Client-reporting starter kit
A small, inspectable example for turning approved figures into a draft client update.
Version 1.0 · 19 October 2026 · AI Agents UK
Scope: fictional source data, a proposed prompt, an editorial example and suggested tests. No importable automation, working app connection or recorded model benchmark is included. The acceptance tests have not been run against a live platform.
Use the kit in this order
- Open the fictional source CSV and checked metrics.
- Read the drafting prompt. Try it only with fictional or permitted data in an approved tool.
- Compare the output with the editorial example.
- Use the reviewer checklist.
- Read the eight acceptance tests. Some require an implemented workflow; a prompt alone does not implement duplicate controls or failure notifications.
- Use the implementation brief if you commission a build.
- Explore your assumptions in the local time and cost calculator.
Check the example arithmetic
Spend changes from £2,000 to £2,200: +10%. Leads change from 100 to 121: +21%. Cost per lead changes from £20 to £2,200 ÷ 121 = £18.1818…: −9.09% after rounding.
Before using real data
Agree your reporting definitions and access rules. Remove unnecessary identifiers, keep source references, and identify who approves the result. The prompt is not a security boundary: control tool permissions and validate both inputs and outputs.
For the full explanation, see the client-reporting guide when published.