Grading PMs without grading the grader.

Police take witness statements separately. Nobody has to be lying for it to matter. The second person to talk always hears the first, and from then on they're agreeing with that story or pushing against it. Either way you've lost what they actually saw.

Most PM reviews skip that step. The manager writes the story, defends it in a calibration room, and the PM hears it at the end. By then it's a verdict. The PM spends the hour defending instead of reflecting, and you end up measuring how well someone argues. That's a skill. The job is something else.

When I built the evaluation framework for a product and design org, I started from a different question: how do you get two honest reads before either person has seen the other's?

One matrix, value from start to finish

The foundation was a grid, four rows by five columns.

The rows were the four things a PM does with value, in the order value lives: identifying it, prioritizing it, creating it, and realizing it once it's live. The columns were five skills a PM carries from role to role: guiding, strategy, vision, ownership and insight. Every cell described what that capability looks like through that skill.

A couple of cells, to make it concrete. Ownership crossed with creation came out as "own the outcome": you're accountable for what ships, you own the result and not the activity, and you can't hand that to design or engineering. Vision crossed with realization meant resisting new work until what you already shipped has delivered, and making the cost of underused product as clear as the cost of not building.

The cell descriptions were behavioral anchors and conversation guides, and nobody rated them directly. That mattered. If the words in a cell map straight to a score, people reverse-engineer the score. Break that link and the description goes back to doing its job, which is helping two people talk about the same behavior.

Two instruments, and only one counts for pay

Same twenty cells, written two ways.

The capabilities version was about the current role. The rater's question: given what my role asked of me, what did I deliver, and was this capability even available to me? It rated impact and opportunity.

The skills version was portable. The question there: what do I actually have in my toolkit, and how far can I take it? It rated ability and potential.

Opportunity is the one people forget. If the role never asked someone to do the work, a low score says more about the role than the person. So opportunity ran from minimal to full, as a description of scope, and impact was always read against it.

Then the line I wouldn't move: calibration, compensation and promotion were based on impact only. Potential, ability and opportunity existed for development and growth planning, nothing else. Mix them into pay and people start managing their potential score instead of telling you the truth. Keeping that line consistent protected the instrument, and it protected the trust of the people filling it out.

Blind first, then the reveal

It ran in six stages.

First, the PM filled out a self-assessment workbook. Second, the manager filled out the same instrument separately. Neither looked at the other's until both were done. That part was non-negotiable, because a self-assessment written after you've seen your manager's is just a reply. Third, feedback came in from partners and coworkers.

Fourth was the reveal. We put the two reads side by side as radar charts, one shape laid over the other. Where they overlapped, we moved fast: you both agree, so there's nothing to litigate. Where they didn't, that was the conversation. The manager didn't deliver a verdict. Both people explored the gap together.

The gaps point two ways.

When a PM rates themselves lower than the manager does, it's usually a confidence problem rather than a skill problem. This happens more than people think. Some of your best PMs are quietly underselling themselves, and a traditional review never notices, because the manager's higher rating just goes in the file.

When a PM rates themselves higher, it's a visibility problem. Either there's work the manager isn't seeing, or there's a bar nobody ever made explicit. Both are fixable. Neither shows up when the review starts with the manager's story.

A big gap is information. Treat it that way and the conversation stays a conversation. Don't average it away, either. Low-self, high-manager and the reverse are two different talks, and a blended number hides both.

Fifth came the final review and rating. Sixth, a standard individual development plan for PMs and product designers, written against the same instruments. So the review ended in a plan, in the same language the review used, which the next review could check.

What we found

After running it, the thing that stood out was the gap between what people felt their aptitude was and what they were able to display.

That started a different kind of conversation: are we putting people in the best position to succeed? Are we giving them chances to shine? Some people were being given work they weren't great at. The question became whether we needed them to be great at it, or whether we should give them more of the work where they're already strong, where their superpower could really help the rest of the organization.

Tailoring the work and the goals that way also let people set their own path toward where their growth was aimed. ("AI should be a supersuit, not a severance package" picks up the superpower idea.)

Ranking is easy

Ranking is easy. Line people up, draw a curve, done by lunch. Development is hard. It means knowing, specifically, what each person should work on next and why.

If your evaluation system can't tell a PM what to work on next quarter, in words they'd use themselves, it's only doing half the job. Probably the less useful half.

That's the test I'd give any founder bringing in a first product leader. Ask how they'll know whether the PMs are good. If the answer is "I'll know," keep looking.