A dashboard gives you accurate numbers, but on a Monday morning a regional manager still has to look at ten charts and work out "what happened this period and what should I do", while senior management wants the story rather than the figures. This post explains how AI narrative summaries in BranchVoice turn period data into prose, which scopes they cover, and how generation cost is kept under control.
What an AI narrative summary contains
A summary for a chosen scope and period has four parts:
- Headline: the period's story in one sentence.
- Findings: the most important changes in the data: which topic rose, which stores diverged, where the negativity comes from.
- Recommended actions: concrete steps that follow from the findings, such as "review the returns process in three stores".
- Caveats: the limits of the data; for example, that a finding should be read cautiously because the period had few reviews or one platform delivered no data.
The summary is generated from processed data rather than from the raw reviews: period statistics, topic distribution, sentiment and risk breakdowns, a comparison with the previous period, and a handful of representative quotes. The AI writes the text from that input; the numbers are the same numbers you see on the dashboard. The classification underneath is described in AI review classification.
Which scopes get a summary
| Scope | Where it appears | When it is written |
|---|---|---|
| Brand | Overview, "AI period summary" card | Automatically when the numbers change, at most once a day |
| Lead centre | Lead Centre | Same rule as the brand summary |
| Store | Store detail page | When you click "Generate summary" |
| Topic | Topic detail page | On request |
| Comparison | Store, city or region comparison view | On request |
Each summary is kept separately per period (the default is the last 90 days) and per interface language, so Turkish-speaking and English-speaking teams read it in their own language. A region comparison summary describes, for example, the Aegean and Marmara regions side by side for the same period; the comparison logic itself is covered in store comparison.
When it is refreshed, and when it is not
Two rules govern generation:
- If the numbers have not changed, the summary is not rewritten. The summary's input is stored as a fingerprint. If no new reviews have arrived, or the new reviews have not changed the period's statistics, the existing summary remains current. When the numbers do change, the card shows a "Numbers changed" badge and "Update" writes a new version.
- There is a daily generation cap. The number of summaries that can be written per company per day is limited. When the cap is reached, the card says today's quota is used up and generation will be retried tomorrow. Reading is always free and unlimited.
Brand and lead summaries refresh themselves within these rules. Store, topic and comparison summaries are written only on request, so a chain with hundreds of stores does not generate summaries nobody will read.
Versions and seeing what changed
Every generation creates a new version; older versions are not deleted. The AI Summaries page under Administration lists the summaries for every scope with their version history. When you compare a version with the previous one, newly added findings are shown in green and findings that dropped out are struck through. That answers "did the returns problem flagged last month disappear from this month's summary" in seconds.
The same narrative also feeds the weekly digest email and the AI section of scheduled PDF reports, so a manager reading a report sees the period's story next to the charts. See weekly digest email for details.
How regional managers and senior management use it
The Aegean regional manager of a footwear chain reads the period summary before the weekly meeting: the headline gives the region's overall direction, the findings name the two stores that diverged, and the recommended actions become the agenda. For any finding they doubt, they drop into the dashboard with the same period filter and see the figure for themselves.
For senior management, the brand summary is the first page of the monthly management meeting. Four short sections instead of thirty charts; and because the caveats section states honestly what the data does not say, trust in the summary grows.
Three practical suggestions:
- Rather than forwarding the recommended actions to store managers as they are, share them with at least one source review; the customer's own words are always more persuasive.
- Before generating a store summary, make sure the period contains enough reviews with text; in thin periods the caveats will say so anyway.
- Make version comparison a monthly routine; findings that dropped out are problems that were solved, and the team deserves to see that.
Next steps
AI narrative summaries sit on top of the classification and dashboard layers in BranchVoice: once the data is there, the brand summary writes itself and store and topic summaries are one click away. No extra setup is required. To see what kind of narrative your own period data produces, request a demo.