Every finance leader running a lean team has had some version of the same conversation in the last two years: a vendor, a board member, or an eager junior hire suggesting that AI can now do the job of two or three people. Some of that is true. A meaningful amount of it is vendor marketing dressed up as inevitability. The useful question for a fintech or SME finance function isn't "should we use AI" — that's already been answered by the tools quietly embedded in the software you already run. The useful question is where it earns its place, and where it still needs a human holding the pen.
Where it genuinely helps
Reconciliation and transaction matching
This is the clearest win, and the one with the least controversy attached to it. Matching bank transactions to invoices, flagging exceptions, and clearing the routine 90% of a reconciliation automatically is exactly the kind of pattern-matching task AI tools now do well and fast. For a lean team, this doesn't eliminate the reconciliation function — it eliminates the hours spent on the matches that were never going to be interesting, freeing the person doing it to focus on the exceptions that actually need judgment.
First-pass categorisation and coding
Categorising expenses, tagging transactions to the right GL account, and flagging anomalous entries against historical patterns are tasks AI handles with increasing reliability — particularly once it's been trained on a business's own historical coding decisions. The output still needs review, but "review a categorised ledger for errors" is a materially faster task than "categorise the ledger from scratch."
Drafting first versions of recurring reports
Variance commentary, board report first drafts, and management account narratives that follow a predictable structure period after period are strong AI use cases. The tool can pull the numbers, draft an explanation of what moved and by how much, and produce something a finance lead edits down rather than writes from a blank page. For a small team producing the same report monthly, this compounds quickly.
Scenario modelling and sensitivity analysis
Once a financial model exists, AI tools are genuinely useful for rapidly generating variations — "what does the cash position look like if collections slow by two weeks," "show me the model at three different growth rates." This doesn't replace the model itself, which still needs to be built with real judgment about the business, but it makes stress-testing an existing model far less labour-intensive.
Document and contract review at first pass
Extracting key terms from a supplier contract, flagging clauses that deviate from a standard template, or pulling figures out of an invoice or a data room document — this is well within current AI capability, and materially faster than manual extraction. It's a first pass, not a final one, but it's a genuinely useful first pass.
Answering "where did this number come from" questions
For a lean team fielding constant questions from founders, board members, or investors about specific figures, AI tools that can query a company's own financial data and explain a number's components save real time — provided the underlying data is clean enough to query in the first place.
Where it doesn't yet hold up
Judgment calls with incomplete information
AI tools are confident by default, which is precisely the wrong trait for the calls that actually define good finance leadership: how conservative to be in a revenue recognition policy given an ambiguous contract, how to size an own-funds buffer against a genuinely uncertain growth trajectory, whether a customer concentration risk is serious enough to flag to the board now versus next quarter. These require weighing incomplete, sometimes contradictory information against real consequences — not pattern-matching against training data.
Anything regulatory where the cost of being wrong is high
Own-funds calculations, safeguarding reconciliation, regulatory capital reporting — these need to be right, explainable, and defensible to a regulator, not merely plausible. AI-assisted drafting can support the process, but the final number and the reasoning behind it still need a human who understands the specific regulatory framework and can stand behind the answer. A confidently wrong AI-generated safeguarding calculation is a worse outcome than a slower, correct manual one.
Relationship-dependent work
Negotiating payment terms with a difficult supplier, having the conversation with a founder about why the burn rate needs to change, managing a board member's anxiety about a cash position — none of this is a data problem. It's a trust and communication problem, and it stays firmly human.
Messy, non-standard, or unstructured source data
AI tools are only as good as what they're reading. A lean SME with inconsistent invoicing, informal intercompany arrangements, or records still partly on paper or in someone's inbox will find that AI tools underperform badly against that mess — not because the tools are weak, but because the input isn't structured enough to extract reliably from. The unglamorous work of cleaning up source data still has to happen before AI adds much value on top of it.
Anything that requires accountability
An AI tool can draft a board report. It cannot be the person who stands behind the numbers when a board member pushes back, or the person whose name is on the regulatory filing. Accountability doesn't delegate to a tool, however good the draft it produced.
Strategic prioritisation under constraint
Deciding which of five pressing finance projects gets the team's limited hours this quarter, or when to bring in outside expertise versus building capability in-house, requires context about the business, its people, and its trajectory that goes well beyond what sits in the financial data. This is judgment layered on judgment — the part of the job that doesn't compress well.
The practical shape for a lean team
The teams getting real value out of AI right now aren't the ones trying to automate the CFO function. They're the ones being specific about which tasks are mechanical enough to hand off, and disciplined about keeping human review on anything that touches a number a regulator, investor, or board member will hold the business accountable for.
A workable split looks roughly like this: let AI take the first pass on reconciliation, categorisation, report drafting, and data extraction — the volume work that eats hours without requiring much judgment. Keep every judgment call, every regulatory submission, every negotiation, and every strategic prioritisation decision firmly with the humans on the team, informed by AI-assisted analysis but never delegated to it.
The bottom line
AI doesn't replace a lean finance team — it changes what a lean finance team can reasonably be expected to cover. The mechanical volume that used to require headcount can now genuinely be absorbed by tools. The judgment, accountability, and relationship work that actually defines financial leadership hasn't gotten any easier to automate, and isn't likely to for a while. The finance leaders getting the most out of this moment are the ones being precise about that line — not the ones assuming the whole function is about to be automated, and not the ones dismissing the tools as hype either.