AI in finance is mostly applied to routine, backward-looking work: reconciliation, checking, contract review and first-pass analysis or modelling. The final review still requires a qualified person, which is why the roles most exposed are the processing ones.
What is genuinely in production
The honest starting point is that the conversation runs well ahead of the practice. Everyone is talking about AI in finance and almost nobody is sharing how to practically apply it.[1] That gap is why I have been building a practical AI in finance guide by interviewing CFOs and finance leaders about what they have actually built, with examples out of tech, automotive, construction and the not-for-profit sector.[2]
What comes back is consistent and fairly narrow. Reconciliation and checking. Routine contract review. First-pass analysis. Connecting live data so board reporting refreshes itself rather than being rebuilt by hand each month. Useful work, and mostly the work that was eating the month-end.
Where it works, and where it does not
Adoption tracks business complexity more than ambition. The good returns are showing up in simpler businesses, software and services, because the underlying model is easy to plug into. It is much harder in businesses with significant inventory, stock and logistics, because the products and the data are more complex.[3] If someone promises you the same rollout timeline for a distribution business that they delivered for a SaaS company, be sceptical.
There is also a capability ceiling that is easy to state. AI can handle 80 to 90 percent of basic finance tasks such as checking, reconciliation, analysis, forecasting or modelling, and the last 10 percent requires a qualified human who understands the challenge and what good looks like.[4] The last 10 percent is where all the risk lives.
The one rule
My rule is that AI produces a first draft, not a final answer. A qualified person checks every number and signs off anything that reaches a board, an investor, an auditor, a client or a regulator, so accountability stays human.[5] That is not caution for its own sake. The obligations ASIC places on directors for financial reporting do not become negotiable because a model produced the figure.
The clearest place this pays off is reporting, where a P&L that arrives late is history rather than management information.
Which roles this exposes
This is the part founders and finance professionals should actually plan around. Any finance role below the level of financial accountant is likely to be in trouble over the next few years, particularly backward-looking processing work in shared service centres and local accounts payable and receivable teams. Forward-looking roles, management accounting, FP&A and business partnering, which depend on working with the business and communicating, will be far less affected.[6]
| More exposed | Less exposed | |
|---|---|---|
| The work | Backward-looking processing: transaction entry, reconciliation, accounts payable and receivable, routine reporting assembly | Forward-looking work: management accounting, FP&A, business partnering, scenario work and commentary |
| Why | The inputs are structured and the output is checkable, which is exactly what automation is good at | It depends on working with the business and communicating, which needs context a model does not have |
For hiring, that has one practical consequence. Employers are not asking for expert-level AI capability in finance candidates. They are asking for people who are adaptable and translatable with technology, with genuine interest in automation and process improvement.[7] That is a reasonable bar and most good candidates already clear it.
If the automation question is really a question about what to hire next, start with how a startup finance team should be structured.
Common questions
How is AI actually used in finance today?
Mostly for routine, backward-looking work: reconciliation, checking, routine contract review, first-pass analysis and connecting live data so board reporting refreshes itself instead of being rebuilt by hand. It is far less used to produce final numbers, because the last stage of any finance output still needs a qualified person to stand behind it.
Can AI do financial forecasting?
It can produce a first pass. AI handles a large share of basic finance tasks including analysis, forecasting and modelling, but the final portion needs a qualified human who understands the specific challenge and what a good outcome looks like. Treat the output as a draft to be interrogated, not an answer to be published.
Which finance jobs will AI replace?
The most exposed roles are below the level of financial accountant, particularly backward-looking processing work in shared service centres and accounts payable or receivable teams. Forward-looking roles such as management accounting, FP&A and business partnering are much less exposed, because they depend on working with the business and communicating rather than on structured processing.
Do finance candidates need AI skills now?
Employers are generally not asking for expert-level AI capability. What they value is candidates who are adaptable and translatable with technology, with real interest in automation and process improvement. Being able to describe something you have automated or improved is worth more than a list of tools.
References
- What I see AI actually changing: everyone is talking about AI in finance, but no one seems to be sharing how to practically apply it.
- From a placement I worked on: it is based on interviewing CFOs and finance leaders about their actual AI builds, featuring examples from tech, automotive, construction and not-for-profit sectors.
- How I describe the automation shift: many of the good returns are in simpler businesses like tech or services, which have easy models, and it is much harder to implement in businesses with significant inventory, stock and logistical issues because of their complex products.
- Tom Hunter on the capability ceiling, speaking on a podcast interview about finance careers and technology: AI can handle 80 to 90 percent of basic finance tasks like checking, reconciliation, analysis, forecasting or modelling, but the last 10 percent requires a qualified human who understands what the challenge is and what good looks like for the outcome.
- How I describe the automation shift: it produces a first draft, not a final answer, and a qualified person must check every number and sign off anything that reaches a board, investor, auditor, client or regulator, so accountability remains human.
- From a finance structure I work with: any finance role below the level of financial accountant is likely to be in trouble in the next few years, particularly backward-looking processing tasks in shared service centres and local accounts payable and receivable teams, while forward-looking roles like management accounting, FP&A and business partnering, which require working with the business and communicating, will be far less impacted.
- From a recent search: they highly value finance candidates who are adaptable and translatable with tech, particularly in automation, process improvement and an interest in AI, without necessarily expecting expert-level AI capability.