home / use cases / data, spreadsheets and reporting / spreadsheet-reconciliation

Reconcile sheets against the database

Every Friday an Aivell lines up your payments sheet against the accounting database — or two sheets against each other — matches the rows, ignores rounding noise, and writes a mismatch list with a one-line explanation for each. You get the summary by email; the source data is never touched.

Watch the demo

starts when: Friday at 17:00 — or the moment the statement lands by email

The problem

Every close hides the same chore: two lists that ought to agree and don’t. The payments sheet against the ledger, the bank export against the bookings, the warehouse count against the stock table. Someone sorts both by amount, drags a highlighter down the screen, and spends an afternoon discovering that most differences are rounding — and that the two real ones are buried in row 214.

It’s slow, it’s error-prone, and it lands in the worst week of the month. Hidden admin like this costs businesses around $17,000 per employee per year — much of it exactly this kind of checking work that produces nothing new, only confirms what should already match.

How your Aivell does it

  1. The trigger is yours. Friday at five, month-end, or the moment the statement arrives by email — the chore starts without anyone remembering to.
  2. Both sides are read in place. The sheet through the Spreadsheets tool, the database over a read-only connection. Nothing is exported, nothing uploaded.
  3. Rows are matched the way a human matches them. Invoice number first, then amount and date as fallback — and your quirks, described in a sentence, are applied: prefixes stripped, gross versus net understood.
  4. Noise is filtered before it reaches you. Cent-level rounding is marked and set aside. Only real differences — over your threshold, or missing entirely — make the list.
  5. Every mismatch comes explained. One line each, on a tab of its own, with the Friday summary in Ingrid’s inbox: matched, mismatched, missing.

The guardrails are the point: the chore can write only its own tab and can never modify either source. It flags; a human decides. Your ledger stays exactly as trustworthy as it was — just checked every single week, behind your firewall.

The chore, as you'd write it.

No flowcharts, no code — a chore is just a message to your Aivell, in your own words. It shapes it into a solid, guarded procedure and follows it to the letter. This one:

Giulia Aivell · reconcile payments sheet ● active

every friday at 17:00, open the payments sheet

pull the week's invoices from the accounting database

match rows by invoice number first, then by amount and date

differences of a few cents — mark as rounding, don't bother anyone

over €50 apart, or a row missing on either side, goes on the mismatch list

write the mismatch list to its own tab, one line of explanation each

email Ingrid the totals — matched, mismatched, missing

never correct either source; flag only

You
Got it — I'll start right away, and check with you whenever something needs a human.
Giulia Aivell

The tools it uses.

Connected with single-click passkeys — no copied tokens, no OAuth hell.

Spreadsheets reads both sides, writes the mismatch tab
Database supplies the week's records, read-only
Email sends Ingrid the Friday summary

Sensible guardrails.

Every tool can be limited to exactly what this chore needs. For this one, you might set:

Spreadsheets Writes only to the reconciliation tab it owns. The payments sheet and every other tab stay read-only.
Database Read-only, full stop. The chore can look, never touch.

Questions, answered.

What does "explaining" a mismatch actually look like? +

A sentence a bookkeeper would write. "Invoice 4411 was paid in two installments — the sheet has one row, the database has two." "Row 87 uses the gross amount; the database stores net." You start from a diagnosis, not from two raw lists.

Can it compare two spreadsheets when there's no database involved? +

Yes. Bank export against bookkeeping sheet, warehouse count against stock list — any two sources it can read. The chore just names them.

This is our financial data. Who else processes it? +

Nobody. Matching and reasoning run onboard the box, inside your network. Neither the sheet nor the database extract is uploaded to any reconciliation service.

Our two sources don't even have the same columns. Does that break it? +

No. You describe the mapping in plain words — "their reference is our invoice number without the prefix" — and the Aivell matches by meaning, not by column position.

Related.

What is an Aivell?

An Aivell is an on-premise AI colleague: a small box that plugs into your network, runs its own AI onboard and takes the repetitive work off your desk. Prompts, documents, data — nothing ever leaves your office. Unplug the internet: it keeps working.

You describe each task as a chore, in plain language. Your Aivell turns it into a solid, guarded procedure and runs it in the background — with guardrails on every tool, a complete audit trail, and approvals in your hand when you want them. One-time setup, fixed monthly fee. No tokens, no overages.

Watch the demo

onboard inference · air-gapped · fixed monthly fee