Campaign post-mortem: 8 sections of a marketing autopsy (with a Claude template you can re-use)

It was 11:47 pm on a Thursday when the Slack thread started. Our Q4 webinar campaign had wrapped three hours earlier and we were staring at a 2.1% conversion rate against a 4% target. Someone posted a Pepe-sad-frog meme. Someone else wrote "we need to talk about this tomorrow." Then someone — a senior PM who had been with the company longer than most of us — typed: "Can we not do the post-mortem thing this time? It always goes the same way."

I've run or sat through roughly 200 campaign retrospectives over 15 years. The senior PM was right about the "always goes the same way" part. Most post-mortems are a 30-minute Slack call where the loudest person blames the channel that underperformed, someone says "creative was fine, the brief was the problem," and the doc gets archived in a Notion folder no one opens again. Nothing changes. The next campaign repeats the same mistake with a new name and a fresh budget.

The reason post-mortems fail isn't effort. It's structure. Without a fixed skeleton, you end up relitigating opinions instead of building a learning loop. So a few years ago I standardized the team's post-mortem around eight sections — a marketing autopsy, not a vibe session. Recently I built the whole thing as a Claude prompt that spits out a fully drafted retrospective from your raw campaign data in about 90 seconds. The doc you ship to the team is 80% done before the meeting starts. The meeting itself is for arguing about the 20% that actually matters.

Below is the framework, the template, and the three things this approach won't catch.

The 8 sections

The order matters. Section 1 anchors the campaign to its original intent. Section 8 is the only one anyone will read six months from now. Don't reorder. Don't merge.

1. The brief, in one paragraph

Paste the original brief. If there wasn't one, paste the Slack message or ticket where the campaign was approved. The point isn't bureaucracy — it's locking in what the team thought it was doing. Without this anchor, the post-mortem drifts into a debate about what success "should have" meant.

Two things to flag explicitly:

• The KPI hierarchy. Was this campaign optimizing for pipeline (sales pipeline, 销售管线), brand lift, or raw lead volume? They trade off against each other and you can't judge creative without knowing which one was the boss.
• The assumptions list. "We assumed our email list was warm enough to convert at 3%." "We assumed LinkedIn CPM (Cost Per Mille, 千次展示成本) had stabilized post-iOS 14.5." If the assumption was wrong, that's not a failure — that's the lesson.

2. What actually happened

A short, factual timeline. The campaign went live on X, the creative shipped on Y, we paused spend on Z because of W. No interpretation. No "and then things went south." A junior should be able to read this section and understand the sequence without any context.

The discipline of writing facts first is what kills the blame game later. You can't argue about whether something "felt off" if the timeline clearly shows the budget was cut in half on day 6.

3. The numbers, in a single table

A grid of: metric → target → actual → variance. Keep it to 6-10 rows. If you have 30 metrics, you don't know which ones mattered.

Include both leading and lagging indicators. A webinar campaign's leading indicators are registration rate and show-up rate. The lagging is pipeline created and closed-won revenue 60-90 days out. The honest version of a post-mortem always says: "we don't know the lagging number yet, we're tracking it." That's a real answer.

4. The audience that actually showed up

This is where the framework diverges from the typical post-mortem. Most teams write "the audience was our ICP (Ideal Customer Profile, 理想客户画像)" and move on. The real question: who did convert, and how is that different from who we targeted?

Pull a 2-column comparison. Target: VP Marketing, mid-market SaaS, 50-500 employees. Actual: VP Marketing, 50-200 employees, US and UK only, with a strong skew toward companies that had downloaded our pricing page in the last 30 days. The second column is almost always more useful than the first. It's the seed of the next campaign's targeting brief.

5. Creative autopsy

Three to five specific assets. Name them. For each: hook, format, channel, top-line metric, and one sentence on why it worked or didn't. Avoid the word "creatively" — it's a non-observation.

The cheapest learning in any campaign is in the copy that lost. A 0.4% CTR (Click-Through Rate, 点击率) headline isn't a failure, it's a free AB test (split test, 对照实验) result you didn't have to set up. Document it.

6. Channel forensics

Where did the money go, and what did each dollar buy? The same campaign often has wildly different CAC (Customer Acquisition Cost, 获客成本) by channel — one of my Q3 campaigns had $42 CAC on LinkedIn and $310 on a sponsored newsletter placement. Without the breakdown, the next budget meeting defaults to "let's shift more to LinkedIn" and you lose the context that the LinkedIn number only worked because of a one-time promo offer.

Include the cheap channels too, especially organic search and email. They don't have a "spend" line, but they have an attribution claim worth defending.

7. Friction log

The things that almost killed the campaign. The landing page went down for 4 hours on day 2. The legal review took 9 business days when we budgeted 5. The CRM (Customer Relationship Management, 客户关系管理) sync dropped 18% of leads for 48 hours. The webinar link in the confirmation email was broken for the first 200 registrants.

Friction is the section that turns a post-mortem into an operational document. Most teams under-report it because the friction was someone else's fault. Force yourself to write it down anyway. The pattern across three campaigns is the pattern that needs a process fix, not a campaign fix.

8. Carry-forward

This is the only section with bullet points, and there are exactly three: do more of, do less of, do differently. One sentence each. Everything in sections 1-7 should be compressible into these three lines. If it isn't, you don't know the lesson yet.

The carry-forward is also where you commit to a specific change for the next campaign. Not "we should test more creative" — "for the next webinar, we'll test two subject lines A/B style, owned by [name], measured on [metric]." Named owner, named metric. Otherwise this section is a wish list.

The Claude template

I built this in about 20 minutes and have used it for 6 campaigns now. Copy it, replace the bracketed inputs, and run it on Claude Sonnet 4.5 or Opus 4.7 with web search off:

You are a senior marketing analyst writing a campaign post-mortem.
Tone: direct, no fluff, no motivational language. Write as if to a
peer who was on the campaign and doesn't want to read 10 pages.

INPUTS (paste below):
- Campaign name: [name]
- Original brief: [paste 1-2 paragraphs]
- Campaign dates: [start - end]
- Budget: [total + channel split if available]
- Target KPI: [primary metric and target value]
- Raw data dump: [paste metrics, audience breakdown, channel performance,
  any notes from the team chat]

OUTPUT: Write the 8 sections in this exact order, using these headings:
1. Brief (1 paragraph)
2. Timeline (factual, dated)
3. Numbers table (markdown table, 6-10 rows)
4. Audience delta (target vs actual, 2 columns)
5. Creative autopsy (3-5 assets)
6. Channel forensics (spend vs CAC by channel)
7. Friction log (bullet list)
8. Carry-forward (3 bullets: do more, do less, do differently)

Rules:
- Quote real numbers from the inputs. Do not invent metrics.
- If a number is missing, say "not measured" — do not estimate.
- The carry-forward must name a specific change with an owner.
- Total length: 600-900 words.

Two design choices worth flagging. First, the prompt tells the model to say "not measured" instead of estimating. This is the single most important instruction — it stops Claude from filling gaps with plausible-sounding numbers, which is the fastest way to make a post-mortem useless. Second, the carry-forward constraint forces a named owner. A lesson without an owner is a feeling, not a plan.

You'll still need to edit the output. The Claude draft typically nails sections 2, 3, and 6, and is weakest on section 5 (creative autopsy) because the model can't read the ad image the way your team can. Budget 15-20 minutes to rewrite the creative section with human eyes. That's a much better use of meeting time than spending 90 minutes arguing about what the campaign was supposed to do.

What this framework won't catch

Post-mortems are a learning tool, not a prediction tool. Three things the eight sections can't do for you:

They can't tell you about campaigns you didn't run. The biggest miss I've had in the last two years was not running a competitor-conquesting paid search campaign because "we never do that." The post-mortem on the campaign I did run told me nothing about the one I avoided. The lesson is structural: post-mortems are a closed-loop on the past. For the gaps in your strategy, you need a separate process — a pre-mortem, a competitor teardown, or just a frank conversation with someone who runs marketing at a competitor.

They can't replace decision quality at the brief stage. I've seen teams run flawless post-mortems on campaigns that were doomed from approval day. The retrospective reveals what happened. It doesn't reveal whether the original bet was even worth making. If your post-mortems keep producing "the brief was wrong" as a recurring carry-forward, the problem isn't the post-mortem — it's the approval process.

They bias toward measurable tactics. A campaign that moves brand sentiment or category narrative will look mediocre in every section above. That doesn't mean it failed — it means the post-mortem framework isn't the right evaluation tool for brand work. Be honest about which campaigns this works for (performance, demand gen, lead-based) and which it doesn't (brand launches, category creation, awareness-only flights).

The actual point

The senior PM who wanted to skip the post-mortem wasn't wrong that most of them are pointless. He was wrong that the answer was to skip them. The answer is to make them cheap enough that doing them properly costs less than the cost of repeating the same mistake.

A 90-second Claude draft plus a 30-minute team review is a quarter of the time of the old process. The team actually reads the carry-forward now because it's three lines. And six months in, when we're planning Q3 and someone says "didn't we learn this already?", the answer is in a doc that took 15 minutes to write.

The post-mortem isn't the lesson. The post-mortem is the receipt for the lesson. Build the receipt, and the learning becomes free.