Nick QuirkMost agencies celebrate wins in Slack and quietly bury losses in an archived folder. That asymmetry is expensive. The campaign that returned 11% of its traffic forecast contains more usable information than the three that hit target, because failure exposes the seams in your process — the assumption nobody tested, the approval that took 19 days, the developer who was never actually booked.
A project autopsy is the structured way to extract that information before it evaporates. Done properly it takes 90 minutes, produces three or four concrete system changes, and pays for itself the next time a similar project comes through the door.
What a project autopsy actually is
A project autopsy is a single-project, evidence-first review of a campaign that missed its stated outcome. You reconstruct the timeline from artifacts — briefs, time logs, Slack threads, ranking data, invoices — and identify the decisions that produced the gap between plan and result.
Three things distinguish it from a normal retro:
- It's triggered by an outcome, not a calendar. You run it because something specific went wrong.
- It's evidence-led. Nobody speaks from memory until the timeline is on the screen.
- It ends in system changes, not sentiments. "Communicate better" is not an output. "Content projects now require a named client approver with a 5-business-day SLA in the SOW" is.
When to trigger one
You cannot autopsy everything. Set thresholds and stick to them. Reasonable triggers for a mid-sized agency:
- The project missed its primary KPI by more than 40%
- Actual hours exceeded budgeted hours by 25% or more
- A client churned within 90 days of a project delivering
- You performed more than 8 hours of unbilled rework
- A deliverable shipped more than three weeks late
At those thresholds most agencies land on two to four autopsies a year. That's the right volume — enough to build a pattern library, few enough that people take them seriously.
Gather evidence before anyone opens their mouth
The single biggest failure mode is holding the meeting first and looking at data second. By then the story is already set, and the story is always some version of "the client was difficult." Sometimes true, rarely the whole cause.
The person running the autopsy — ideally not the project lead — assembles a pack in advance. It should take two to three hours and contain:
- The original scope and success metric, exactly as written. Not as remembered. If the SOW said "improve organic visibility" with no number attached, that's finding number one.
- The planned vs. actual timeline. Phase start and end dates side by side.
- Budgeted vs. logged hours by role and phase. This is where the truth usually lives. A technical audit budgeted at 18 hours that consumed 47 tells you the discovery step was wrong, not the execution.
- The decision log. Every scope change, every "quick favour," every pivot, with a date and who asked.
- Outcome data. Rankings, traffic, conversions, pipeline — whatever the metric was, charted against the forecast.
- Client-side latency. How long approvals actually took, measured. Not "they were slow."
If assembling this takes you two days of digging through email, you've already learned something. Agencies that run project phases and time tracking in one system can pull the first four items in twenty minutes; those juggling a task board, a spreadsheet, and a separate timer usually cannot. That's part of why we built PeakKR around agency project structures rather than generic task lists — the autopsy pack should be a byproduct of doing the work, not an archaeology project.
How to run the 90-minute session
Invite everyone who touched the project — strategist, SEO lead, writer, developer, account manager — plus one person with no involvement to ask naive questions. Cap it at seven people.
Minutes 0–10: rules and framing
State the two rules out loud: we critique decisions and systems, never people, and we judge decisions on the information available at the time. Then the project lead names their own biggest mistake first. This single move does more to defuse defensiveness than any amount of psychological-safety language.
Minutes 10–30: build the timeline together
Walk the evidence pack chronologically. No interpretation yet — just "on March 4 the brief was approved, on March 19 the first draft went to the client, on April 8 the client responded." Watch for the moments where people say "wait, I didn't know that happened." Those gaps are findings.
Minutes 30–60: find the decision points
Identify the three to five moments where a different choice would have changed the outcome. For each one ask:
- What information did we have?
- What information did we assume?
- What would have made the missing information visible?
Here's a real pattern. An agency forecast a 60% traffic lift from a 40-page content refresh for a SaaS client. Result after five months: 8%. The autopsy timeline showed the keyword research was done in week one, the client's product pivoted in week three, and nobody revisited the target list. The decision point wasn't the research — it was the absence of a checkpoint between phases. The fix was a 30-minute strategy re-validation gate at the end of every discovery phase. Cost: half an hour. Value: the next quarter's roadmap didn't ship against stale assumptions.
Minutes 60–80: separate causes into three buckets
Sort every finding into system (our process allowed this), skill (we didn't know how), or situation (genuinely outside our control). Be honest about the third bucket but suspicious of it. If more than a third of your findings are "situational," the group is still protecting itself.
Minutes 80–90: commit to changes
Pick a maximum of three changes. Each needs an owner, a date, and a location — the template, the SOW, the onboarding checklist, the automation. A change that lives only in the meeting notes will not happen.
The five failure patterns you'll keep finding
After a dozen autopsies, the same causes surface across agencies:
1. The unfalsifiable goal. "Increase brand awareness," "improve SEO health." If the goal can't be missed, it can't be hit either. Every project needs one number and a date.
2. Undated dependencies. The content calendar assumed developer availability that was never booked. Cross-team work fails at the handoff, not the task — which is why managing cross-functional SEO, content and dev teams needs explicit ownership at every boundary.
3. Accreted scope. No single request was unreasonable; the eleventh one broke the budget. If your autopsies keep landing here, the fix is upstream in how you handle scope creep in marketing projects — logging and pricing changes at request time, not at invoice time.
4. Client-side latency treated as noise. A 14-day average approval cycle on a 10-week project consumes 20% of the timeline. If you didn't budget it, you didn't plan the project — you planned your half of it.
5. Progress reported as activity. Twelve of fifteen tasks complete tells you nothing about whether the strategy is working. Teams that track how uncertain work actually is — see hill charts for marketing projects — spot stalls in week three instead of week nine.
Closing the loop so it isn't theatre
An autopsy that produces no durable change is worse than none — it teaches the team that honesty is performative. Three habits prevent that:
Write the finding into a template, not a document. If the lesson is "validate CMS constraints before scoping technical fixes," add it as a required field in your SEO audit project structure. The next project inherits the lesson automatically.
Review commitments at 30 days. Five minutes in a leadership meeting. Owner, change, done or not.
Keep a one-page pattern log. One line per autopsy: project, primary cause, change made. After six entries you'll see your agency's actual weakness — and it's almost never what you'd have guessed.
Share a sanitised version with the client where it's appropriate. "We reviewed why this underperformed and here's what we changed" rebuilds more trust than a discount does. Roughly a third of the time, the client volunteers their own contribution to the problem, which is the fastest route to a better working relationship on the next engagement.
Project autopsy checklist
- Define your trigger thresholds in writing — KPI miss, hours overr
