How Tabs Is Building AI Agents for the CFO's Office with Deepak Bapat.

Welcome to Revenge of the Nerds. We’re skipping the hype and going straight to the builders. In this edition, we talked about:

• Tabs breaks revenue automation into three layers—a knowledge graph, a platform of financial primitives, and an agentic experience—each essential to maximizing cash collection and passing revenue audits.

• The accounting talent crisis (75% of CPAs nearing retirement, new graduates declining 30% over a decade) is accelerating demand for platforms that let a single controller do the work of an entire AR team.

• Vertical AI in finance wins over general-purpose tools because it can own proprietary contract data, build domain-specific intelligence, and ultimately stand behind the liability of a clean audit.

Let’s dive in. No floaties needed…

Revenge of the Nerds

Deepak Bapat, CTO & Co-Founder of Tabs

Deepak Bapat is the CTO and co-founder of Tabs. Founded in 2023, the company hit 5x ARR growth in its last year and closed a $55M Series B in September 2025 to fund its push into agentic AI for billing, collections, revenue recognition, and reporting.

Before Tabs, Deepak spent over six years at Latch. In this smart-access hardware-software company, he rose from software developer to director of software engineering. What makes him tick is a problem most engineers would scroll past: the messy, gray-area world of B2B revenue, where a customer's happiness with your product directly affects whether they pay their invoice.

Deepak doesn't just want to automate the spreadsheet—he wants to make the finance team's judgment scalable.

Give us a quick overview of Tabs & what you're building.

Tabs is a revenue automation company. The way I describe what we do is three layers: a knowledge graph, a platform, and an agentic experience sitting on top.

The outcome we're trying to drive is a world where any B2B company, once a contract is signed, doesn't have to worry about how to maximize cash against that contract into the bank. They don't have to worry about revenue reporting or understanding their financials. And they don't have to worry about what we call revenue compliance—things like ASC 606, GAAP reporting.

Fundamentally, the revenue automation platform is built on the idea that once someone signs a contract, the finance team's goal is to maximize the cash in the bank against that contract in a way that is realistically possible. Another way of putting that is minimizing revenue leakage. The second goal is being able to pass a revenue audit without any significant bumps along the way—all while minimizing additional human capital as your company scales.

Why does it still take finance teams so long to close books in 2026?

There are a couple of reasons. A lot of companies are trying to solve the financial close, getting to a continuous close or even a zero-day monthly close. That's all helpful and powerful. But when you think about all parts of the general ledger, we break it down into about three sections.

The first is all the human capital and people ops functions that go into your ERP and general ledger. Those are companies like Rippling, Gusto, and Justworks. The second bucket is everything from accounts payable to expense to procurement—those are the Ramps and Navans of the world. Those two buckets are actually simpler. Human resources is very rules-based: you must pay X, you must pay this social security, if you have a 401k, you must do Y. And in the AP and expense world, everything can also be very rules-based and works reasonably well.

On the revenue side, specifically, there's a lot more gray area. Revenue-related finance functions are, in some ways, a customer relationship problem just as much as they are a finance problem. A good example: if you have a customer who's not happy with your product and hasn't paid, how does that affect your ability to get them to pay? And then if you can't get them to pay, how does that affect the revenue on the other side when you go for your GAAP-compliant audit? You've sent out this invoice, and under ASC 606, there are guidelines regarding the likelihood that the customer will actually pay. There's so much more art than science on the revenue side.

That's why, even in 2026, the revenue side remains unsolved in the way AP has been solved, or the way procurement has been solved. Those problems are more standardized. You can build a workflow and it holds. Revenue doesn't work like that.

You've recently launched AI agents for billing & collections. How does that work in practice?

It really comes down to the data that informs billing and collections. A lot of that comes from an underlying knowledge graph where we've consolidated business entities across our merchant base into a single concept. Think of it like every business having a Tabs business ID. That's the first layer—the insights that drive the agentic layer. The second piece is the primitives: invoices, credit memos, email templates, and communication objects. You need to build the road on top of the soil. Then you have the agents sitting on top of all of that.

When a contract goes closed-won, the first job is processing it correctly and generating invoices. One thing we do really well is complex revenue—I think it's one of our differentiators. We handle everything from flat-price billing to metering and usage, which matters because so many companies have moved toward usage-based models. We need to make sure usage data is correct, flag anything anomalous, and apply it to invoices. The most important thing is getting that invoice out quickly. If I receive an invoice from a vendor with wrong information, I ask them to fix it, and my net 30 doesn't start until I get the corrected version. The same thing applies to our merchants' customers. Speed matters.

From there, we move into what we call pre-collections—before the invoice hits 90 or 120 days overdue and enters traditional collections territory. This is where the art starts. The agent pulls in signals from the knowledge graph, things like product analytics and support tickets, to gauge how hard you can push to collect. If that doesn't work, it escalates into full collections mode: stronger language, phone calls, whatever it takes. And the last piece is cash-in—connecting to the bank feed, matching payments to invoices, and shutting off reminders once something is paid.

Two more things run in parallel across all of this. AP portal automation, which is increasingly important as you go upmarket. Fortune 100 companies have AP portals that are almost impossible to get through—they'll keep finding something to ding you on so the invoice doesn't get paid. Once we know what it takes for one merchant to clear a specific customer's AP portal, we can apply that knowledge to every other merchant billing the same customer. The other is inbound inquiry handling: simple stuff like W-9 requests or PO number changes we can resolve automatically, while more complex disputes still get escalated to a human. Over time, the system learns how you respond and starts handling more on its own.

Walk us through what happens on day one when a customer plugs into Tabs

Today, if someone joins Tabs, the first thing we do is ask you to provide your source of truth for all your contracts. You always get an implementation manager to help you set up. And we've found that this is where we first need to help our merchants. Where's the source of truth? Is it in HubSpot? A Google Drive? PandaDoc? The number of merchants doing $10M, $20M-plus in revenue who don't necessarily know where all their contracts are is surprising. So the first thing we do is consolidate all of that into Tabs.

We then run what we call a representative set workflow on the list of contracts. We use principal component analysis to break down the entire set and figure out what the representative contracts look like. You may have flat-price contracts, usage-based contracts with commitments, and pay-as-you-go plans. Each becomes a bucket that we sample from. Depending on complexity, it's usually somewhere between five and ten contracts. We'll pull a sample from each bucket and ask you to provide the ground truth—go into each contract and give us what you want the output to look like if we were to process it.

We then run an agent with a couple of tools that generate a context object for each bucket, so we can apply one of those context objects to any new contract that comes in. The most important thing is that we understand how to process your contracts correctly, because if we do, sending invoices becomes much easier and faster.

Once we have that, we can apply it to our knowledge graph for every single customer. We start to understand your customer base so that, when you're ready to send out invoices, we have first-level information about each customer. Then you connect your CRM, pull in your contracts, connect your ERP or general ledger—whether that's QuickBooks, NetSuite, Sage, Rillet, or others—and we start to understand what your business looks like.

There's talk of an accounting crisis coming—CPAs retiring, graduates declining. Is Tabs part of the solution?

I definitely see it as an opportunity. Part of what makes this product so useful is what we've seen from our customers. Companies are expected to grow incredibly fast right now, especially in the venture-backed space. That means keeping really lean back-office teams. If you have one controller, they're sometimes expected to manage billing and all of the accounting. They might have one or two outsourced resources in a different part of the world. The scope of their domain is enormous—not only because you're trying to reduce costs, but also because there aren't enough seasoned, reliable accountants, given the shortage. You're finding under-resourced, overworked accountants, controllers, and AR analysts.

One of the things we talk about a lot is that our job is to help you scale your business without struggling to find the next controller or AR analyst. I think this is an opportunity for those people to focus on what actually matters—passing the full audit and being more strategic. Up until now, controllers and accountants have been so focused on manual rote flows in Excel or in early SaaS products that required a lot of manual work that they haven't been able to be more strategic. You've seen a lot of people coming out of FP&A and business ops roles and becoming heads of finance. I want that to come out of controllership, as those people love numbers and have a solid foundation to build on.

Why should finance teams leverage vertical AI over general-purpose tools like ChatGPT or Claude?

There's been some interesting conversations on Twitter about this. I'm going to go in a couple of different directions. First, I think it's really powerful that you can do a ton of complex reporting using Claude and MCPs without needing to understand how to code or write Python. Tabs is launching an MCP server shortly, which will allow our merchants' finance teams to connect to Tabs to get the information they need for reporting. For very company-specific use cases—such as understanding your cash flow and building dashboards—general-purpose tools are incredibly powerful.

The second piece is the knowledge graph. There is still something to be said for proprietary information that we have access to, even if it's aggregated and anonymized. A lot of the reason general-purpose tools are so good at coding is that there's more code in public repos than probably anywhere else, and they have access to it. That's not true of contracts. Enterprise contracts, enterprise billing relationships—they don't have access to that. Everyone's on an enterprise or team plan, so they can't train their models on it. We can. We can leverage insights from all the data in our platform to make the agents smarter. When you're drafting an email to a customer you know isn't happy with the product, what you actually want is someone to tell you what you should say based on a thousand lifetimes of that information, and then you tweak it.

The third one is liability. Tabs is going to get to the point where we can stand on our ability to collect cash and pass a revenue audit. A general-purpose model will not say that using its tool guarantees you'll pass your revenue audit, no-questions-asked. That's not part of their value proposition. For us, it is.

You were an early hire at Latch and helped scale from Series A all the way through going public. What learnings are you applying as a founder?

It was a very interesting time. The first thing I actually realized is that in-person matters. We started in person when we founded the company back in 2022. Having run engineering teams and Ali having run operational teams in a remote world, we recognized very early on that the iterations we pulled off in person were substantially different from what we could do remotely. So that's the first thing.

The second thing: Latch was a hardware-software company. A lot of the early work was manufacturing-related—once you standardize on something, you have to stick with it for at least one manufacturing round. That's not true in software. On the software side, we really tried to move as fast as possible. That's where half my mental energy goes. How do I get the team, even with more people and more customers relying on us, to still keep iterations really tight and quick

And one more thing. It's all fixable. At Latch, we focused on smart access for apartment buildings. There were times where if something went down—whether it was the internet in the building or a lock—you were basically stopping someone from getting into their home. That's a pretty big problem. But it was always fixable. If you just went head down and pushed through the pain, you could solve it. That gave me confidence that every problem is fixable and no problem is truly as bad as you think it's going to be.

Finance isn't the 'coolest' industry. How do you attract & retain a top engineering team?

I think a lot of companies struggle with this, and to be fair, this is something I work through every single day.

From a recruiting perspective, if you talk about billing as billing, it is boring. But if you can figure out what gets that person excited—whether it's a thousand edge cases and how we leverage AI to handle the ambiguity, or the fact that 90% accuracy isn't good enough and we must get to 100, or the scaling challenges of processing usage data in real time—finding what makes that person excited is incredibly important.

On the retention side, I talk with each of our 30 engineers to understand what they're trying to achieve and help them get there. I don't have any other answer than getting really personal with every single person. Presenting a problem they can project their interests onto in a boring industry is the only way I know how to do it. And I struggle with it every day because there's no magic bullet.

Keeping that empathetic, authentic connection with every individual is the only way I've found to retain the people I want.

If Tabs succeeds, what does a finance team look like in five years?

There are two variations, and I'm not sure which one it will be.

The first is a one-person finance team. On the revenue side, Tabs will be a module within an entire set of back-office modules, so you can have one person run the entire back office for your company. You've got your HR AI, your operations AI, your supply chain AI, your expense AI. And now you've got Tabs as your revenue AI.

The other option is you have one person and then a couple of finance engineers. Their job is to make the system work for you, but they're in-house, and they're actually engineers. They're thinking algorithmically about where to apply AI to build the workflows you need.

What should people take away from Tabs?

The three layers—the knowledge graph, the platform, and the agentic experience—are all equally important. If you have a knowledge graph and agents, but only that, especially in finance, it only goes so far. You still need the primitives. If you have a platform with primitives and agents, you'll lack depth—you'll be a very shallow platform that can't drive real insights. And if you have a knowledge base and a platform but no agents, you're just a SaaS company, and someone is going to put an agentic experience on top of you. You have to have all three. You can't get away with two.

What we're doing at Tabs is building all three layers across the entire revenue stack. Right now, the only two outcomes that matter are when a deal closes and someone signs, and when you are maximizing cash in the bank and passing your revenue audit, minimizing additional human capital labor against that contract. That's it. And you need all three layers to do that.

Additional Reads

• The Tabs’ origin story on YouTube: A deeper look at how the company is rebuilding revenue automation from the ground up.
• Lightspeed on building with Tabs: The investor case for why revenue automation is the next big back-office category.
• Tech Optimist Podcast, Episode 93: Tabs joins a16z to discuss how AI is rewriting B2B revenue.

Quick Poll

Rate This Edition