How it works

Weekly Ops Report is a four-agent pipeline. Each agent has one job and hands a typed artefact to the next, so a failure is attributable to a specific stage rather than hidden inside a single model call.

1. 1

   Data Collection Agent

   Pulls the week’s key metrics from the configured sources — user growth, revenue, activity and retention — and records source, frequency and last-update time for every value.

   Handoff → Raw metric set + source ledger

2. 2

   Data Cleaning Agent

   Checks for anomalies and missing values. Each flag carries a detection method, a severity and candidate root causes. Anything repaired is recorded as a repair, never silently patched.

   Handoff → Cleaned dataset + anomaly register

3. 3

   Analysis Agent

   Computes week-over-week and year-over-year change, classifies the multi-week trend for each metric, and distils the movements into ranked insights with confidence and evidence.

   Handoff → Trend series + insight table

4. 4

   Report Agent

   Assembles the four sheets — Data Overview, Trend Analysis, Anomaly Annotation, Operations Recommendations — into a single Excel workbook.

   Handoff → weekly-ops-report.xlsx

What the pipeline never does

• • Never invents a metric value — every number traces to a collected value.
• • Never presents an anomaly explanation as confirmed — causes are labelled candidates.
• • Never substitutes rule-based output for model output without labelling which it is.

← Back home