Weekly metrics reporting agent

TL;DR: Build a hands-off weekly metrics agent that pulls Friday data from Google Drive, computes KPIs, auto-plots charts, drafts a narrative, and drops a ready-to-share business report into your Slack or inbox. ## Connecting to live data The first step is to give the agent permanent access to your source of truth. In the demo the file lives in Google Drive, so we add a Drive connection and mark it “Agent-owned”. - **Agent-owned** means the connection uses a service-account style token instead of an individual user’s OAuth. - Once granted, the agent can read Sheets (or any Drive file) on a schedule without anyone staying logged in. - No more “download CSV → drag into notebook → rerun script” every Friday. ## Letting ChatGPT refine the workflow Instead of hand-coding every instruction, we describe the goal in plain language: “On Fridays pull last week’s data, calculate our three core KPIs, create visuals, and write a short narrative.” ChatGPT suggests wrapping the KPI logic into a **Skill** called “Metrics Calculation”. - Skills are reusable sub-routines that encode team definitions (e.g., how churn is counted, what “active user” means). - They turn ad-hoc prompts into deterministic steps, so the agent doesn’t improvise each week. - The skill is version-controlled and can be unit-tested independently. ## Scheduling the run We open the agent’s scheduler and set: - **Trigger**: every Friday at 09:00 UTC - **Starter message**: “Run analysis” That single sentence is enough; the agent loads the Skill, pulls the latest Sheet, and executes the full pipeline unattended. Team members no longer need calendar reminders or Slack bots asking “Who’s compiling this week’s numbers?” ## Observing execution in Activity History Every run is logged under **Activity History**. Opening an entry reveals: 1. Files accessed (direct links to the Sheet revision) 2. Code cells executed (Python snippets that compute KPIs and generate charts with matplotlib/seaborn) 3. Console outputs (row counts, any data-quality warnings) 4. Final artifacts (PNG charts + markdown report) If numbers look off, you can replay the exact code against the same Sheet snapshot for debugging. ## End-to-end output After computation the agent: - Generates 3–4 charts (weekly active users, revenue, churn) saved as high-resolution PNG. - Writes a concise markdown narrative: highlights week-over-week deltas, flags anomalies, and adds context bullets. - Bundles everything into a single PDF or posts directly to a Slack channel or email alias. The PM opens Slack on Friday morning and the report is already there, ready to forward to leadership—no copy-paste, no formatting. Source: [OpenAI YouTube – Weekly metrics reporting agent](https://www.youtube.com/watch?v=AwCNb_xI4fI)