AI Reporting: Trello, an LLM and SLAs in a spreadsheet

Context

Agile boards hold the truth about how work is actually flowing, and almost none of it is legible to the people who have to decide with it. The phase lives in a list name, the reasoning lives in the comments, and the SLA lives in nobody's head.

Problem

Reading a board by hand happens weekly at best, and the enrichment that makes it useful (which phase is this really in, what did the discussion conclude, is it breaching its SLA) is exactly the part a human keeps skipping.

Approach

AI Reporting synchronises Trello cards into Google Sheets automatically and enriches them on the way: an LLM classifies the phase, summarises the comments and monitors the SLA, so the sheet holds a current, decision-ready view instead of a raw export.

Non-obvious decisions

• Land in a spreadsheet rather than in a BI tool. The people making these calls already live in Sheets, and a report they can pivot themselves is used more than one they have to request.
• Enrich on ingest, not on read. Classification and summarisation happen once per sync, so the model cost is bounded by the number of cards rather than by the number of readers.
• Let the model read the comments. The phase label is often stale and the comment thread is where the real status is. Summarising it is the step that turns an export into a report.

Outcome

Trello activity arrives in Sheets already classified, summarised and checked against its SLA, centralising the insight needed for agile business decisions.