Why we bet Adopt AI on agents running the office of the CFO

Why we narrowed down

A prospect ended a call last spring with a question I could not answer well: "If I had to tell my CFO in one line what you do, what would I say?" I gave a paragraph. She was polite. We did not close.

That question is the reason we rebuilt the company. About six months ago we made a decision that felt like shrinking it: we stopped building agents for everything and started building agents for one thing, finance and accounting work. Today we launched a new website that says exactly that. I want to explain why, what happened after, and where it goes next.

What we had

We had real technology. ZAPI, which we patented, looks at any software system, discovers its API, and generates a working connector in under 24 hours, even for systems nobody has integrated before. On top of it we had agents that could act across a customer's whole stack. It was impressive in a demo. It was also almost impossible to sell. Every conversation started with "so what do you actually do?" and ended with a custom scoping exercise. We were good at a lot of things and essential for none. We had powerful technology and no sentence.

What we saw

When we looked at where the agents were actually earning their keep, the answer was embarrassingly consistent. Month-end close. Bank and intercompany reconciliations. Accounts payable. Collections. The reporting pack. Bookkeeping and tax for companies that had no finance team at all. The work was high-volume, rules-based, spread across five or six systems, and done by people who would rather be doing anything else. It was the work our technology was built for. We had just never said so.

What we did

We said so. We wrote the sentence: AI agents that run your finance operations end-to-end, in the systems you already have, with your team reviewing exceptions. Then we deleted everything that did not fit it. We rebuilt the website around the work itself, a close, a reconciliation, an AP run, instead of around a platform diagram. And for companies without a finance team, we paired the agents with our own accountants and tax experts and started showing up as their finance function.

What happened

The pipeline changed first. Inbound grew 55x in six months. The people filling out the form were controllers, CFOs, and firm partners with the precise problem we solve, and they came in already knowing what they wanted. Demos went from "let me show you what's possible" to "here is your close running." We now run more than a hundred qualified demos a month and close new customers every week. Effectively all of our current revenue was added after the change.

We did not spend more on marketing to get there. Our cost per lead is low enough that we could triple inbound on budget alone, and we have deliberately not done that, because the constraint right now is how many deals a small sales team can run at once, not demand. The product changed too: one roadmap instead of one per prospect, and a connector technology that finally had a job description.

What I got wrong

I believed a broad platform was the safe choice: more surface area, more ways to win. In practice breadth was the risk. Nobody can champion you internally if they cannot describe you in a sentence. Narrowing down did not make us smaller. It made us legible, and legible companies grow. The lesson is uncomfortable for anyone who likes optionality: if a buyer cannot repeat your pitch to their boss, you do not have one yet.

Where we are headed

For the next quarter we are heads-down converting the demand already in the door and adding sales capacity to match. After that, the ambition is simple to state: the agents run the entire back office, and the finance team becomes the reviewers and decision-makers rather than the operators.

The next few years

Our customers still close the month, and that is the next thing we want to change. When agents categorize, match, and reconcile every day, your books stay no more than a day behind, and month-end turns into a review of the few exceptions the agents could not resolve. We are building that now.

With a ledger that is current every morning, you can run a rolling cash forecast and alerts that fire when something changes, so a CFO hears about a problem that week instead of in next month's pack. You can also run controls on every transaction. A duplicate payment, a vendor's bank details changing the day before a payment run, revenue nobody invoiced: today an auditor might find those in a yearly sample. An agent can flag them the day they happen.

As an agent builds a record on your books, you hand it more workflows, one at a time, and you can undo anything it did.

Your team's job changes as well. Accountants who stop assembling numbers get time to write the rules the agents follow. I think the best accountants of the next few years will be the ones who can teach an agent how their business handles an exception.

Saying no

We now turn down agent projects outside finance, including some that would pay well. We will not build a chatbot for accountants. Every engineer at Adopt works on accounting problems now, so when a controller tells us an agent handled an exception badly, the fix reaches every customer.

Send us your worst reconciliation

It is the end of November. How much of September is reconciled? If the answer is "give me a week," send us the reconciliation your team dreads. We will run it on your data and show you the finished reconciliation with every step the agents took.

If you run a small company, you can start free at adopt.ai and connect your bank.

Deepak Anchala, Co-founder and CEO, Adopt AI