By the end of this you will have a weekly performance summary written from a Sheets export, with player identifiers stripped before the model sees anything and the metric labels checked before the summary goes to anyone.
Two steps in this workflow are not optional. Step 1 is the anonymisation. Step 5 is the label check. Everything else you can adapt.
This is for BI, CRM, retention and country managers who write the same report every Monday.
The setup
A Google Workspace account with an eligible Gemini plan. Gemini in Sheets works best on native Google Sheets files, and Google states that Excel files need converting first. If your BI tool exports .xlsx, convert to Sheets before you start rather than wondering why the panel is unhelpful.
You also need to know what your own columns mean. That sounds obvious and it is the thing that breaks this workflow more than any model behaviour.
Step 1: strip the player identifiers
Do this before anything else, in the sheet, as a step you perform rather than a rule you remember.
Delete these columns outright: player ID, username, email, phone, full name, address, postcode, IP address, device ID, payment reference, any account number.
If your analysis genuinely needs a per-player row, replace the identifier with a sequential number in a new column and delete the original column. Keep the mapping in a separate file that never goes near a prompt.
Aggregate where you can. A report on cohorts, markets, segments and channels needs none of the above and answers the same questions.
Then check the sheet again for the identifier that hides in a free-text column. Support notes, bonus comments and manual adjustment reasons are where a player name survives a column deletion.
Step 2: fix the column headers before you prompt
Rename every metric column to something unambiguous. GGR, NGR, deposits, withdrawals, bonus cost, FTDs, actives, handle. Spell out anything a colleague would have to ask about.
A column called “revenue” is the single most expensive header in iGaming reporting, because three teams mean three different numbers by it.
Add a units row or put the currency in the header. Mixed currencies in one column with no marker produce a summary that adds euros to pounds without comment.
Step 3: open Gemini in the sheet
Open the spreadsheet and click the Ask Gemini button at the top right. A side panel opens where you enter prompts.
Start with a description of the data rather than a request, so you can see whether it read the sheet the way you did.
Describe this sheet before analysing it. List every column, what you
understand it to contain, and the units or currency you think it is in.
Flag any column whose meaning is unclear.
Do not summarise the data yet.Read that list against your own understanding. Any column it has misread is a column that would have produced a wrong number in the summary.
Step 4: ask for the summary against named metrics
Write a weekly performance summary from this sheet.
Cover, in this order:
1. Week on week movement for each of: GGR, NGR, deposits, FTDs, actives
2. The three largest movements by percentage, with the segment or market
they came from
3. Anything that looks like a data problem rather than a performance
change: a zero, a duplicate, a value an order of magnitude out
Rules:
- Use the column names exactly as they appear in the sheet. Never
substitute one metric name for another.
- State the currency with every figure.
- If a movement could be explained by fewer days of data, say so.
- No recommendations.Section 3 earns its place every week. A model reading a spreadsheet is good at noticing the row where a decimal went missing, which is a job nobody enjoys doing by eye.
Step 5: check the labels before the numbers
Read the summary once, looking only at metric names.
The model takes your headers at face value. A column headed GGR that actually holds NGR produces a summary that says GGR, and the totals will look plausible either way because both are revenue-shaped numbers of a similar order. Nothing in the output can catch that. Only someone who knows the export can.
So the check is on your side of the line: confirm that each metric named in the summary matches what that column actually holds. Then check the arithmetic on the single largest movement against the sheet.
Step 6: move it into a Doc
Copy the summary into a Google Doc for circulation, and add two things the model cannot: the reason behind the largest movement, and what you are doing about it.
Note in the Doc which week the data covers and when it was pulled. Reports get forwarded, and a summary with no date attached gets read as current 3 weeks later.
Where it breaks
Charts do not update
Google states that charts generated by Gemini do not automatically update when the source data changes. Build a chart on Monday, refresh the export on Tuesday, and the chart still shows Monday’s picture while sitting in a sheet that now holds different numbers. Regenerate charts with every refresh, or build them with the normal Sheets chart tools instead.
The conversation disappears
Conversation history in the Sheets side panel is not saved automatically and is lost if you reload the browser or close the spreadsheet. A refined prompt you spent 20 minutes on goes with it. Keep your working prompts in a text file outside the sheet.
Excel files need converting
Gemini in Sheets works best with native Google Sheets files. An .xlsx dropped into Drive and opened in compatibility mode is where the panel starts giving vague answers, and the cause is the file format rather than the prompt.
Mislabelled columns propagate silently
The failure in step 5, restated because it is the one with consequences. Every downstream reader of the summary inherits the label error, and a summary is more likely to be forwarded than the sheet it came from. Google’s own guidance is that Gemini may suggest inaccurate information and should not be relied on as professional advice, which covers the reporting case squarely.
The limits
Nothing that identifies a player goes into the sheet you point Gemini at. Step 1 is the workflow, not a caution attached to it.
This produces an internal reading of your own numbers. It does not produce regulatory reporting, and a figure from this workflow does not go into a licence return, a tax filing or an investor update without coming from your reporting system with the normal checks on it.
Where your data protection team has rules on which systems player data may touch, those rules decide this, and a report being internal does not change that.
AI for iGaming is a recurring series from The iGaming Europe. Bartosz, our Head of Content, shares the prompts, setups and workflows he has tested on real iGaming work. One workflow per issue, including what went wrong. No tools to buy, no affiliate links.
Source: Google Workspace Help
