The bank recs, accrual tie-outs and invoice ticking that taught every accountant what "wrong" looks like are now done by software. The Accountant Residency is a practice floor where new grads and career changers close simulated books, catch AI's mistakes, and get reviewed by real seniors, so firms can hire people who already have judgment.
Of US accountants are nearing retirement age, with CPA candidates falling.
Ⓢ Randstad salary guideEvery AI-drafted reconciliation still needs a person who can tell when it is wrong. That instinct used to come from years of doing the work by hand. If firms stop hiring juniors now, around the early 2030s there is no one experienced enough to check the machine.
Like a medical residency: structured cases, supervised by people who have done the work, ending in a credential employers trust.
More residents attract firms. Firm hiring attracts residents. Both attract mentors.
Hired residents return a year or two later as mentors, so mentor supply grows with the user base instead of being recruited.
Every graded attempt and mentor correction shows where people get stuck, which sharpens the next simulation.
An AI agent categorized March bank activity for a small e-commerce client. Some of its calls are wrong. Flag the lines you would send back, say why, then check your review. This is the core loop: residents train as reviewers of AI, not replacements for it.
| Flag | Date | Bank description | Amount | AI categorized as |
|---|
The AI does the work with planted errors. Find them, explain each one. Trains review judgment.
Race the AI on a judgment call (cutoff or accrual?), then compare your reasoning with its reasoning and a mentor's.
One fictional company whose books get messier each month. Fraud cases unlock at higher levels.
Cohort leaderboards and a "Certified AI Reviewer" profile firms can verify before an interview.
| Scoring rule | Why it matters to a firm | Points |
|---|---|---|
| Error caught | Recall: the reviewer finds what the AI got wrong | +25 |
| Correct reason given | Judgment, not luck: explains the accounting | +10 |
| False flag on a correct line | Precision: over-flagging wastes partner time | −10 |
| Error missed | Shows exactly which skill to drill next | 0, logged |
First pilot: one 4-week simulated month-end close for a fictional e-commerce client, 10 to 20 residents from a university cohort, 3 to 5 mentors, then one question to two or three firms: would you interview from this list?
Marketplaces die when no side shows up first. Answer: AI grading makes it useful alone on day one; universities seed residents.
highThe data angle could read as exploitative. Needs explicit consent and a clear value exchange.
highIf platform hours counted toward CPA experience requirements it would be a moat. Requires state-board research.
unknownPartners are busy. Retired CPAs and returning alumni may be the real supply.
mediumThey will, for themselves. The lane is the mid-size firm that cannot.
manageablePick the piece you are reacting to. Short and blunt is perfect. Firms, educators, students, career changers and investors all see different holes.