§ Playbook · Finance
The AI playbook for Finance
Finance AI adoption flatlined last year - 58% to 59% - because the easy hopes met real controls. This playbook covers where the money is (accounts payable, forecasting, the close), what a hallucinated number costs, why the Big 4 cutting graduates is your problem too, and a 90-day route that your auditors would sign off.
Free · no form, no follow-up sequence
Deloitte refunded part of a AU$440,000 government contract because its report contained AI-invented citations and a fabricated court quote. A Big 4 firm, paying cash for a hallucination. Every number a machine produces in your name now needs a chain of custody.
58% → 59%
finance AI adoption 2024 to 2025 - the S-curve stalled mid-slope.
Gartner, November 2025
$2.75 vs $13.11
cost per invoice, automated versus manual; best-in-class runs 52.8% touchless.
Ardent Partners benchmarks, 2025
−44%
Big 4 UK graduate accounting job listings in one year, as AI takes the grunt work.
City AM / Accountancy Age, 2025
§ Part one
What actually changes.
Finance's AI story is unusually honest, because finance measures things. Adoption went 37% to 58% to 59% - a sprint, then a stall, as knowledge-management demos met data quality and control frameworks. Meanwhile the evidenced wins are specific: automated invoices cost $2.75 against $13.11 manual, best-in-class AP runs 52.8% touchless while the average shop sits at 25%, and Unilever's demand forecasting cut error by around 30% and freed some $300m in holding costs.
Notice what those have in common: transactional volume, structured data, clear rules. That is where finance AI works today. The top reported use case across the market is still knowledge management - search-and-summarise - which is pleasant, and not where the margin is. The gap between the 25% average and the 52.8% best-in-class is the actual opportunity, and it is mostly process and data work, not model work.
01
AP and AR go touchless, gradually
Invoice capture, matching, exception routing, payment scheduling. The best run above 70% touchless with sub-day cycles; the average is 25%. Closing that gap is worth more than any pilot on your list, and it is mostly plumbing.
02
Forecasting gets measurably better
Unilever's 30% error reduction is the reference case; FP&A teams report 15-20% accuracy gains and much faster revision cycles. The machine assembles and projects; humans argue about the assumptions - which was always the valuable part.
03
The close compresses
Reconciliations, accruals, variance commentary drafted by machine and reviewed by people. Unilever publishes FP&A commentary three times faster. Days come out of the close without anyone heroically not sleeping.
04
Expense and spend get audited completely
Every transaction reviewed rather than sampled. Anomalies surface in days, not at year-end. Mildly terrifying for a few colleagues; excellent for everyone else.
05
Judgement work stays yours
Capital allocation, going-concern calls, the estimate that requires a view. No credible evidence supports automating any of it, and the regulators - PCAOB and FINRA both now publish AI guidance - would like a word if you try.
06
Search-and-summarise is a starter, not a strategy
Knowledge management tops Gartner's use-case list, which mostly reveals how early this still is. Nice to have; nothing like 52.8% touchless. Do not let it absorb the budget.
§ Part two - the honest bit
What it costs you.
The role change is at the bottom of the pyramid, and it is compounding a crisis that already existed. AP clerks and junior accountants do exactly the structured, rule-based work that automates first. The Big 4 cut UK graduate intake sharply - KPMG by nearly 30% - and graduate accounting listings fell 44% in a year. Meanwhile accounting degrees are down 6.6%, CPA candidates have declined for a decade, and around 75% of CPAs are at or near retirement. AI is removing the training jobs from a profession that was already failing to replace itself. Your future controllers were supposed to come from somewhere.
Inside corporate finance, though, note the actual evidence: 88% of CFOs report no AI-driven headcount reductions so far. The cuts are at the Big 4 and the BPO layer; in-house, it is redeployment - so far. Say that honestly, and say what would change it.
01
The transactional layer thins
AP, AR, expense processing and reconciliations need fewer hands as touchless rates climb. In-house it has mostly meant redeployment to date - but that is a choice you have to keep making on purpose, with retraining attached.
02
The pipeline problem is now yours
If the Big 4 train fewer graduates and your own entry-level work automates, where does your next financial controller learn the craft? Building a deliberate development path just became a finance responsibility, not an HR nicety.
03
Hallucination is a financial control now
Deloitte refunded real money; Gartner found 67% of enterprises with production AI had a hallucination incident last year. Any machine-produced number needs lineage, review and sign-off - the discipline you already apply to humans, extended to software that never doubts itself.
04
The ERP estate faces both ways
SAP and Oracle are bolting agents onto systems you have customised for a decade, while agent-native challengers pitch bypassing the ERP screen entirely. Those customisations and workarounds are organisational capital - decide which are worth carrying forward, and which are sunk.
05
Offshore finance centres get repriced
The labour-arbitrage model is directly attacked by sub-dollar transaction costs; providers are pivoting to 'digital employee' pricing before their clients notice. Your shared-services and BPO contracts are negotiating leverage with a renewal date.
06
The regulators arrived early
PCAOB inspection priorities target AI-assisted audits; FINRA's 2026 report has a dedicated GenAI section. Governance is not the brake on finance AI - it is the licence for it.
§ Part three
What good looks like.
Good finance runs a mostly-touchless transactional core, a forecast humans argue about rather than assemble, and a control framework that treats machine output with the same scepticism as human output - documented, reviewed, explainable. The team is smaller at the bottom, more analytical in the middle, and still has a way of growing its own controllers.
01
A touchless core with governed exceptions
AP and AR above 50% touchless and climbing, exceptions routed to people with context, cost per invoice on a dashboard someone actually owns. The gap to best-in-class is a plan, not a benchmark you read once.
02
Numbers with a chain of custody
Every machine-produced figure traceable to source data, model and reviewer. Auditors get lineage instead of assurances. Nothing goes to the board that could turn out to be confidently invented.
03
FP&A on scenarios, not assembly
Analysts spend their time on drivers, scenarios and the business conversation - the work the assembly used to crowd out. Forecast accuracy is tracked and improving, and everyone knows why.
04
A pipeline you rebuilt on purpose
Entry-level roles redesigned around review, analysis and exception judgement rather than data entry - with rotations and mentoring that turn them into your controllers, because nobody else is training them now.
The change, across all five pillars
People
The bottom rung redesigned around review and exception judgement - not deleted and regretted for a decade.
Processes
Controls, lineage and sign-off designed before automation switches on, with auditors as design partners.
Data
Master data fixed first, because automation industrialises whatever quality it finds.
Platforms
ERP agents evaluated on evidence, with customisations priced honestly as assets or anchors.
Partners
Shared-services and BPO contracts repriced against sub-dollar transaction economics at the next opening.
§ Part four
The first 90 days.
Start where finance always should: with the measured gap. Your touchless rate versus best-in-class is the business case; your data quality is the constraint; your controls are the route to permission.
Ⅰ
Weeks 1-2
Benchmark the transactional core
Touchless rate, cost per invoice, close duration, exception volumes, forecast error - and an honest read on master-data quality, because it will decide everything downstream.
Ⅱ
Weeks 3-6
Automate one process, controls first
Pick AP matching or one close reconciliation. Design the review, lineage and sign-off before switching anything on - your auditors as design partners, not a surprise at year-end.
Ⅲ
Weeks 7-10
Prove it against the baseline
Touchless rate, cycle time, exception quality, incidents. Zero unexplained numbers is a pass/fail criterion, not an aspiration. Publish the result to the team and the audit committee.
Ⅳ
Weeks 11-13
Scale, and redesign the bottom rung
Extend to the next process, take the benchmark gap as a rolling target, and publish the redesigned entry-level path - what juniors will do, learn and become now the data entry is gone.
The day-90 scorecard · what you should be able to show
§ Part five
The traps, so you can skip them.
Trap 01
Trusting a confident number
Machine output arrives fluent, formatted and occasionally fictional. Deloitte paid a refund to learn this in public. Lineage and review on everything, or eventually you sign something invented.
Trap 02
Automating on dirty master data
Gartner's top adoption barriers are data quality and literacy. Automating a process fed by bad data industrialises the badness. Fix the data first; it is duller and it is the actual work.
Trap 03
Stopping at search-and-summarise
A chatbot over your policy documents is pleasant and saves minutes. Touchless AP saves millions. If knowledge management is your flagship use case after a year, the strategy is stalling politely.
Trap 04
Cutting the rung you climbed
Deleting junior roles saves money this year and costs you controllers for a decade - into a talent market the Big 4 are draining from the other end. Redesign the jobs; do not just delete them.
Trap 05
Letting the ERP decide
Waiting for SAP or Oracle to solve this preserves the licence fee, not your options. Evaluate the embedded agents against alternatives on evidence, and price your own customisations honestly - some are assets, some are anchors.
Trap 06
Treating governance as friction
The controls are why finance gets to scale AI while other functions re-run pilots. A framework your auditors trust is a competitive speed, not a tax on it.
§ Before you go
Six questions worth an honest hour.
Score yourself before someone with a rate card does. If more than two answers are uncomfortable, that is not a crisis - it is an agenda.
The self-assessment
§ Start with the workshop
Book your
Catalyst Workshop.
Bring the strategic change you have been circling. In 1-2 executive days, we will turn it into the first move, the case for it, and the mobilisation plan.