's come down to when people think invisible, what do they think about and does that give them trust? And which is why it goes back to my earlier point. We're selling trust at the end of the day, so. So we have to invest in building that trust. And a part of building that trust is to actually build a brand.

CJ (46:00)

Ad tech is a crazy space. Do you have any wild stories or just learnings things that stuck with you from being there?

Varsha Uday Ubanu (46:07)

I am acutely aware of, like, you know, I think unit economics day in and day out. It's like one of those things that I spent a lot of time thinking about. I think a lot about, like, market effects, my marginal scale, marginal cost, et cetera comes in from there because, like, you know, there's just so much of, like, that halo effect and market effects. I think it's helped me really think about what is most important. And, you know, sort of like going back to what I was saying earlier, when we bring all of those experts and like, you know, for helping train a foundational model, what is it that you want to differentiate yourself on? So for us, it's very much the credibility and the execution and the quality of the data that we bring on because, like, we know we only have to afford the once as a company, we prioritize. And I specifically put a lot more emphasis on like, quality is more critical than like us actually in that moment, short term making revenue.

CJ (46:56)

Sometimes I'm like a dog with a bone. Can you say more about when markets commoditize? Because I think there have been moments in mobile, in web development, like it's happened in other markets too. I'm just curious what you meant by that because. Because it sticks with me.

Varsha Uday Ubanu (47:10)

I think there's literature out there which says that, you know, data labeling can effectively be a commodity. Like, you don't really. There is nothing special there. You're effectively bringing a pool of people

CJ (47:18)

together if it's not differentiated.

Varsha Uday Ubanu (47:20)

And that is really what I mean by like, it could get commoditized if you do not think about exactly what you're differentiating on and how well you're building differentiation in. So that's really what I was referencing when I talk about, like, markets can commoditize. It's a very hot industry today and tomorrow it's commoditized. So you just have to be very careful about really focusing in on the value that you're adding and the differentiation that you're bringing in there.

CJ (47:41)

It's such an astute point because it's not just your industry, it's not just about data labeling. It's just how you look at the landscape and how quickly others can come in and copy what you were doing and offer it for a lower price.

Varsha Uday Ubanu (47:52)

Again, Invisible has like a book that we seven powers and like, you know, helmers, we talk about all the time. And like, the Toyota example is often quoted within the company where like, Toyota was very happy to say, this is exactly how we do it. Come see how we do it. But not a soul could replicate it because it was organizational memory. And I think a lot of where we effectively see differentiation and when it comes to like labeling and providing data at quality, is that organizational memory? Yes. What we simplistically do is find some experts who can effectively give you the input. Like if I were trivializing it. But that process and how you do that reliably short period of time is what the organization has learned how to do really well.

CJ (48:31)

Varsha, I'm going to take you into what we call our long ass lightning round. First question I ask to every successful finance leader is, what's an example of something you've messed up before on the job? It could be here or any previous role.

Varsha Uday Ubanu (48:43)

Very funnily for us, I actually drank Kool Aid for too long before I actually enforced economic discipline. I think about this example very often because this was one of those things that like now if I think about it, I'm like, how could I have ever done that? We had a consumer business. We used to rent space is how I would explain it. But like we were on lock screen, we were a consumer platform on the lock screen. So we would pay all the OEMs to actually be on the lock screen. And when celebrity brands became a thing, we're like, oh, you know what's a great place to actually show these celebrity brands? It's on the lock screen. Extremely expensive endeavor. It was super cool because at that time celebrity brands were a thing. Like there were some great examples of those brands having launched. And we're like, yeah, okay, let's go ahead, let's make this acquisition. It wasn't an acquisition, was like an acqui. Hire of a team that had done that. So we're like, okay, let's do this. Let's throw money into distribution. Let's effectively pay these celebrities a lot of money. Let's actually start having them sell goods. So we would be effectively ended up investing in like inventory and sourcing and returns and you know, the full commerce chain, which is expensive to run. And all along it was very cool. The ROI was just not there and it took years. And I sitting there on the as a finance leader was like, yeah, oh this is cool. We should do this. This is going to unlock so much without actually spending time on saying how long is the payback? What is the cost we are truly incurring? What do we need to get right for this to be successful? When we were able to list out what we needed to get right for this to be successful, which was about a year later, you know, because I didn't want to sit there and say no. Like I took my time. But I think all it needed was a constraint applied early which would have let us actually have long term ambitions of actually being in it. Maybe it would have let us really solve for that problem early enough. But yeah, I think about that problem very often, which Is why like our 8 week solution sprints now? I like them because I'm like, I am time bounding it. I know exactly for how long I'm taking that bet.

CJ (50:31)

Do you remember who any of the celebrities were?

Varsha Uday Ubanu (50:33)

It was in India. Yes, I do remember some of the celebrities. It was cool to do.

CJ (50:37)

That's the thing. It was cool. That's why it's compelling and it's hard to say no. Well, thanks for sharing that because I'm sure there are analogs to other people's roles where they're like, but we get tickets to the Celtics game. It's like, yeah, but we haven't landed a single sale since going.

Varsha Uday Ubanu (50:49)

It's one of those things that you have to be very clear about. I think knowing what it takes to get something right, screwing up really taught me really provided that clarity to say I think about what we're doing something for and if it's worth spending the amount that we're spending to actually get that right.

CJ (51:05)

Next one I got for you. If you could give your younger self advice, knowing what you know today, what would you tell her?

Varsha Uday Ubanu (51:10)

Cash is king. You know, you can use a lot of things. Revenue, gross margin, ebitda. They can be delayed, they can be adjusted, they can be re explained. Cash. Cash tells you how much time you actually have to fix your mistakes. This is very interesting because on our books we were like, oh yeah, we're making a lot of revenue. We're like great. And this is like many, many years ago I had just stepped into like having to run finance. It was like one of those like defining moments for me. The part that I actually didn't realize is like, yeah, oh look, we have so much, you know, we're making so much money. Didn't understand working capital. There is revenue. I'm going to take two months to collect it. And all it took was two of my top customers and two, we had a country to country set up. We had a China business. We had like a global business. China doesn't let you take money out. Can't pay bills. How much cash do you actually have in hand today?

CJ (52:00)

Cash is king. That's the whole podcast right there. Next one I got for you, more of a technical one. What tools does your team use to get the job done today?

Varsha Uday Ubanu (52:08)

I think a lot of it is like boring tools just aggressively well used. We are on netsuite. We use Data Rails as our FP and a tool. We have numeriq to help us with all our closes and nihilist to help us with like cash Forecasting did get nihilist numeric and data rails all within the last year. It's all something that we have sort of like invested in over the last year. The way I think about it is the tool isn't actually the power, it's the discipline of actually using it and doing it better and better and better. And like anything that will help me improve my forecast accuracy. Accuracy, even a slight bit is worth its weight in gold.

CJ (52:38)

What about expense management? Travel procurement?

Varsha Uday Ubanu (52:40)

Ramp is what we use for all of that for all of our expense management and DAC plus ClickUp is our procurement tool. Not ideal, but that's our procurement tool. Today all our HRIs is rippling.

CJ (52:50)

Last one I got for you. I bet there's one from the ad tech days. What's the craziest thing we've ever had someone try to expense movies?

Varsha Uday Ubanu (52:57)

I've had French lessons. I've had subscriptions. That should clearly not be on a company card. As important as hinge might be to someone's mental health, I don't think it fits as it fits into the wellness stagnant.

CJ (53:08)

That's a first. We've never had a dating app on the show.

Varsha Uday Ubanu (53:11)

That's a fun one. I will remember that for the rest of my life.

CJ (53:14)

Varsha, this has been an absolute pleasure. Thanks for carving out time for us.

Varsha Uday Ubanu (53:18)

Yeah, no, this was great. Thank you. Thank you for having me Run the

CJ (53:21)

Numbers is a mostly media production yelling an intro by Fat Joe. Artwork by Meg Delesandro. Sho is executive producer produced by Ben Hillman. Nothing said on this podcast is intended to be business or investment advice. It's the sole opinion of me. A guy who feeds his dog way too much ice cream and has a history of net operating losses. Lol. If you like this podcast, hit subscribe and give us five stars. It will take like two seconds and our algorithm overlords love it. Drink water, call your mom and have a great day. Peace.