Ask ten finance leaders where to start with agentic AI and nine will say "somewhere safe." Then they'll list close tasks that save an hour a month. The tenth will say revenue recognition, and they'll be right.
Rev rec is the highest-leverage workflow in the office of the CFO. It's also the one most teams are least willing to touch. Both things are true, and both are why it's the use case that separates real programmes from theatre.
Why it's the right target
Revenue recognition is judgement-heavy, contract-heavy, and policy-heavy. Every quarter, someone reads the same 40-page MSA, identifies the same performance obligations, and writes the same memo with slightly different numbers. The work is structured enough for an agent to draft, and consequential enough that the time saved actually matters.
A well-scoped rev rec agent doesn't post anything. It reads the contract, extracts the obligations, proposes the SSP allocation, drafts the memo, and hands a finance reviewer a one-page summary with citations back to the contract clauses. The reviewer signs off in minutes instead of hours.
Why teams stall
Three objections come up every time:
- "The auditor will hate it." They won't, if you can show the eval harness, the human-in-the-loop sign-off, and the citation trail. Auditors hate opacity, not automation.
- "Our contracts are too bespoke." They're not. They're bespoke in the same six ways. Agents are good at the long tail of structured variation; that's the whole point.
- "We tried OCR five years ago and it was useless." This isn't OCR. The failure mode of an LLM reading a contract is wrong judgement, not wrong text — which is a much more useful failure to evaluate against.
A staged rollout that actually deploys
- Read-only memos. Agent drafts the rev rec memo for new contracts. Finance reviews. Zero system writes.
- Suggested allocations. Agent proposes SSP and timing. Finance approves into the sub-ledger.
- Standing contracts. Agent monitors modifications and flags reassessment triggers. Still no auto-posting.
- Routine reposts. Only after two clean quarters of evals — and only for contract types the agent has demonstrably mastered.
Most teams never get past step two, and that's fine. Step two alone takes a week of quarter-end work and gives it back to the business.
The point isn't to take the accountant out of the loop. It's to give them back the part of the job they trained for.