How to use DataMimi

The whole thing in three steps

1. Attach a spreadsheet with the paperclip, or paste a photo of an invoice or receipt. It reads .xlsx, .xls, .csv and PDF, including workbooks with many sheets.

2. Ask a question in plain English. Not a formula and not a column name — the question you would ask a colleague who had the file open.

3. Read the answer, then turn it into something you can send: a slide deck, a written report, or a cleaned Excel file.

You do not have to know what is in the file

After the upload it tells you what it found — every sheet, how many rows, which columns look like dates or money, and what is blank. If the file is a mess, that summary is often the useful part on its own.

Start here if you are not sure

This works on any sales or transaction export and is the fastest way to see whether the tool is worth your time.

Say what you want to know, not how to compute it

“Which three customers grew the most this year” works. “Group by customer and sort descending” works too, but you do not need it — and the first version lets the analysis pick a better method than you specified.

Name the period and the measure

“Revenue by month” is fine. “Revenue by month for 2026, excluding refunds” is better, because it removes a decision that would otherwise be made for you and mentioned in small print.

Keep going in the same conversation

Follow-up questions cost fewer credits and keep the same file and the same context. “Now split that by region” is a follow-up; re-uploading the file and starting again is not.

The suggestions under each answer are real follow-ups, generated from what was just computed. Clicking one is usually faster than typing.

Light, normal or thorough

The picker beside the send button decides how much work goes in. Thorough plans the analysis before writing any code and retries further when something looks wrong — use it on multi-sheet workbooks and forecasts. Light is for a quick count on a file you already understand.

Messy is expected

Merged cells, a title above the header row, totals in the middle of the data, blank columns between sections — these are normal in real exports and are handled. You do not need to tidy a file before uploading it.

Several files at once

Attach more than one and ask a question that spans them — orders against a product list, this month against last month. The panel on the right lists every file and every sheet, with tick boxes for what to include. Paid plans allow more files in one analysis.

Photos and PDFs

A photograph of an invoice or a receipt is read into a table, and the arithmetic is checked: line items against the subtotal, subtotal and tax against the total. When those do not agree it tells you rather than quietly using the printed number. Reading a photo needs an account and costs credits, because it is a real cost on our side.

How long files are kept

Thirty days, then deleted automatically. Conversations stay — reopening an old one after that asks you to attach the file again. Charts you have embedded on a website keep working: they are stored separately from the file they were made from.

Four things you can produce

A slide deck, ready to present, in one of several designs. A written report in Word. A cleaned Excel file, formatted rather than dumped. And an emailed summary on a schedule, if this is something you do every month.

The buttons appear under the newest answer once there is something to build from.

Sharing and embedding

Share turns the conversation into a read-only link. Anyone with it sees the analysis and cannot ask questions or spend your credits.

Any chart has an Embed button, which gives you an iframe to paste into a website or a blog. It stays live. Turning sharing off takes every embed down with it.

If an answer looks thin

Open the data table underneath it — the working is there, and “How I calculated this” says in plain English what was done. If the answer still does not cover what you meant, say so in the same conversation: “that is not what I meant, I want …” costs a fraction of starting again.

What a credit is

One unit of work. A simple question on a small file costs a couple; a thorough analysis of a large workbook costs more; a follow-up in the same conversation costs less than the first question, because the context is already loaded.

The estimate appears beside the send button before you send, and anything unusually large asks first.

A failed run is not charged

If something does not produce an answer, you are not billed for it. If you think you were, say so and it is put back.

Running out

Analysis pauses until your period resets. You are never charged automatically. You can move up a plan, or buy a one-off top-up from Settings — top-up credits do not expire, unlike the monthly allowance.