"AI bookkeeping" is having a moment, and a lot of the marketing oversells it. The honest version is more useful: modern AI is genuinely excellent at some parts of bookkeeping and deliberately kept out of others. Knowing the difference is how you use it well.

What AI is genuinely good at

  • Reading documents. Extracting supplier, dates, net, VAT and gross from an invoice or receipt — including messy phone photos and multi-page PDFs — is something today's models do reliably.
  • Decoding bank statements. Turning a wall of transactions into cleaned-up merchant names, amounts, directions and running balances, tuned to how UK banks lay things out.
  • Proposing categories. Suggesting the right account for a line, and staying consistent with the calls your team has already made.
  • Spotting duplicates. Catching the same invoice arriving twice before it doubles up.

Done well, this removes the bulk of the manual data entry — the part nobody enjoys.

What AI should not do on its own

  • Post to the ledger blind. Every extracted figure should carry a confidence score, and anything uncertain should wait for a human.
  • Make judgement calls. Whether an expense is allowable, how to treat an unusual transaction, what a client meant — that's professional judgement, not pattern-matching.
  • Be the final word on tax. AI helps produce the numbers; the accountant is accountable for them.

The pattern that works: review by exception

The sweet spot is simple. Let AI read and categorise everything, attach a confidence score, and route only the ambiguous items to a review queue. Your team confirms the uncertain 10% instead of typing the obvious 90%. You get the speed of automation with none of the "the software posted something wrong" risk.

How BookEnu is built

BookEnu uses Anthropic's Claude to read documents and OpenAI embeddings to make them searchable — but it's designed around review by exception. Low-confidence extractions, line-total mismatches and missing periods go to a queue rather than the ledger. The accountant stays in control; the AI just does the typing.

Curious how that feels on real client work? Apply for early access.

General information, not tax or software-selection advice.