Over the last two years, finance has quietly crossed a line. Claude, OpenAI, and others can now build 3-statement models from exports, check that they balance, and extend them with scenarios in minutes. They draft board commentary, variance explanations, and investor memos that your FP&A team used to grind on for hours. They summarize earnings, filings, and internal reports with context and reasonable judgment.

Anthropic is rolling out specialized finance agents for month-end close, statement auditing and GL review, KYC screening and compliance, model building, earnings review, and pitch prep. At the same time, the big firms are all converging on the same message:

• Gartner finds a majority of finance functions already use AI in some form, and CFO optimism about its long-term impact keeps rising even as adoption levels off.
• KPMG talks explicitly about a future where AI sits “at the core of how work gets done” in finance — breaking down silos across ERP, analytics, and planning into continuously learning, automated processes.
• Deloitte describes “digital controllership” where automation and AI fundamentally change how close, controllership, and reporting operate.
• EY is helping clients use GenAI and hyperautomation to build agile, cloud-based finance functions — emphasizing that AI must be embedded in the operating model, not bolted on the side.
• PwC reports that nearly six in ten CFOs are already investing in AI and analytics, yet most CEOs say they’ve seen limited or no financial benefit. The few who do are those who built operational and governance foundations first.

Why assistants failed finance

So there’s a paradox.

The intelligence is here. The tools are here. But most finance teams are still running a 2010 operating model with 2026 technology.

Assistants made individuals faster, but they never owned a workflow — so the calendar, the close and the BvA cycle look exactly the same. That gap is exactly where digital workers live.

The new baseline: super-analyst in a tab

Between Claude and OpenAI, the new baseline for FP&A and CFO teams looks like this:

• Upload a P&L and balance sheet, get a working 3-statement model with errors flagged and ties checked.
• Ask for first drafts of your board pack narrative and get a decent, structured story.
• Have AI review earnings, benchmark you against peers, and surface key metrics.

All of that used to require well-trained analysts, manually curated exports, and days of back-and-forth revision. Now it’s minutes. Gartner’s latest AI in Finance survey reflects this: nearly 60% of CFOs and senior finance leaders say their teams use AI today, and the major consulting firms are no longer asking if AI will transform finance, but how fast and with what governance.

And yet:

• Month-end is still a brutal calendar event.
• BvA is still manually reassembled every period.
• Forecasts are still refreshed only when someone fights for the time.
• Data hygiene is still owned by a handful of heroic individuals.

You’ve added AI assistants. You haven’t changed the way work flows through finance.

Where most finance functions are stuck

If you’re candid about your own setup, it might look like this. A few power users — you, your top FP&A lead — have:

• A Claude or OpenAI workspace.
• Connectors into NetSuite, SAP, or Oracle, the data warehouse, maybe CRM.
• A growing library of prompts and skills for BvA, board prep, and modeling.

When they remember to use it, work is faster, first drafts are much better, and some analyses that used to be “too expensive” to run now happen. But structurally:

• Roles haven’t changed.
• Calendars haven’t changed.
• Ownership of workflows hasn’t changed.

Everything still depends on people remembering to open the AI, people re-running the steps every cycle, and people stitching outputs back into the same old spreadsheets, decks, and tools.

This is exactly the pattern PwC and others are calling out: a big majority of CEOs say they’ve seen little to no benefit from AI so far, despite heavy investment, because most organizations haven’t moved beyond pilots and are treating AI as a bolt-on tool, not a redesign of how work gets done.

You have smarter helpers. You don’t yet have a new operating model.

What a finance digital worker actually does

Almost all current AI in finance is assistant-based. You log into Claude or an internal app. You ask, “Help me with this model, pack, or analysis.” You get a better answer, faster. But AI doesn’t own anything. It helps the person who does.

A digital worker flips that. It has a clearly defined role:

• “Own BvA and forecast refresh for this region or P&L slice.”
• “Own commissions prep according to this incentive plan.”
• “Own GL hygiene and reconciliations for this ledger.”

It connects to the systems it needs — ERP, planning, CRM, billing — with its own scoped identity. It runs on triggers and schedules, not human memory: a new month tick, a new data batch, nightly or weekly jobs. It produces standard outputs your team already expects: BvA packs, updated forecast views, exception lists, reconciliation reports.

Your team’s relationship shifts from “I’ll go run BvA in Claude” to “Emily, our BvA and forecast worker, has run this cycle. We’re reviewing and adjusting.”

That might sound like semantics. It isn’t. It’s the difference between occasional acceleration and reliable execution capacity.

How this fits your existing stack (Copilot, ERP, planning)

Digital workers don’t replace your ERP, your planning tool, or your Copilot/Claude license. They sit on top. The worker reads from NetSuite, SAP, Oracle or your warehouse, writes back into the same planning models your team already uses, and hands finished outputs to humans inside Excel, Slack, or your BI tool. Your assistants stay for ad-hoc questions; the worker takes over the recurring cycle.

Nothing in your stack gets ripped out — the worker just becomes the named owner of the cycle that used to live in someone’s head.

This is the world KPMG describes when they talk about a finance function that’s predictive instead of reactive, with AI embedded into processing, close, and planning across a unified ecosystem that continuously learns and improves.

Governance and control: why this is safe for CFOs

CFOs are rightly worried about more than speed:

• “Who is this AI operating as?”
• “What systems can it actually touch?”
• “What’s the blast radius if something goes wrong?”
• “Can I explain this to my audit committee?”

A finance digital worker is built to be defensible in front of an audit committee — scoped identity, least-privilege access, and a full audit trail of every action it takes.

Big firms are emphasizing the same thing. KPMG’s latest AI research stresses that AI is now an “operational reality” and that pressure is shifting to security and control as companies race to deploy multi-agent systems. Deloitte’s Digital Controllership work focuses heavily on governance, risk, and control as automation moves deeper into close and reporting. PwC’s surveys show that the small minority of companies seeing real AI ROI are the ones who built strong AI foundations — responsible frameworks, clear environments, enterprise integration — before piling on use cases.

The default agent setup is dangerous

Most generic agent setups today look like this: the agent runs with the same effective access as the human user. Connectors are broadly scoped (“read/write this whole dataset”). There is no clear separation between “what Nessy can see” and “what this agent should see.”

That might be acceptable in experiments. It is not acceptable in core ERP, cash and debt workflows, management reporting, or anything that ends up in front of a board, lender, or regulator.

Workers carry their own identity

Digital workers are designed to have their own, task-scoped identities. Each worker has just enough access to do its job — no more. Its actions are logged as its own — not as a human pretending to be a bot. If it misbehaves, the damage is limited to its scope, and the investigation is straightforward.

This is the controllable, auditable structure Deloitte’s next-gen controllership and KPMG’s future-of-finance narratives keep pointing toward: a human-led, AI-operated finance function with clear controls and traceability.

What digital workers actually change

Digital workers don’t just speed up tasks — they remove entire blocks of recurring work from your finance calendar.

1. You can delete tasks from calendars

“Prepare BvA pack” becomes “Review BvA pack.” “Build forecast views” becomes “Decide what to do about them.” “Clean and reconcile this data” quietly disappears as a recurring line item. You’re not just shaving minutes off tasks. You’re removing entire blocks of work.

2. You can prove what happened

Every action is tied to a worker with a specific scope. Audit trails exist at the workflow level, not spread across inboxes and spreadsheets. When something looks off, you can trace which worker touched it, what inputs it saw, what logic it followed.

This is the difference between “we use AI” and “our close and reporting processes are AI-orchestrated under a control environment we can defend” — exactly what surveys like PwC’s CFO Compass and Deloitte’s controllership work say is becoming the top challenge and success factor for modern CFOs.

3. You can scale without linearly adding headcount

More entities? Clone or re-scope existing workers. More volume? Increase cycles, not staff. More complexity? Capture new rules and edge cases in the worker’s playbook once, then benefit every cycle. This is what PwC’s latest AI performance research highlights: a small minority of companies capture the majority of AI’s financial gains because they use AI to reshape business models and operations, not just to make individuals work faster.

Where to start: BvA, forecast, and close

Don’t boil the ocean — pick one recurring cycle (BvA, forecast refresh, or month-end close) and stand up a single named worker to own it end-to-end.

Pick one workflow you care about:

• FP&A: BvA and forecast refresh for a region or business line.
• Close: month-end hygiene and reconciliations for a ledger.
• CFO office: board pack prep and narrative scaffolding.
• Treasury: short-term cash and covenant monitoring.

Now ask yourself:

• Who owns this now? If they’re on vacation, do you panic?
• How much of it is rebuilt manually every period, even if AI helps?
• Could you describe it clearly enough that you’d trust a named non-human worker to run 80–90% of it, with your team in review?
• If that worker didn’t run one month, would anything important break?

If your honest answers are “it lives in one or two people’s heads,” “we reassemble it every time anyway,” “I’m not confident I could hand it off cleanly,” and “if they miss it, we scramble” — then you have AI-enhanced humans, not AI workers.

AI is everywhere. But returns are concentrated in the minority who have reshaped workflows, controls, and operating models around it.

The deeper question for CFOs and FP&A leaders

Stripped of buzzwords, here’s the question that matters over the next 12–24 months:

Which parts of your finance function will still rely on people doing the work, and which will shift to people managing digital workers who do it?

The brains — Claude, OpenAI, others — are already broadly available. The big consultancies are all telling you the same thing: AI needs to be embedded in the core of finance, with the right control environment, or you will be left behind. Gartner is clear that AI will increasingly power the “run” of finance, while humans focus on “change” and “decide.”

The differentiator won’t be “who has the best model?” — it will be who turned those models into reliable, governed digital workers first. It will be:

Who turned those models into reliable, governed digital workers — workers that own recurring workflows, sit safely on top of existing systems, and free humans to focus on judgment, narrative, and partnering with the business?

If your AI plan today stops at better assistants, you’re not wrong. You’re just stopping one crucial layer too early — exactly the layer Gartner calls “digital workers” and the one the Big 4 keep pointing at when they talk about the future of finance.

That missing execution layer is where the next wave of advantage — and real, measurable P&L impact — will come from.