How to Brief an AI Report: Writing a Focus That Works
A short brief turns a generic AI report into one that answers your question. How to write a report focus: audience, question, measures, comparisons and limits.
· 6 min read · Summarix team
An AI report is only as useful as the question behind it. In Summarix you choose a dataset and can answer one optional question: What should the report focus on? Leave it blank and you get a broad overview. Write a short brief that names the audience, the question, the measures and the comparisons you care about, and the AI plans its KPIs, charts and commentary around it, while every number is still calculated by code from your data.
What the focus actually changes
It helps to know how the report is built. First, the AI plans: a title, up to six KPIs and up to six charts. Each KPI is one column with one calculation (sum, average, minimum, maximum, count or distinct count), and each chart groups one column by another as a bar, line, area or pie chart. Then Summarix's own code computes every figure from your dataset. Finally, the AI writes an executive summary, up to eight insights, up to eight recommendations and up to five notes on data quality, from those computed results.
The focus steers the planning and the writing. It decides which columns become headline KPIs, which breakdowns get a chart, and what the commentary pays attention to. It cannot change the arithmetic, and that is deliberate: keeping AI from making up numbers depends on the model choosing what to calculate, never calculating it.
| A focus can | A focus cannot |
|---|---|
| Choose which columns become the headline KPIs | Filter the rows to a date range (scope the dataset instead) |
| Ask for a breakdown by region, product, owner or month | Divide one column by another (add a calculated column first) |
| Set the audience and the level of detail | Use data that is not in the dataset |
| Ask the commentary to look for risks, gaps or outliers | Turn the report into a statistical forecast |
The five parts of a good brief
- Audience: who reads it and what they already know. 'For the board, non-technical' produces different writing from 'for the warehouse manager'.
- The question: the decision the report should inform, in one sentence. 'Which regions should get extra stock next month?' beats 'analyse sales'.
- The measures: the columns that matter, named exactly as they appear in your data, such as `Revenue`, `Units` or `Gross margin`.
- The cuts: how to break the numbers down: by region, by product line, by month, by sales rep.
- What to flag: risks, unusual values, missing data, or anything you want called out even if it is bad news.
Weak and strong briefs compared
| Team | Weak focus | Stronger focus |
|---|---|---|
| Sales | Analyse our sales | For the sales manager: Revenue by Rep and by Region, top 10 Customers by Revenue, and which reps have the most Deals with no Close date |
| Operations | Show delivery performance | For the ops meeting: Orders by Status, Late deliveries by Depot and by Week, and the routes with the most Late flags |
| Support | Tickets report | For the head of support: Tickets by Priority and by Channel, average Satisfaction score by Agent, and categories with rising ticket counts |
| Finance | Expenses | For the owner, plain language: total Amount by Cost centre and by Month, the five biggest Suppliers, and any month far above the others |
The stronger versions have one thing in common: they name columns that exist. If you ask for 'margin' and the dataset only has `Revenue` and `Cost`, the report cannot show margin. Add a `Margin` column in the spreadsheet before you upload; our data cleaning checklist covers that kind of preparation.
Scope the data before you write the brief
KPIs are calculated over the whole dataset, and charts with a date on the axis are grouped by month. So 'last week's revenue' is not something a focus can ask for if the file holds a year of orders: the KPI will be the year's total. Give the report a dataset that matches the question instead: a file or sheet tab with only last week's rows, a Jira filter or API link that returns only recent records, or a `Week` column you can chart by.
How you keep that scoping right week after week, especially for scheduled reports, is covered in keeping report data fresh.
Check what the AI did with your brief
Every report keeps a step-by-step build log: the profiling of your data, the masking of personal information, the plan the AI chose and any warnings. If a KPI is not what you expected, the log shows which column and calculation the plan picked. If a report fails, you can try again with a different focus, which creates a new report from the same dataset.
- A KPI used the wrong column? Name the right one exactly in the next brief.
- A chart is too crowded? Ask for the top 10 by a measure.
- The summary is too technical? Say who the reader is.
- A data-quality note mentions blanks or odd values? Fix the source before you rely on the numbers.
On paid plans, Ask AI lets you chat with the finished report for follow-up questions, answered from the report's own numbers. It is the quickest way to test whether a different focus would help before you run another report.
Reuse a good brief in schedules
Once a brief works, keep it. Scheduled reports have their own Focus (optional) field, so each daily, weekly or monthly run follows the same brief and the reports stay comparable over time. Our guide to scheduled email reports covers the timing and recipients.
Pick a dataset, write a three-line brief and see the report it produces.
Free plan: 5 AI reports a month, no card needed.
Putting it together
A good brief is short and specific: who reads it, what they need to decide, which columns matter, how to break them down and what to flag. Scope the data to the question, check the build log, and reuse briefs that work. For the writing that comes out the other end, see how to write an executive summary from data.
Frequently asked questions
Do I have to write a focus at all?
No. The field is optional, and without it you get a broad overview of the dataset. A focus helps when you have a specific question or audience, and it makes scheduled reports more consistent from one run to the next.
How long can the focus be?
Up to 2,000 characters, but three or four sentences usually work better than a page. Name the audience, the question, the columns and the breakdowns, and leave the rest to the plan.
Can the focus make the AI change the numbers?
No. The AI chooses which columns and calculations to use; Summarix's code calculates the results from your data. A focus changes what is measured and how it is described, not the arithmetic.
Why does my report ignore the date range in my focus?
Because a focus does not filter rows. KPIs cover the whole dataset. To report on a period, give the report a dataset that contains only that period, or add a column such as Week or Month to break the numbers down by.