Your clients' AI closes the books. Your signature says they're right. This is the layer in between.
Accounting-Firm Copilot runs an independent verification pass across a client book: it re-derives the close from the same source data, diffs it against what the client (or their AI bookkeeping tool) produced, and reports agreement or named disagreements, line by line. Coding audit, AR/AP reconciliation, bank rec, financial-statement attestation, and a trust-account line for firms serving attorneys. Under your firm's brand, with your partner signing off on every release. Software does the re-derivation; your people do the judgment.
Talk to us about a firm pilot See a sample disagreement report
THE PROBLEM
Two curves are crossing on every managing partner's desk. Client books are increasingly closed by AI bookkeeping tools, faster and cheaper and mostly right, and "mostly" is doing heavy lifting. Meanwhile the bench that used to catch "mostly" is thinning: fewer accounting graduates, fewer exam candidates, seniors stretched across more engagements every year. So the review that protects the firm's signature quietly becomes sampling, and sampling is exactly wrong for AI-produced output, which fails in confident, plausible, systematic ways rather than obvious ones. When a mis-coded quarter or an unreconciled trust ledger surfaces later, nobody asks the software; they ask the firm that signed. The tools producing the books will happily grade their own homework. That grade is worth what self-grading is always worth.
HOW IT WORKS
The Copilot is an engagement you run per client, per period. It works the verification in order: first the coding audit, every transaction's categorization independently re-derived and diffed against the book, disagreements listed with the account and the reason. Then AR/AP: transaction reconcile, bank statement reconcile, and an aging report, with unreconciled items named rather than netted away. Then the close itself: we rebuild the bank reconciliation and financial statements from the same source data and compare them to the client's close, line by line. Then the attestation: the diffed statement pair, the artifact you can put in the workpapers to show the numbers were independently verified. Optionally, a client-facing board pack assembled from the now-verified numbers. Every line returns agree, disagree, or undetermined with the gap named, and a line that cannot complete never quietly passes. At the end, the whole bundle goes to your reviewing partner, and nothing is released to a client until a partner signs off. Firms serving attorneys can add the trust-account (IOLTA) reconcile line, the same discipline pointed at the ledgers that end careers when they drift.
WHAT YOU GET
Per client, per period, one engagement record your firm can stand behind: - the disagreement report, agree / disagree / undetermined per line, with named items and named gaps, the thing a reviewing partner actually reads, - the re-derived artifacts, our independently built bank rec and financial statements next to the client's, in the workpapers, - the attestation, the diffed statement pair documenting that verification happened, - the partner gate, nothing releases without sign-off, and the record shows who signed, - your brand on all of it, white-label per-partner license, IP licensed, never assigned.
WHO IT'S FOR
Accounting firms and CAS practices running 20 to 500 client books, especially where clients have adopted AI bookkeeping the firm is now expected to stand behind. Fractional-CFO shops and bookkeeping agencies delivering monthly closes at volume. Firms serving solo and small-firm attorneys who carry trust-account exposure. If you want an AI that does the bookkeeping itself, this is not that; it is the layer that checks whoever does, human or AI. And it is verification tooling for your engagements, not a PCAOB audit platform.
PRICING (the ladder)
Per-partner licensing, priced as a ladder; every number is a hypothesis we validate with you, not a commitment. - Firm pilot, fixed price: we run the full engagement line across 3 to 5 of your client books for one period, and your partners review every disagreement report before you commit to anything. - Book metered, per active client book per month once the pilot proves the verification holds, all lines included, trust-account add-on for firms that need it. - Platform, embed the verify layer in your own client portal and sell verification as your own branded service line.
IP is licensed, never assigned. Any step that takes your money is gated and confirmed by a human before it runs, nothing charges silently.
THE PROOF (dogfood)
AYA is its own first client book. The same verification spans this product white-labels are the ones we point at our own finances: the board-pack line reads our real bank account (balance, transactions, burn, forecast), and the close-verify, categorization-audit, attestation, and AR/AP lines are the committed deliverables of our Phase 1 and 2 finance products, verified on disk and quality-gated before this page was written. The engagement pattern that sequences them was built the same way everything here is built: composed from existing verified pieces, scored by our quality gate, with the verification receipts kept.
HONEST NOTE
We would rather under-promise, especially to buyers whose license is on the line. What exists: the five verification lines and the trust-account workflow, committed and structurally verified, plus the new engagement pattern that runs them per client with a partner-review gate. What has NOT happened yet: no accounting firm has run this, no client book other than AYA's own has been touched, and the engagement pattern has not yet executed end to end on a live mesh (structural validation only, and we keep that distinction in writing). The white-label packaging a firm rollout needs, the cross-book dashboard, hard per-client isolation surfaces, the engagement-letter workflow, does not exist yet and we say so. The first firm pilot is the test this product has to pass. Where a line can be decided we decide it and attach the evidence; where it cannot, it comes back undetermined with the gap named, and an undetermined never quietly becomes a pass. For a firm whose signature is the product, that refusal to fake certainty is the whole point.
Talk to us about a firm pilot
*This page is a specification. The capability it describes is not built yet, and nothing here is a claim that it runs today.*