Spreadsheet cleanup

Every company runs on spreadsheets that three systems export differently. Cleaning them is mechanical, repetitive, and exactly what an agent should do, as long as every change is listed rather than applied invisibly.

The job on one screen

Runs whenA file is dropped in the intake folder.
ReadsThe file, the target schema, and the rules for dates, currencies, and names.
DecidesWhat each column is, how to normalize it, and which rows cannot be saved.
ProducesA clean file, a change log, and a rejects file with reasons.
Stops whenClean file written. The rejects always go to a person.
Tools it needsFile parser, schema definition, storage.
Autonomy to start atApprove. It does the work and stops before the irreversible step.

Optional: load a real model

Run it

The fields it pulls

Required fields are never guessed. If one is missing the run stops and a person gets the document, which is the difference between an extraction agent and a wrong database row.

FieldRequiredHow it is found
Supplieryes^\s*([A-Za-z][\w&.\- ]+?)\s*[;,]
Dateyes(\d{1,2}/\d{1,2}/\d{4}|\d{4}-\d{2}-\d{2})
Amountyes([A-Z]{3}\s?[\d,]+\.?\d{0,2})
Termsoptional\b(net-?\d+)\b
Referenceoptional\b([A-Z]{2}-?\d{3,})\b

Where this one goes wrong

Ambiguous dates

03/04/2026 is two different dates depending on the country of the person who typed it. Never guess: require the source's format or reject the row.

Silent rounding

Currency conversion inside a cleanup step introduces errors nobody can trace. Keep the original column.

Fuzzy name merging

Acme Ltd and Acme Limited are probably the same. Probably is not good enough for a payment run.

How you would know it is working

MeasureWhy that one
Rows requiring manual fixingThe number to drive down, and the honest measure of the rules.
Silent correctionsShould be zero. Every change is in the log or it did not happen.
Rejects that were actually fineToo many and your rules are stricter than reality.

Earning more rope

AssistProduce the clean file next to the original for comparison.
ApproveWrite the clean file into the pipeline with the change log attached. Right level.
AutoFully automatic for a source whose format you control, never for one that arrives by email.
Log every transformation. The value of this agent is speed, and the thing that makes it survivable is that a human can reconstruct exactly what it changed.

Related: Document extraction · CRM hygiene · Invoice matching · all agent jobs · Agent Lab home

Free from AI School - no signup, everything runs in your browser.