Why your finance team hasn't felt the AI revolution yet

It isn't a tools problem

The average finance team uses 7+ software applications to do its job, according to Gartner's 2023 CFO technology survey — and that number keeps climbing. The stack is bloated, not bare.

The reason finance still runs on spreadsheets isn't that nobody built the tools. It's that none of the tools talk to each other, none of them hold the full picture, and none of them were built for the way modern AI works.

When engineering wants to use AI, the code lives in GitHub — clean, structured, accessible by API.

When finance wants to use AI, the data lives in:

• 14 different bank portals, half of which don't have APIs
• An ERP that was last upgraded in 2019
• A general ledger no one fully trusts
• A folder of Excel files named Cash_Forecast_v17_FINAL_USE_THIS.xlsx

You can't point an AI assistant at that and get something useful. The data isn't ready. And nobody's done the years of plumbing required to make it ready.

‍

Banks were not built for AI

Here's the part most people outside finance don't realize: the financial system was built before the internet, let alone before AI.

Most banks still don't offer real APIs. The ones that do have inconsistent formats and unreliable feeds. The standard way to get bank data into a system is still SWIFT messages, host-to-host file transfers, or — in 2026 — somebody logging into a portal and downloading a CSV.

This is why every "AI for finance" demo you've seen looks great in a sandbox and falls apart in production. The demo assumed the data was already there. In real finance, the data is the hardest part.

Live, validated bank data across multiple entities and currencies isn't something you can prompt your way to. It takes real infrastructure - and it has to keep working as bank formats change, feeds break, and credentials expire.

Until that foundation exists, AI on top of it is decoration.

‍

Finance has requirements that AI doesn't meet by default

Even if the data problem were solved tomorrow, finance has a second hurdle most other functions don't.

Engineering reviews every line of AI-generated code before it merges. Marketing can tolerate an AI draft that gets edited. Finance can't tolerate "mostly right." A forecast that hallucinates a vendor is worse than no forecast. A reconciliation that's 95% correct is a reconciliation that failed. A payment posted to the wrong entity is a board-level problem.

General-purpose AI tools — ChatGPT, Claude, Gemini, Copilot — weren't designed for this. They were designed to be helpful and creative. By default they:

• Give different answers to the same question on different days
• Have no memory of your specific company, accounts, or policies
• Leave no audit trail of how they got to an answer
• Don't enforce who is allowed to see what
• Don't know which numbers are real and which they invented

That's a fine tradeoff if you're drafting a blog post. It's a non-starter for closing the books.

So finance teams experiment with ChatGPT, get burned once, and quietly go back to Excel. The technology didn't fail them — the technology was just never built for what they needed.

‍

The work itself is harder than it looks

There's a third thing worth naming. Finance work isn't one type of work.

Sometimes it's deep and collaborative — a multi-entity forecast that takes weeks of judgment, assumptions, and stakeholder input. Sometimes it's repeatable — the same reconciliation, the same report, the same close, every period. Sometimes it's a fire — the CFO asks why cash dropped $4M last week and needs an answer in an hour.

Each of these needs a different kind of help. The first needs a system that holds context over time. The second needs reliable automation. The third needs instant, flexible analysis on top of full financial context.

Most software picks one of these and ignores the other two. That's why finance teams end up with seven tools and still default to spreadsheets — Excel is the only thing flexible enough to handle all three modes, even badly.

The AI revolution your finance team is waiting for isn't a smarter chatbot. It's a system that adapts to all three modes, with real data underneath and real controls around it.

‍

Execution without context is just faster mistakes

A lot of the AI conversation in finance has fixated on speed. Faster reconciliations. Faster forecasts. Faster reports.

Speed is the easy part. The hard part is making sure the system knows what it's doing before it does it.

Consider a simple treasury decision: moving $2M from a US entity to a UK subsidiary to cover payroll. A fast system executes the transfer. A smart system knows that the UK entity has a £1.5M receivable landing Thursday, that the FX rate is unfavorable today and improving next week, that there's an intercompany loan already open between those two entities, and that company policy requires CFO approval over $1M cross-border.

Same action. Completely different outcome.

This is why context has to come before execution. An AI agent that can post journal entries, send payments, or update forecasts is only as good as the financial picture it's operating from. Without live cash positions, entity-level visibility, policy awareness, and a trail of past decisions, automation just means making the wrong call faster.

The finance teams getting real value from AI right now aren't the ones with the flashiest agents. They're the ones who built the context layer first — and then let execution happen on top of it, with judgment baked in.

‍

What changes when this actually works

Picture the morning routine without the spreadsheets.

The treasurer opens her laptop and the cash position is already there — accurate, current, reconciled across every entity and currency. Anomalies are already flagged. The forecast already updated overnight based on yesterday's actuals. The categorization is done. The variance commentary is drafted.

She spends her morning on the things that actually require her judgment: where to move cash, how to hedge an exposure, whether to draw on a credit line. The logistical work — the hours that used to go to moving data around — happened while she was asleep.

That's not a futuristic vision. It's what happens when the foundation is in place: live connections to every bank and ERP, a unified data layer, agents that maintain accuracy continuously, and controls built into every step.

The AI revolution hasn't skipped finance. It's been waiting for someone to do the unglamorous work first.