The task
Every Monday at 10am you drop a link to the team dashboard in #growth. Three people click. Nobody comments. The dashboard has 23 charts and the team is not going to read all 23, so the data sits there and decisions get made on vibes instead. You want a 4-line weekly takeaway pinned to the channel that tells the team what actually moved, why, the surprise nobody noticed, and the question that matters for this week — short enough that people read it in Slack without ever opening the dashboard. The dashboard becomes the receipt; the takeaway is the headline.
Where AI helps — and where it does not
AI is good at compressing 23 charts into a 4-line headline once you tell it which 2-3 metrics actually drive decisions. It is also good at the tone shift between audiences — the takeaway for leadership reads different from the one for the eng team, even with the same underlying numbers. Where AI fails: judging which metrics matter this week. Your team cares about activation right now because of a specific launch; next month they will care about retention because of a churn spike. The model does not know your current priority. You pick the metrics; the model writes the takeaway.
A common failure mode: the model treats every weekly takeaway like the most important week ever — every line ends with “significant,” “notable,” “alarming.” Train your team to ignore the takeaway in 4 weeks. Tell the model to call boring weeks boring, and save the language of urgency for actually-urgent weeks.
What to feed the AI
- The 2-3 metrics that actually drive product or growth decisions this week (not 15 vanity metrics)
- Each metric’s this-week value vs last-week, and ideally vs last-month average
- Known context — planned product changes that landed this week, ad spend changes, holidays, outages, a feature toggle
- The audience for this specific takeaway — leadership, eng, marketing, the full company; each has a different lens
- The 1-2 questions the team is currently trying to answer (the takeaway should advance one of them or admit it cannot)
- Whether this week is a “we shipped something” week or a “nothing changed” week — calibrates urgency
- Last week’s takeaway if you have it (so this week’s reads as continuity, not isolated)
- A boring-week clause — explicit permission for the model to say “nothing moved enough to matter; revisit next week” when that is the truth
Copy-ready prompt
Write a 4-line dashboard weekly takeaway for a Slack post.
Core metrics (with this-week vs last-week vs last-month avg): {paste}
Context (what landed this week, outages, holidays, ad spend changes): {paste}
Audience: {leadership / eng / marketing / full company}
Open questions the team is trying to answer: {1-2}
This week's type: {"we shipped something" / "nothing changed" / "anomaly week"}
Last week's takeaway (for continuity): {paste or "first week"}
Return exactly 4 lines (no more):
Line 1 — Headline. The one metric that moved most OR matters most, in this week's frame. Lead with the number; the change is the headline.
Line 2 — Why. Connect the move to context (launch, ad spend, seasonality). If no clear cause, write "no clear cause yet" — do not invent one.
Line 3 — One surprising number. The thing nobody else noticed in the dashboard. This is what earns the read.
Line 4 — Action or question for the team. Specific. Not "let's discuss."
Total under 60 words. No corporate hedging ("we will continue to monitor closely"). If the week is genuinely boring, the model is allowed to say "nothing moved enough to matter this week — see you next Monday." Do not fake urgency.Shorter variant — monthly recap version
Same inputs as above, but this is a monthly takeaway, not weekly. Write 6 lines, not 4.
Add: Line 5 = the one decision we made this month based on this data. Line 6 = the metric we should watch in the coming month.
Keep total under 100 words. Same rule: do not fake urgency.Sample output
A strong weekly takeaway: “Activation: 16% (was 12%) — biggest weekly jump since June. Cause: A/B variant B rolled to 100% on Tuesday. Surprising: weekend signups doubled, mostly mobile. Action: should we revive the mobile-onboarding cleanup we paused in Q2?”
A useful boring-week takeaway: “Nothing moved enough to matter this week — activation, retention, and signups all within last-month noise. Surprise: support tickets dropped 30% (probably holiday week). Next: revisit Monday with the launch numbers.” — much better than fake urgency.
A useful anomaly-week takeaway: “Refund rate: 4.1% (was 1.8%) — 2.3x last week, biggest single-week jump in a year. Cause: investigating; correlates with the v2.3 release. Surprising: refunds concentrated in the EU plan, not US. Action: pause v2.3 EU rollout until we have a read.” — leads with the alarming number, names the suspected cause without inventing one, surfaces the segment detail, gives a specific action.
A useful single-question version: “Q being asked: ‘is the new pricing hurting trial-to-paid?’ This week: trial-to-paid 18% (was 17%) — within noise. No signal yet. Surprising: enterprise trials up 40% on the new pricing page. Action: hold judgment until end of month.”
How to refine
- If the takeaway is dry / “nothing happened”: “Line 3 must make a reader say ‘huh, really?’ even on a boring week. Surface a small surprise — a segment, a hour-of-day pattern, a cohort. Surprise is what earns the read.”
- If lines feel padded: “Shorten each line to under 12 words. If a line needs more, the line is doing two jobs and should be split or one job dropped.”
- If you keep crying wolf: “If the move is within last-month noise, do not call it significant. Use ‘within noise’ or ‘too early to tell.’ Save ‘significant’ for moves that are actually outside noise.”
- If actions are vague: “Replace ‘let’s discuss’ with a specific question to a specific person (‘@dave — can you check the refund-by-region split before Wed?’). Vague actions get ignored.”
- If the headline buries the lead: “Line 1 must lead with the number. ‘Activation jumped 4pp’ beats ‘Activation has shown notable movement this week.’”
Common mistakes
- “Lots happened this week” framing: content-free; kills engagement. Either name what moved, or name that nothing moved.
- Burying the action in line 4: Slack readers do not scroll; if the action is the most important line, it should be line 1 in extreme cases (alarming spike) or line 4 in normal weeks.
- Crying wolf weekly: “alarming,” “significant,” “concerning” used every week teaches the team to mute the channel; reserve urgency for actually-urgent weeks.
- Posting the dashboard link without the takeaway: readers click less than 10% of the time; the takeaway is the dashboard’s billboard.
- Including too many metrics: 4 lines on 4 metrics is no signal; 4 lines on 1 metric (with one surprise from another) is signal.
- Letting AI invent a cause: “due to seasonal trends” when you have no data on seasonality is fabrication; the model defaults to inventing a cause when none is named. Force “no clear cause yet” as an allowed option.
- No continuity with last week’s takeaway: readers reading the second post should see the thread; “last week’s open question — answered” earns reads more than fresh headlines.
- Same takeaway across audiences: leadership wants decisions; eng wants causes; marketing wants levers; one post serves none if you do not pick.
FAQ
- How often should I post takeaways?: Weekly works for most teams. Monthly is fine for stable products. Daily takeaways train people to ignore the channel — the noise drowns the signal.
- Pin the message or use a thread?: Pin the latest weekly takeaway, archive the previous one (or move it to a #data-archive channel). Threads work for active discussion, not for headlines.
- Should the takeaway include a chart screenshot?: One small chart per takeaway, only if it adds information beyond the number. Multiple charts in a Slack message are noise.
- What if the team disagrees with my “why” interpretation?: Good — the takeaway opened the conversation, which is the entire goal. Be willing to update next week’s takeaway with the corrected interpretation, in public.
- How long until the team actually reads it?: Three to four weeks. The first one feels ignored; by week 4, people are using “did you see the takeaway” as shorthand. Stick with the format and a consistent post time (Monday 10am works for most teams).