A study of business spreadsheets found that 94% contain at least one critical error. Citigroup once wired $900 million to creditors because of a spreadsheet mistake. The "London Whale" trades that cost JPMorgan $6 billion were partly the result of a copy-paste error in Excel. And in 2020, England's public health system lost track of 16,000 COVID-19 test results because a spreadsheet quietly hit its row limit.
Now every one of those error-prone spreadsheets is being handed to an AI assistant and asked to "just analyze this." That is the uncomfortable subtext behind a claim making the rounds in finance circles this year: traditional FP&A (financial planning and analysis) software, built for humans clicking through menus, is obsolete. The provocation came from Didi Gurfinkel, CEO of Datarails, the Tel Aviv-based FP&A company that built its name solving "Excel hell" for finance teams. His argument, in short: AI can now build models, run analysis, and write reports faster and better than a human ever could — so any tool that limits what the AI can touch is already behind.
Whether or not you buy the "dead" framing, the underlying shift is worth paying attention to if you run the books for a small business, a lean startup finance team, or even just your own company's numbers in a spreadsheet.
What Datarails Actually Announced
Datarails — which has raised $175 million in venture funding, including a $70 million Series C round in early 2026 — launched a platform called FinanceOS. It's less a single product than a positioning shift: instead of building yet another walled-garden dashboard, Datarails is betting that finance teams will keep using whatever AI tool they already prefer (Claude, ChatGPT, Microsoft Copilot) and just need the data underneath to be trustworthy.
FinanceOS connects to hundreds of source systems (accounting software, ERPs, CRMs, payroll), performs the unglamorous work of consolidation — eliminations, currency conversion, reconciling numbers that come from three different systems and don't quite agree — and exposes the result to AI models through a governed, auditable layer using the Model Context Protocol standard. Once an AI-built model exists, it can be locked so it stays consistent even as the underlying data keeps refreshing.
Gurfinkel's pointed comment about competitors was that older vendors "are already gone... they don't have enough cash or energy to rewrite the technology," while newer AI-native rivals invested heavily in slick interfaces but underinvested in the unglamorous data-consolidation layer — which he argues is actually the durable competitive advantage. Datarails is also shifting its pricing model away from per-seat licensing toward usage-based pricing, on the theory that AI agents, not humans, will increasingly be the ones "using" the software.
The AI Adoption Paradox in Finance
Datarails isn't the only company noticing this. Industry surveys on AI use in corporate finance keep turning up the same puzzling pattern: adoption is rising, but reported impact is barely moving. One 2025 industry benchmark found that the share of finance teams actively using AI ticked up only one percentage point year over year — from 58% to 59% — while 91% of finance teams said the AI tools they'd already adopted had delivered low impact on their actual work.
That's a strange result for a technology this powerful. The usual explanation offered by CFOs and finance leaders isn't that the models are bad — it's that the data being fed into them is. Teams are connecting a genuinely capable AI to genuinely broken source data and then wondering why the output doesn't help.
The Real Problem Isn't a Lack of AI — It's a Lack of Governed Data
Here's the part that matters more than any single vendor's product launch: AI adoption inside finance teams has been rising, but the impact of that AI hasn't kept pace. Multiple 2026 industry surveys point to the same gap — teams are turning AI on faster than they're getting value from it, and the most commonly cited reason is data quality. One widely cited 2026 CFO survey found that 63% of finance leaders believed they had full visibility into their own spend data, while in reality only about 5% actually did.
That gap is the whole story. If your books are messy, plugging in an AI tool doesn't fix the mess — it just processes the mess faster and hands you a confident-sounding answer built on bad inputs. AI models don't reliably know when they're wrong. They will cite a tax rule that sunset two years ago, or summarize a spreadsheet tab that was never updated, with exactly the same tone of authority as when they're right.
For a small business, this shows up in mundane but costly ways:
- An AI-generated cash flow summary that quietly excludes a bank account no one told it about.
- A "clean" expense report built from a spreadsheet where 3–5% of formula cells contain silent errors — a rate that's been consistent in spreadsheet-error research for two decades and only compounds as sheets grow.
- A confident answer about margin that's actually averaging two different fiscal periods because a tab got copied instead of linked.
None of this is really an AI problem. It's a data governance problem that AI has made more visible — because now the mess gets summarized into a paragraph you're more likely to trust at face value than you would a raw, obviously-messy spreadsheet.
What "AI-Ready" Financial Data Actually Requires
Whether you ever touch a product like FinanceOS or not, the underlying discipline it's selling is worth adopting on its own:
1. One source of truth, not a folder of spreadsheets. If "the real numbers" live across four workbooks with different last-modified dates, no AI layered on top will reconcile them correctly — it will just guess which one is current.
2. An audit trail for every number. When a figure changes, you should be able to answer why it changed and who (or what) changed it. This is exactly what spreadsheets are worst at — a formula overwritten with a hardcoded number leaves no trace.
3. Locked, versioned models. An AI-built projection is only useful if it stays stable while you review it. If the underlying data keeps shifting silently underneath the model, you're reviewing a moving target.
4. Governance before automation, not after. The instinct is to bolt AI onto whatever process already exists. The teams getting real value are doing the reverse: cleaning up the chart of accounts, standardizing categorization, and only then pointing AI at it.
5. Permissions that don't rely on "don't touch that tab." In a shared spreadsheet, access control usually means a verbal agreement not to edit certain cells. That's not governance — it's a habit that breaks the first time someone new joins the team. Whatever system holds your numbers should enforce who can change what, not just hope everyone remembers the rules.
6. A record that survives the tool, not just the vendor. FinanceOS, like most FP&A platforms, is a proprietary system — useful, but your historical data lives inside someone else's product. Worth asking of any tool you adopt: if you left tomorrow, could you take a complete, readable financial history with you, or would you be re-entering years of transactions by hand?
Would This Actually Change Anything for a Five-Person Company?
It's easy to read "FP&A platform" and assume this is a large-enterprise story. In practice, the failure mode is the same at any size — it's just less visible in a five-person company because nobody's auditing your spreadsheet but you.
Picture a small agency that tracks project profitability in a shared spreadsheet: one tab per client, hours pulled in from a time tracker, expenses pasted in from a card statement. Ask an AI assistant "which clients were profitable last quarter?" and it will happily read every tab and give you a confident ranked list. What it won't tell you is that two client tabs use a different overhead formula than the rest (copied from an older template), and one client's numbers are three weeks stale because that tab wasn't updated after the last invoice run. The AI has no way to know that — it just sees numbers in cells and reports on them.
The fix isn't necessarily buying enterprise FP&A software. It's making sure that before any AI touches your numbers, there's one place those numbers live, one consistent format they follow, and a record of when each figure last changed. That's a governance problem, not a company-size problem.
The Case for Plain-Text, Version-Controlled Books
This is precisely the gap plain-text accounting was built to close, long before "AI-ready data" became a buzzphrase. A ledger kept in a version-controlled plain-text format has, by construction, the properties finance teams are now scrambling to bolt onto their spreadsheets:
- Every change is tracked. Git history is the audit trail — no separate change log to maintain, no formula silently overwritten without a trace.
- One canonical file, not a folder of "final_v3" workbooks. There's no ambiguity about which version is current.
- Structured, parseable data. Because entries follow a consistent double-entry format instead of freeform spreadsheet cells, an AI tool reading the ledger is working from structured, auditable records rather than guessing what a merged cell means.
If you're a small business owner, freelancer, or indie developer currently managing books in a spreadsheet and wondering whether to hand it to an AI copilot, the more durable fix is usually to fix the underlying record-keeping first.
Keep Your Books AI-Ready From Day One
Whatever finance tooling wins the next few years, the winners will be the ones whose underlying records were trustworthy before an AI ever touched them. Beancount.io offers plain-text accounting that's transparent, version-controlled, and structured for exactly this kind of scrutiny — no black-box spreadsheets, no mystery formulas, just auditable records any tool (AI or human) can trust. Get started for free and build your books on a foundation that's ready for whatever comes next.