home / use cases / data, spreadsheets and reporting / data-hygiene

Customer data cleanup you can audit

Once a month an Aivell sweeps your customer table for duplicates and drift — same company twice, three phone formats, dead email domains — writes every proposed fix to a change log, and pauses for approval. Only after a human says yes does anything change, and deletes are blocked outright.

Watch the demo

starts when: First Sunday of the month, the sweep begins

The problem

Customer data doesn’t break; it rots. A company signed up twice under slightly different names gets two invoices and takes offense at both. Phone numbers live in five formats, mailings bounce off domains that died in 2023, and every report is a little bit wrong because “ACME GmbH” and “Acme Gmbh.” are counted as two customers.

Everyone agrees it should be cleaned. Nobody does it, because cleanup is a weekend of squinting at near-duplicates — and because the tools that offer to help want your entire customer base uploaded first. For records full of personal data, that trade is the wrong way round: keeping the processing in-house is privacy by design under GDPR Art. 25, not an inconvenience to engineer around.

How your Aivell does it

  1. The sweep runs itself. First Sunday of the month, the Aivell scans the customer table — no export, no upload, read in place over your own network.
  2. Matching is tiered, not clever-by-force. VAT number matches are solid ground; name-plus-address similarity is a suggestion. The chore treats them differently on purpose.
  3. Everything becomes a proposal first. Merges, normalized phone numbers, corrected country codes — each written to the change log as a before-and-after pair. At this point, the database hasn’t changed.
  4. A human holds the pen. Greta gets the list and the chore pauses. She approves from her phone — the whole batch or line by line — and only then do updates run.
  5. Nothing is ever erased. The guardrail blocks deletes outright: a merge archives the losing record, and the append-only log keeps the trail permanent.

That pause is the feature. The tedious part — finding, comparing, formatting — is done by the box; the judgment stays with a person. Your records get cleaner every month, and every single change has a name, a date and an undo path.

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 · monthly customer cleanup ● active

first sunday of the month, scan the customer table for duplicates

match on VAT number first, then fuzzy on name plus address

normalize phone numbers and country codes to one format

obvious duplicates — same VAT, same address — propose a merge

anything fuzzy goes on the review list, never auto-merged

write every proposed change to the changes sheet, before and after

send Greta the list and pause — nothing runs until she approves

after approval, apply the merges and archive the log

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.

Database finds duplicates, applies approved fixes
Spreadsheets keeps the before-and-after change log
Email sends Greta the approval request

Sensible guardrails.

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

Database Updates run only after an explicit approval. Deleting rows is blocked entirely — merges archive, never erase.
Spreadsheets The change log is append-only. History cannot be rewritten, not even by mistake.

Questions, answered.

Could it merge two customers that only look alike? +

Not on its own. Exact matches — same VAT number, same address — are proposed; fuzzy matches go on a review list and are never merged automatically. And nothing at all changes until Greta approves the batch.

We'd never ship our customer base to a data-cleansing service. How is this different? +

It never ships. The matching, the normalizing and the reasoning all run on the box, inside your network. The customer table is read in place and updated in place — no export, no third-party processor to add to your records.

What does it actually fix? +

The slow rot — duplicate companies, five phone formats, country codes that disagree with addresses, email domains that stopped resolving. It proposes; the change log shows each fix as a before-and-after pair.

Can a change be undone? +

Yes. Every applied change is in the append-only log with its previous value, so reversing one is a documented step, not an archaeology project.

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