| name | pm-metrics-critic |
| description | Critique a metrics dashboard, success-criteria section, or proposed North Star metric against the repo's metrics guide. Use when the user shares a dashboard, a list of KPIs, a success-criteria slide, or asks "are these the right metrics?" Surfaces the three most common metric failures: aggregate metrics that hide segment-level failure, vanity metrics decoupled from strategy assumptions, and standalone metrics with no falsifiable counter. Returns a metric-by-metric verdict and a recommended replacement set. |
PM metrics critic
Reviews a set of metrics โ a dashboard, a success-criteria section, a North Star proposal โ against decision-making/metrics.md. The core test: does each metric, if it moved, tell us something strategic, or does it just tell us the product is being used? Most metric sets fail this test and the skill's job is to say so concretely.
When to use
Use this when:
- The user shares a dashboard or KPI list and asks "are these the right metrics?"
- The user is locking success criteria for a launch or a quarterly OKR
- The user is proposing or revisiting a North Star metric
- A metric "moved" and the team is debating whether the underlying assumption was actually validated
Don't use this when:
- The user wants a full PRD critique โ use
pm-prd-drafter
- The user wants to build a metric tree from scratch (mapping inputs โ outputs โ outcomes) โ that's a longer exercise; reference
decision-making/metrics.md ยง metric trees and walk it manually
- The user is debugging a specific dashboard query or instrumentation bug โ that's an engineering task
How to apply
-
Pull the strategy assumptions. Reference decision-making/metrics.md ยง "Match metrics to strategy assumptions." For the work in question, list the load-bearing strategic assumptions. If the user can't articulate them, surface that gap before evaluating any metric โ you can't grade metrics without strategy.
-
For each proposed metric, ask: what assumption would this falsify? If a metric can't falsify any assumption โ if it can only confirm "the product is being used" โ it's a vanity metric. Mark it as such.
-
Run the segment-level decomposition. Aggregate metrics hide segment failures. If the strategy assumption is segment-specific ("top creators will adopt"), the metric must be segment-specific. Top-line DAU does not validate a top-creator assumption.
-
Run the standalone-vs-counterweight check. From decision-making/metrics.md ยง counter-metrics: every primary metric should have a counter-metric that catches the obvious gaming path. Engagement up + retention down is a different story than engagement up + retention flat. Flag any metric that ships without a counter.
-
Run the leading-vs-lagging check. A success criterion of "ARR by Q4" tells you nothing in weeks 1โ8. Identify the leading indicator that would predict the lagging metric early enough to act on it.
-
Recommend a replacement set. Not "track more metrics." A focused list of 3โ5 metrics that each map to a strategic assumption, with a counter-metric for each.
Output structure
## TL;DR
[One-paragraph honest take. Are these metrics measuring the strategy or the activity?]
## Strategic assumptions in play
1. [Assumption 1 โ load-bearing for this work]
2. [Assumption 2]
3. ...
## Metric-by-metric
| Metric | What it would falsify | Verdict |
|---|---|---|
| [Metric A] | [Assumption / nothing โ vanity] | ๐ข strategic / ๐ก weak / ๐ด vanity |
| [Metric B] | ... | ... |
## Failure modes flagged
- **Aggregate masking segment failure:** [Which metric, what segment is hidden]
- **No counter-metric:** [Which metric, what the obvious gaming path looks like]
- **Lagging-only:** [Which metric, what leading indicator should sit alongside it]
- **Decoupled from strategy:** [Which metric, what assumption it claims to validate but doesn't]
## Recommended replacement set
1. **[Primary metric]** โ falsifies *[assumption]*. Segment: [specific]. Threshold: [specific]. Counter-metric: [specific].
2. **[Primary metric 2]** โ ...
3. ...
## What an exec will press on
[The 1-2 metrics questions the team will be asked that they're not yet ready for.]
What good looks like
- Every recommended metric is named with a segment and a threshold. "Adoption" is not a metric. "Adoption among top-decile creators reaching 25% in 60 days" is.
- Each metric has a counter-metric. No exceptions.
- The North Star metric (if one is proposed) is a single metric the entire org could agree to optimize, not a basket of five things.
- Vanity metrics are called out explicitly. "DAU is a vanity metric for this strategy because the assumption is about which users adopt, not how many" โ that level of specificity.
- Leading indicators are named. The team should not have to wait until Q4 to know if Q4 is on track.
Anti-patterns
- Approving a dashboard because it has "good metrics on it" without checking what they would falsify.
- Recommending more metrics. Metric proliferation is itself a failure mode โ most dashboards need fewer, sharper metrics, not more.
- Accepting "engagement" as a metric. Engagement is a category. Engagement-with-what, by-whom, doing-what, is the metric.
- Letting a North Star metric ship without naming the strategic bet it represents. If the team can't say "if North Star moves, it means [specific strategic claim] is true," the North Star isn't load-bearing.
- Ignoring the counter-metric requirement because the team "would never game it." Gaming is rarely intentional; it's structural. The counter-metric exists to catch the structural drift.