AI Bookkeeping, Honestly: What It Is and What Ours Does

AI bookkeeping is software that reads your financial paperwork — receipts, invoices, bank lines — and turns it into accounting records without you typing anything. That is the honest definition, and it is genuinely valuable: the typing was never the part of bookkeeping that needed a human being.

The judgement was. So the question that separates AI bookkeeping products is not "can the AI read a receipt?" — they all can, most of the time — but what stands between the AI's reading and your books. This page explains what our answer looks like, in both directions: what the platform does, and what it deliberately does not.

What AI bookkeeping is (in general)

Under the marketing, most products in this category do some mix of: extracting data from documents (dates, amounts, counterparties), proposing a category or a ledger entry, matching transactions against bank statements, and flagging anomalies. Modern models do this impressively well on clean, familiar paperwork — and less reliably on smudged photos, unusual layouts, ambiguous purchases, and anything requiring knowledge of your business rather than your documents. Both halves of that sentence are true everywhere, whatever any vendor implies.

What ours does

Elizabeth Bookkeeping is built around one controlled workflow: AI drafts. AgentLedger validates. People approve.

• AI drafts. Evidence arrives — uploaded, read from connected email (read-only, one-click disconnect), collected from vendor portals by our browser extension, or photographed over WhatsApp or Telegram. The AI proposes a full double-entry transaction and attaches a confidence score: its own estimate of how sure it is.
• AgentLedger validates. Every draft must pass our plain-text ledger kernel's deterministic checks — postings balance to zero, accounts exist, dates are sane. Failures are rejected with reasons, never quietly repaired. No model sits in the validation path.
• People approve. A human accountant approves, corrects, or queries every entry under an agreed review policy, working a queue ordered doubtful-first. Suspected duplicates are held for a human decision, never auto-posted or auto-discarded. Where you agree a per-account automation policy, routine low-value entries can post automatically within explicit confidence and materiality thresholds — logged, reversible, and manual-by-default until you opt in.

Every step lands in a permanent audit trail: source document, draft as proposed, score, edits, approver, timestamps. "Why is this entry here?" always has a reconstructable answer.

What ours does not do

We think the refusals are as much a feature as the capabilities, so here they are plainly.

• It does not file your taxes by itself. Making Tax Digital updates are prepared from your approved books and submitted to HMRC only with your explicit authorisation. Nothing files autonomously.
• It does not move money. No payment execution on AI output, full stop. A manipulated document can at worst produce a bad draft for a human to catch — not a transfer.
• It does not post on its own say-so. Outside the narrow automation policies you explicitly agree, nothing enters your ledger without a person's decision.
• It does not replace your accountant. It removes the data entry and orders the review work; the judgement — what is claimable, what is miscategorised, what deserves a question — remains human, and a named human is accountable for every entry.
• It does not promise perfection. Models are sometimes confidently wrong; humans occasionally miss things. The controls exist to make errors rare, visible, and correctable — not to pretend they are impossible.

Why the controlled workflow matters

Anyone evaluating this category should ask one question of every product: what happens when the AI is wrong? Because it will be, sometimes, and demo reels never show that day. Our architecture assumes it from the start — a deterministic validator that cannot be persuaded, a human decision on everything material, duplicates held rather than guessed at, and an audit trail that keeps every automation decision explainable to you, your accountant, or HMRC.

If you want the detail, how it works walks the workflow end to end with a worked example, and our buyer's checklist for AI bookkeeping software gives you the questions to put to any vendor — including us.

Check eligibility / Ask for Details