How to automate weekly reporting with AI

Why freeze the format?

Reports earn trust through sameness.

Same sections, same order, same definitions, so readers compare week to week without re-reading the structure.

Prompt: “Here’s our weekly ops report format, remember it exactly. Section 1: the five headline numbers (new pipeline created, deals closed won, MRR, website signups, open support tickets) each with the change vs. last week in absolute and percent. Section 2: what moved and why, max four bullets, only for changes over 10%. Section 3: what’s at risk next week. Section 4: nothing else. If a number is unavailable, write ‘unavailable’. Never estimate.”

“Never estimate” is the most important line in any reporting prompt. A plausible wrong number is worse than a gap.

How specific do the sources need to be?

Name the dashboard, the filter and the date range. Vague sources are what make automated numbers drift.

Prompt: “Pull each number from these exact places: pipeline and closed-won from the ‘Team pipeline’ HubSpot dashboard filtered to this week Monday–Sunday; MRR from Stripe’s billing overview; signups from the GA4 ‘Signups’ report; open tickets from the Zendesk views page. Put the raw numbers in the ‘Weekly metrics’ sheet with today’s date before you write anything else. I want the audit trail.”

Ask for causes, not adjectives

Ask for causal candidates and the companion goes looking: the deals that closed, the campaigns that ran, the tickets that spiked and what they were about.

Prompt: “For every metric that moved more than 10%, go find the likely cause before you write about it. Check the deals that closed, the campaigns that ran, the tickets that spiked and what they were about. Write each bullet as: what moved, by how much, the most likely cause, and how confident you are. If you can’t find a cause, say ‘cause unclear’. Don’t narrate.”

Schedule it for Friday afternoon

Prompt: “Every Friday at 3pm, produce this report in our format, append it to the ‘Weekly Ops’ Notion page under a new heading with the date, and post the five headline numbers plus a link in #leadership. Leave the ‘what’s at risk’ section as a draft for me. I’ll finish it before Monday.”

Leaving one section for a human is worth stealing. It keeps an editor in the loop, the same split our operations users run across recurring jobs.

What trends will it spot on its own?

After a few months, the reports are a dataset. Ask the companion to read the whole run.

Prompt: “Read the last 12 weekly reports in the Notion page. What’s trending that no single week made obvious? Give me three things, with the weekly numbers behind each.”

Frequently asked questions