Two years ago, I was managing three concurrent projects. All three ran over budget. Two missed their deadlines. One got cancelled.
I blamed external factors at the time: changing requirements, staffing issues, unclear leadership direction.
Last month, I fed all three post-mortems into an AI and asked it to find the patterns. What it found was uncomfortable.
The Experiment
I had post-mortems from three projects that went sideways. Each was 2-4 pages of honest documentation.
I gave them to an AI with this prompt:
Here are three project post-mortems from the same project manager.
Read them and identify:
1. Patterns that appear across all three failures
2. Root causes vs. symptoms
3. Early warning signals that appeared in all three
4. Be honest. I want genuine insight, not validation.
What It Found
Pattern 1: I was confusing status with progress.
All three post-mortems mentioned 'good communication' and 'regular check-ins.' The AI noted that while I had excellent visibility into WHERE the projects were, I consistently lacked early clarity on whether the PACE was sustainable. Lots of green status updates that turned red in weeks 8-9.
I thought I was managing. I was reporting.
Pattern 2: Scope was agreed but never defined.
In all three cases, stakeholders had signed off on scope. But the documentation left room for interpretation — phrases like 'a basic version of X' or 'initial integration with Y.' When reality hit, everyone had a different mental model of what 'basic' meant.
I knew scope creep was a risk. I didn't realize I had built ambiguity into the scope itself.
Pattern 3: The first 4 weeks set the ceiling.
The AI pointed out that decisions made in weeks 1-4 constrained what was possible for the rest of the timeline. Architecture choices, stakeholder alignment decisions, and resource allocations made early became increasingly hard to reverse.
'Your post-mortems consistently describe problems in weeks 8-12 that were seeded in weeks 1-4.'
What I Changed
After this analysis, I made three concrete changes:
1. Replace status reports with pace reports. Instead of 'here's where we are,' I now include 'here's whether we're on trajectory to hit the deadline at current pace.'
2. Define scope with examples of what's NOT included. For every major scope item, I document one or two things that might seem included but aren't. Explicit exclusions prevent the most common scope debates.
3. The Week 4 review. I now run a mandatory retrospective at week 4, specifically asking: what decisions made so far will be hardest to undo? Are we comfortable with those?
The Uncomfortable Part
What made this exercise valuable was that AI doesn't have a stake in protecting your ego. When you debrief with colleagues, there's social friction around saying 'you consistently made the same mistake.' With AI, there's no friction. It just says the thing.
If you're interested in using AI as a genuine thought partner in project management — not just a writing assistant — I write a free weekly newsletter about exactly this:
Ahead of Schedule: practical AI habits for project managers. One thing you can use this week, every week.
Subscribe free: https://buttondown.com/marcustillerman
Top comments (0)