"FP&A software is dead."

That's a bold claim.

But it's even bolder coming from the guy who spent a decade building it. That's Didi Gurfinkel. He built Datarails. Now he's burning it down and building FinanceOS.

FinanceOS is your governed financial data layer, making every AI output trusted, repeatable, and audit-ready.

See FinanceOS in action

My all-time favorite FP&A platform was a little tool called Adaytum, which was like a supercharged back-end database for Excel. Incredibly easy to use and brilliantly powerful. Well, until Cognos bought it in 2003 and wrecked it. No… I don’t forgive easily.

In the twenty-plus years since, FP&A software has been all about trying to ‘kill Excel’. The pitch went something like this: ditch the spreadsheets, move to our platform, and get your life back. You are now free from Excel jail.

And oftentimes the products were cool and… tempting. I’ve bought a few. You get charmed by the improved version control, workflow control, stronger access, and control permissions.

Then you get a few months in, and you want to change something in the chart of accounts. Or add a new location. And you find out you need to take an IT ticket or pay a consultant to do that for you. 

Or you want to add more operational data to feed your KPI dashboard, and you have to get it structured, cleaned, and labeled for the data warehouse first.

OK, maybe another day then…

WTF? No one told you all of those shiny new features cost you speed, flexibility, and responsiveness. Oh well, too late now…

But the thing is, Excel isn't a bad habit we need to break away from. Finance thinks in rows and columns. Excel's a way of thinking about and articulating financial problems and solutions. No amount of venture capital or hype has changed that over the last five decades. 

As Didi puts it, Excel is to finance what Word is to lawyers. It's the medium, not the message.

"Everyone who deals with numbers goes to Excel because if you want to present numbers with the relationship between the numbers. You need the canvas," he said. 

So, while most other FP&A platforms were trying to corner businesses into their native software, Datarails did the opposite. The company met businesses where they were… in Excel.

Didi and the team decomposed the different roles Excel plays inside finance workflows, and then built the Datarails platform around those different roles: 

In short, Datarails built the tool that sits between the data and the ability to analyze it.

"Who is crazy enough to build a platform for Excel?" Didi told me. "To build a deep technology with algorithms to transform Excel into a database. You need to be a lunatic."

So, did Didi set out to make life easier for FP&A teams?

No… not quite. In fact, not at all.

After a long and successful corporate career (latterly at Cisco), Didi’s founding thesis for Datarails was that Excel needed a friend to take the load.

“I came with the concept, you cannot beat Excel. Let’s try to make Excel something that enterprise can manage. The main problem with Excel has been the version control, the auditability, the trust,” he said. “No one complains about the calculation in Excel, no one complains about the presentation of Excel. They complain about owning the data."

From that insight, Datarails built a platform that could connect all of an organization's Excel files into a single, centralized database that could read the structure of any spreadsheet, follow its data flows, understand its formulas, even interpret its cell formatting and color coding, and map it all into something structured and queryable. 

Founded in 2015, Datarails’ first product lines centered on compliance for financial institutions. When that fell flat, there was a pivot to ‘spreadsheet management’, followed by a dabble with Datarails as a collaboration and productivity tool. Didi spent the next five years trying to find someone who needed the solutions he’d built. 

He flew around the world trying to drum up business from anyone who touched Excel at scale. He got meetings and plenty of blank stares. And certainly no customers.

For five years, Datarails had basically zero sales. Frankly… Didi had a solution in desperate search for a problem.

“We thought, 'This is the best thing ever,’” he said. “I tried to start selling the solution. Then the problem was that we had the platform, not the solution.”

By 2020, Datarails had just a few months of runway left… the end was nigh. But there was one glimmer of hope. On Didi’s wander around the heaviest corporate Excel use cases, he’d stumbled into meetings with a few FP&A teams. And this time, he saw a different level of engagement. Something more concrete. 

“We came up with this approach: we know how to transform Excel from personal use to something reliable and enterprise-grade. Let’s go to the number one users in Excel in organizations - FP&A.”

With the Datarails 13-week cashflow forecast looking undeniably grim, Didi ran back to his investors with a proposal… one last pivot, into FP&A. 

"I came back to my board and said, ‘Look, we have one last shot. We should shift focus to business FP&A teams. I give you the choice. You can shut the door and say, ‘Give me my money back.’ And fair enough,” he said.

“They asked, ‘Are you convinced that this is the right move?’. I was honest with them: ’I’ve been convinced like three, four times before. So how can you ask me this question?’"

The investors figured it was worth one last roll of the dice.

The rest, as they say, is history. The accidental advantage of spending years building around Excel was that Datarails arrived at FP&A’s door with a product no competitor thought to replicate. 

This product was firmly counterpositioned in a sea of products hell-bent on removing that little green X from your desktop.

Less than a year after that fateful board meeting, Datarails already had $1m of ARR. And since then, it’s been a rocketship. Didi claims they are on track to hit $100m of ARR during 2027.

An overnight success story, one decade in the making…

So here's what I kept coming back to: Why would Didi want to upend his own product category after waiting so long to find success?

Well, it comes back to that epiphany Didi had back when he first saw LLMs. He immediately saw how AI could turbocharge the technology he had built and hand even more control to the user.  

And he gambled that foundational models would improve quickly enough to power Datarails into a new era. With LLMs, the Datarails platform, and Excel, he could build something truly transformational for FP&A teams.

That creative leap set off three years of work to build a new AI-native iteration of Datarails called FinanceOS, which was launched just a couple of weeks ago.

Didi personally led me through a behind-the-scenes tour of FinanceOS just days after it launched. He brought Datarails' own Head of FP&A, Ari Ben Zoor, for a demo on how they use the product on their own live financials. Not a client. Not a case study. Their own numbers.

Ari walked us through building board decks, flux analyses, and variance commentaries using Claude prompts in real time.

The first thing that struck me was that FinanceOS didn’t fit into a neat box:

FinanceOS is kinda all of those things, and none of them all at once.

So, what is it then?

What I saw broke the FP&A software trade-off by eliminating that layer entirely. FinanceOS gives you all the flexibility to do what you want with Excel, along with all of the version control and trust of native FP&A software. And then turbocharged it all with Claude.

In other words… FinanceOS is an AI-powered data warehouse for finance teams that handles any data (however clean or messy) and serves it straight into a controlled Excel environment, where you do whatever the hell you want with it via Claude (or any other AI).

Steven Goyne, a lifelong FP&A leader-turned solutions consultant at Datarails, showed me how he’d produce a board deck in just a few moments directly from a short Claude prompt. OK, nice enough, but I’m reading about that on LinkedIn every day. It’s good to see it on live data - rather than a sanitized case study - but the real question is where does that information come from, and how do I know it’s right?

And that’s where it got interesting.

FinanceOS was connected to Datarails' payroll system, their ERP, CRM, an Excel-based budget model, and their other core data sources. Claude wasn't generating outputs from a static file someone had uploaded.

It was querying a live, governed database using Datarails' own formula taxonomy, a specific structured language for retrieving data from FinanceOS. And Claude had been trained to look only to Datarails for its information source. Claude had guardrails and zero permission to hallucinate.

So you could say, ‘Tell me what is driving February’s payroll variance to budget of $150k’ and Claude would:

Which begs the question… how the f*ck did it do that?

The answer was in the architecture. 

Claude was following a consistent, repeatable set of instructions for how to pull data, which means the outputs are auditable, and the numbers tie back to the source every time.

Steven showed me what was happening under the hood.

Underneath all of this, six agents were running in the background, each with a defined role in the process.

Each of these agents is programmed with a specific role inside this workflow and nothing else. By narrowing the context and the responsibilities, they are able to focus their computational power and reasoning on much narrower problems. Kind of like putting a magnifying glass up to the sun to harness immense power on something very specific.

Compare that to working with a raw LLM, which is trained on the entire internet and, in theory, prepared to answer any conceivable question about anything.

And these agents all show their work, giving you that all-important traceability, meaning that the audit trail and version control of your spreadsheets (another thing Excel is bad at) are all baked into the Datarails platform.

It’s the kind of thing that, once you see it, you can’t unsee it.

When I hear something is AI-native or AI-powered, I’m immediately cynical.

For software founders, putting ‘AI’ in their fundraising deck has been a great way to guarantee funding. It’s quickly become the most abused phrase in enterprise software marketing, so when I see a new product, I have a new favorite question for these moments… “Yeah, but what’s AI about that?’

So, I want to be clear about where AI is actually doing real work inside FinanceOS.

As I see it, the AI is doing three distinct jobs here.

The first is at implementation. Anyone who has been through a finance system implementation knows the particular misery of data mapping. Chart of accounts reconciliation. Dimension definitions. Source system quirks that nobody documented. You will also know how important the implementation is to get right.

It is, in the conventional telling, a long list of small, painful problems that consume enormous amounts of time and goodwill before you ever get to the point of actually using the thing you bought.

Let me be clear: implementing Datarails is still real work. Your people, along with Datarails’ team, have a lot of heavy lifting to do during the process. Mapping your chart of accounts, defining your dimensions, and establishing your data architecture are crucial decisions requiring human judgment. You are defining the foundations of how your business understands itself financially. 

But with FinanceOS, AI is embedded in that implementation process in a way that changes its character. AI is able to quickly parse what used to be a million small, fiddly data architecture problems and present them back to you as a much smaller number of leadership-grade questions.

So, the implementation burden is more concentrated on the big things that really matter.

The second job is in the agent layer. The agents running inside FinanceOS are continuously working to make sure everything ticks and ties within the database. When something doesn't, they don't just flag it and dump a list of exceptions in your lap. 

They take it through a resolution workflow built of multiple layers of checks and only bring it back to you, the human in the loop, when they need a decision. And when they do bring it back, they tell you in plain language what the two or three real issues are and what they recommend. 

You respond in natural language. The days of receiving a data exception report that requires its own analyst to interpret are, at least in principle, over.

The third job is the one that's most visible, and in some ways the most straightforward: AI as the interface. Claude, sitting on top of a rich, multi-dimensional, governed database, is able to build whatever analysis you need in minutes or seconds. 

The critical detail - and this is the one that separates FinanceOS from simply pointing Claude at a spreadsheet - is that Claude is querying Datarails through a specific formula taxonomy. There are no sprawling nested formulas. No iterated prompt-on-prompt constructions that nobody can audit six months later. The outputs are clean, repeatable, and tied to the source.

That combination of AI in the implementation, AI in the governance layer, and AI as the interface is what Didi means when he talks about FinanceOS as infrastructure rather than software. 

It is not a tool that does one AI thing well. It is a system in which AI is load-bearing at every level.

When I saw FinanceOS for myself, the real elephant in the room was this: with the right permission controls in place, it's not a big leap to see how the business could become self-serve for a lot of financial information. And doesn't that crush the FP&A function? So is he replacing FP&A software… or FP&A itself?

I put this directly to Didi. The answer is… he doesn’t really care. 

"Financial planning and analysis will be done forever. I don't know who will do it. It could be AI or a person. But it's not my problem,” he said. “I take the responsibility to build the right platform."

Refreshingly honest… let’s break that down.

The traditional FP&A function plays several roles. It does the analysis. It manages and controls access to financial data. It translates that data for the rest of the business. It builds the board decks, the management reports, and the variance commentaries. And most importantly, it turns that into C-suite grade level insights and trade-off decisions.

Of those roles, the analytical work of financial planning and scenario modeling judgment, the kind that requires judgement and commercial context, has enduring value. That work will always need a human who understands the business behind the numbers.

But gatekeeping information? That has no future. If a sales director can query the database directly, run their own scenario, and get a board-ready output in seconds, the version of FP&A whose primary function is to be the gatekeeper between the data and the rest of the business is not an enduring job. 

As our conversation came to a close, Didi shared the purest version of his vision for how the business consumes financial information: "You don't need anything else in the middle. You have data and you have AI.” 

I think we are a long way from that. But for the first time, I've seen an AI-native finance tech stack that has given me data I could trust - with audit-grade traceability - without denying me the flexibility to see information and build models the way I want. 

Hungry for more? Click here for more extracts of my conversation with Didi.

The most powerful AI in the world is working off last month's spreadsheet

It doesn't know a hire was made, a sale closed, or a cost hit the books. 

No governed data. No audit trail. No connection to your numbers.

FinanceOS fixes that. Every AI output is trusted, repeatable, and audit-ready.

See it in action