Data entry and extraction
An agent that turns forms, PDFs, scans, emails and messy spreadsheets into clean rows in your spreadsheet, CRM or database, and asks a person only when it is unsure.
Companies keep paying people to retype information that already exists. Data entry is the second most in-demand admin skill in Upwork’s In-Demand Skills 2026 report.
It is what businesses hire for
Upwork lists the usual projects: form entry and transcription, database cleanup and de-duplication, e-commerce product listings, data migration between systems, and CRM entry.
UpworkIt is paid by the hour, and it comes back every week
Data entry specialists cost $10–$20 an hour, with projects from $50–$300 for basic entry up to $300–$1,200 for migrations.
UpworkIt is slow, error-prone, and rigid tools break
People get tired, skip fields and mistype numbers, and someone has to check the output anyway. Template-based extraction stops working when a form or supplier changes its layout.
What the agent does
- 01
Takes documents from where they arrive
On a schedule, or whenever your system sends them, a code step picks up new files from an inbox, shared folder, form tool or upload.
- 02
Reads any layout
An extraction agent runs on every document in parallel and fills the exact fields you define, such as name, date, amount, address and product code, whatever the layout. Scans and photos are read as images.
- 03
Checks its work
Code steps validate formats (dates, emails, totals), look up existing records to avoid duplicates, and standardise values such as state names, units and phone formats. The agent also reports how confident it is in each field.
- 04
Writes to your system
Clean records are added to Google Sheets, Airtable, your CRM or your database through a tool.
- 05
Cleans what you already have
The same workflow can run over an existing spreadsheet to de-duplicate, fill gaps and standardise.
Checked before anyone sees an answer
The guardrail
Nothing is saved on the agent’s word alone. Code re-checks formats and totals and looks for duplicates, and a record that fails a check or has a low-confidence field is held back from your system and sent to review.
Where a person steps in
Low-confidence records and failed checks go to the Human Review inbox with the document and the fields side by side, editable. Everything else goes straight through. For the first week, everything is reviewed.
What is recorded
Every document has a trace: what was read, which checks passed, who approved it and what it cost. You are emailed if anything fails.
The steps, in order
A one-off clean-up or migration can skip the source connection in step 5: upload the files and run the workflow once.
| Step | What you do in Studio | Typical time |
|---|---|---|
| 1Define the fields | Create the extraction agent and its output schema, which is your spreadsheet’s columns. | 20–30 min |
| 2Add your rules | Create a knowledge base with your formats, category lists and mappings. | 20–30 min |
| 3Test on real documents | Run 20–30 samples in the Playground and save them as a test suite with the correct answers. | 1 hour |
| 4Build the workflow | Trigger, fetch, Agent (for each), validate, Branch, Human review, save. | 1–2 hours |
| 5Connect your systems | Connect the source (inbox, folder or form tool) and the destination (Sheets, CRM or database). | A few hours per system for a developer |
| 6Run in review mode | Send everything to Human Review for the first week, then only the exceptions. | 1 week |
A worked example
A business entering 1,500 forms or documents a month. The inputs are assumptions about a typical business; replace them with your own numbers.
| Line | Assumption | Per year |
|---|---|---|
| Manual entry today | 1,500 documents a month × 5 minutes each, which is 125 hours a month at $15 an hour, the middle of the $10–$20 range | $22,500 |
| Review time with Studio | The agent handles every document; people review about 15% at 2 minutes each, which is 7.5 hours a month | $1,350 |
| Labour saved | Manual entry today, less the review time that remains | ≈ $21,150 |
| Faster turnaround | Records ready in minutes, not at the end of the week | Not counted |
| Total value | ≈ $21,150 | |
| Krazimo Studio cost | ≈ $1,400–$1,600 a year: the Starter plan at $99 a month (about 3,000 runs a month) plus model usage of about a cent or two per document | |
About 1,400 hours a year of typing, replaced by 7.5 hours a month of reviewing the documents the agent was unsure about.
See it working
on your own information.
Tell us about your business and we will walk through this use case with you in Studio.