How I Automated My Weekly Business Review With AI (And Saved 2 Hours Every Monday)

The weekly review isn't just a status update. It's a communication artifact that serves three audiences: your engineering team (what shipped, what's blocked), your leadership (are we on track), and your future self (what did I learn this week).

Most PMs approach it the same way:

1. Open Jira. Filter by last updated this week. Scroll.
2. Open Slack. Search for channel-specific decisions. Scroll more.
3. Open meeting notes. Try to remember which decision was final.
4. Open last week's report. Try to remember what you said you'd do.
5. Write, edit, rewrite, ship.

The problem isn't the writing. It's the information assembly — pulling fragments from five different tools, reconciling conflicting statuses, and structuring it into a narrative that makes sense to a VP who hasn't been in the details.

The Productboard State of AI in Product Management report (October 2025) found that PMs spend 43% of their work week on documentation and administrative tasks. The weekly review is the most repetitive of those — same structure, same audience, every week. That predictability is what makes it automatable.

Most AI automation tutorials for PMs recommend a stack of paid tools: n8n, Zapier, Make, API connectors to Jira and Slack. That stack works. It also takes an afternoon to set up, requires maintenance when APIs change, and often needs engineering approval.

This workflow uses two things:

1. Claude Projects (free tier works — Claude Haiku is available at no cost)
2. Your previous week's report

That's it. No Jira connector. No Slack API key. No automation platform.

The insight is counterintuitive: you don't need real-time data integration to automate your weekly review. You need your brain dump, structured well. Claude reads your raw notes and last week's report, then produces a draft that matches your voice and format.

This works regardless of your company's tool stack — Jira, Linear, Asana, Notion, or a whiteboard and sticky notes. The AI doesn't care where your data lives. It cares about the pattern in how you communicate.

Claude Projects are persistent workspaces. Everything you add — reference documents, custom instructions, conversation history — stays within the project. This means you set up your review format once, and every subsequent week's conversation inherits the same instructions.

Go to claude.ai → Projects → Create Project. Name it "Weekly PM Review" or similar.

In the Project Knowledge panel, upload last week's report as a PDF or text file. This is the single most important step — it gives Claude a working example of your voice, your section structure, and the level of detail your stakeholders expect. Without this, the AI output sounds generic. With it, the AI output sounds like you wrote a first draft in a hurry.

In the Custom Instructions field (under Project Knowledge), add:

• Your product name and what it does (one sentence)
• Your role and who receives your weekly review (e.g., "VP of Product, Engineering Lead, 2 squad leads")
• The sections your review always includes (e.g., "Shipped This Week, Blocked/At Risk, Next Week's Focus, Key Decisions Made, Metrics Watch")
• Your preferred tone (e.g., "Direct, no jargon, data-backed, call out risks explicitly")

These instructions persist across every conversation in the project. You'll never re-explain your format.

Each week, start a new conversation in your Claude Project. Paste this prompt:

Weekly Review Prompt Template

I need to generate my weekly PM review for the week of [DATES]. Here's my raw brain dump:

What shipped this week: [Bullet points — features, bug fixes, experiments, docs]

What's blocked or at risk: [Bullet points — with owners and expected resolution dates if known]

What I'm planning for next week: [Bullet points — top 3-5 priorities]

Key decisions made: [Bullet points — include rationale in 1 sentence each]

Team/morale notes (optional): [Any context a VP should know but wouldn't see in Jira]

Metrics worth watching (optional): [Any numbers that moved — good or bad]

---

Instructions for output: Use last week's uploaded report as your style reference. Match its section structure, level of detail, and tone exactly.

Rules:

• Write in first person ("I", "we")
• No corporate passive voice ("was completed by the team" → "we shipped")
• Flag risks explicitly — don't soften problems
• Keep each bullet to 1-2 sentences
• Do NOT invent information. If my brain dump doesn't mention something, don't add it from general PM knowledge.
• If something is unclear, flag it: "[PM to verify: ...]"

That's the whole prompt. The magic is in the style reference — by uploading last week's report and telling Claude to match its format, you eliminate the hardest part of writing AI prompts (describing tone and structure in text). You just show the example.

Claude generates a complete draft in under 2 minutes. It will match your section structure, mirror your vocabulary, and flag gaps with "[PM to verify:]" markers.

Your job now is a 15-minute editing pass:

1. Fill in the [PM to verify] gaps — usually 2-4 items where you were vague in your brain dump.
2. Add stakeholder-specific context — the VP who cares about one specific metric, the engineering lead who needs a blocker unblocked.
3. Remove anything that doesn't sound like you — the AI sometimes adds a sentence of corporate filler. Delete it.
4. Verify the numbers — if you mentioned metrics in your brain dump, double-check them. The AI won't hallucinate numbers you didn't provide, but it might phrase them differently than you would.
5. Add your personal take — one sentence at the top or bottom that only you could write. "This week felt chaotic, but we landed the integration — huge credit to the platform team." This is the human signal that makes the review unmistakably yours.

The total is roughly 15 minutes — compared to the 90-120 minutes of the manual process. A Productboard survey of 379 PMs found that AI users save 1-2 hours per day on internal tasks. The weekly review accounts for most of one day's savings.

Here's the complete prompt, ready to copy into your Claude Project each week:

I need my weekly PM review generated for [WEEK OF DATES].

BRAIN DUMP:
- Shipped: [list]
- Blocked/At Risk: [list]
- Next Week Priorities: [list]
- Key Decisions: [list]
- Team Notes: [optional]
- Metrics: [optional]

OUTPUT RULES:
1. Match the structure and tone of last week's report (uploaded to Project Knowledge)
2. First person ("I", "we"). No passive voice.
3. Flag risks directly. Don't soften problems.
4. 1-2 sentences per bullet max.
5. Don't invent information. Use [PM to verify: ...] for anything unclear.
6. End with: "Next week's focus: [1 sentence]"

Week 1 — The AI output was structurally correct but slightly formal. It hadn't absorbed my voice from a single reference report. I spent 25 minutes editing. Still faster than manual.

Week 2 — Noticeably better. With two reports in the conversation history, Claude had more to learn from. Editing dropped to 15 minutes. The "[PM to verify:]" markers caught two things I'd forgotten — things that would've been missing from a manual draft too.

Week 3 — The AI started anticipating patterns. It correctly inferred that a "blocked on legal review" item from week 2 should carry forward to week 3's report without me explicitly saying so.

Week 4 — My brain dump got sloppier (I dictated it on a walk). The AI output was still clean. Editing was 12 minutes. I realized the workflow was training me — I was getting better at giving the AI what it needed.

Three things surprised me:

1. The 15-minute rule matters. I gave myself a hard stop: if editing took longer than 15 minutes, something was wrong with my prompt or brain dump. Time-boxing prevented perfectionism — the enemy of shipping a weekly review. A good review at 10am beats a perfect one at 4pm.

2. The AI caught things I missed. Twice, Claude flagged a dependency I'd forgotten to mention. Once, it noted that a "shipped" item from my brain dump contradicted an "at risk" note from the previous week's report. The AI isn't just drafting — it's pattern-matching across time.

3. Stakeholders didn't notice. After three weeks, nobody asked "did you write this?" The output, with 15 minutes of editing, was indistinguishable from my manually written reviews. That's the real benchmark.

If you want to apply this approach to your own workflow, here's the framework distilled into four decisions:

This workflow automates the writing of your weekly review. It does not automate the data collection. You still need to know what happened this week. The AI can't read your Jira (not without API integrations), can't attend your meetings, and can't know which Slack thread contained the critical decision.

For most PMs, that's the right trade-off. Data collection requires judgment — which items are report-worthy, which technical details need translation for leadership, which risks are real. The writing is pure labor. This workflow automates the labor and leaves the judgment to you.

If you do want full data automation — Jira → AI → Slack auto-post — that requires an API-connected stack (n8n or Make, plus organizational API approvals). That's a topic for a future post. But it won't produce better reviews than this workflow. It'll just save you the 5-minute brain dump.

The 2-hour time savings are the obvious benefit. The less obvious one is schedule psychology.

When your weekly review takes 90-120 minutes, you either write it Friday afternoon (ending your week on admin) or Monday morning (starting your week on admin). Friday reviews mean you're the person who stayed late writing a status report while the team went for drinks. Monday reviews mean your week starts with a retrospective instead of a forward look.

With the 15-minute AI workflow, you can write your review whenever your brain dump is complete. Dictate bullet points on Friday after standup. Run the Claude prompt during a coffee break. Edit between meetings. Ship before anyone asks.

Monday mornings become yours again — for strategic thinking, customer calls, the work that actually moves the product forward.

The Claude Project template — with the exact prompt, Custom Instructions format, and section structure — is available below. Set it up once, use it every week.

[CTA: Weekly Review Automation Template (Claude Project + exact prompt)]

Related reading: Explore more AI for product managers workflows. If you're new to using AI as a PM, start with AI-First PM course Module 1 — it covers this workflow plus PRD generation, user research synthesis, and stakeholder comms automation.

Sources & Further Reading:

• Anthropic. Claude Projects documentation.
• Productboard. State of AI in Product Management Report (October 2025).