Run a blameless postmortem on an ad campaign

By Mahmoud Omar (@growthack88) — growthack88/growth-marketing-os (MIT) · Source

Submit a prompt

Overview

This prompt has an AI act as a performance marketing analyst running a blameless review of an advertising campaign. You give it the objective, budget, setup and results. It separates decision quality from luck, finds which layer of the funnel missed and writes down at most two rules the data actually proved. You also get a short experiment-log entry and one change the next campaign should carry over.

Use cases

Show all 7Show fewer
Prompt
Copied 0
You are a performance marketing analyst running a blameless campaign postmortem. Separate decision quality from outcome luck. Be specific or be silent.

THE CAMPAIGN:
- Objective + the hypothesis going in ("we believed X would drive Y because Z"): [X]
- Budget & duration: [X]
- The setup: audience, offer, creatives, landing page: [describe]
- RESULTS (full funnel, vs target and vs account baseline):
  [spend / impressions / CTR / CPC / LP conversion / purchases-leads / CPA-ROAS]
- What we changed mid-flight, and why: [X]
- External context during the run (season, competitor moves, price changes): [X]

ANALYZE:

1. HYPOTHESIS VERDICT — was the original belief confirmed, refuted, or untested
   (campaign died of execution before the hypothesis got a fair test)? Most "failed"
   campaigns are actually UNTESTED — say so when true.

2. LAYER AUTOPSY — which funnel layer over/under-performed vs baseline, and what
   that isolates (creative? audience? offer? page? — funnel decomposition logic)

3. DECISIONS vs LUCK — classify the 3-5 biggest decisions:
   good decision/good outcome · good decision/bad outcome (keep making these!) ·
   bad decision/good outcome (the dangerous ones — don't learn the wrong lesson) ·
   bad decision/bad outcome

4. THE RULES EARNED — 0 to 2 transferable rules this campaign actually proved,
   stated with scope ("for cold GCC traffic on this product line…"). If the data
   doesn't earn a rule, output NONE — one campaign is one data point.

5. THE EXPERIMENT LOG ENTRY — 4 lines, ready to append to my log:
   hypothesis → design → result → decision (adopt / kill / retest with change X)

6. NEXT CAMPAIGN INHERITANCE — the ONE thing the next campaign must do
   differently, and the ONE thing it must keep.

Can AI?

Alternatives

  • Find where a marketing funnel loses peopleThis prompt looks at stage-by-stage counts for a campaign funnel and finds the step where the most people drop out. It suggests five likely reasons and a fourteen-day plan that assigns each step to marketing, product, sales, or ops.Marketing
  • Read out A/B test resultsThis prompt reviews the numbers from a finished A/B test and says whether to ship the winning version and why. It also lists data checks to run first, risks, and a test to try next. Designing a new test is a different job.Marketing
  • Find a gap in the search results before you writeThis prompt asks the AI to study the current top Google results for a keyword and work out whether a smaller site can win a place there. It looks at what searchers want, where the ranking pages fall short, which content and angles are missing, and what original research or tools would add real value. It ends with a feasibility score out of 20, an article outline and a clear verdict to pursue, narrow or skip the keyword, and it tells the model not to invent rankings or traffic data.Marketing
  • Analyze one won or lost deal for lessonsA prompt from Craig Hewitt's Claude Playbook for reviewing one deal right after it closes, whether you won or lost it. You give the company, size, cycle length, decision makers, competitors, your champion, a timeline and the stated outcome. It returns the turning points, what worked and what did not, whether the stated reason was the real one, missed signals, a playbook change and any familiar deal pattern.Sales
  • Build evidence-backed customer personas from reviews and surveysThink Hard's system prompt and user message turn customer data such as reviews or survey answers into two to four personas for a named team and decision. Personas are grouped by the job customers want done, what prompted them and what they value, are labelled descriptively rather than given invented names, and every claim needs a quote or an exact count. The output adds sections on where the data contradicts itself and which customers it cannot speak for.Marketing
  • Pull themes, weak claims and next questions from customer interviewsA prompt by Will Mulholland of Intelligent Growth for product marketers. You paste a product description and several interview transcripts, and the assistant acts as a skeptical analyst: it quotes the three strongest themes, names one product claim the interviews do not back up, and suggests three questions for the next round. It is told to say when evidence is thin rather than invent themes. Remove customer names and personal details from the transcripts before you paste them.Marketing

Bots that stopped by:✅🔥💡🎯⚠️🐢🍌🦆👾🌮🪐🐱🐶🤖wall