Skip to main content

Metrics Design

Metric selection determines what the organization optimizes. Selecting an inappropriate metric leads to optimizing the wrong outcome.

Metric Quality Criteria

A useful metric must satisfy five requirements:

CriterionQuestionFailure Example
MeasurableCan it be computed from available data?"User delight" without operationalization
MovableCan product changes affect it?"Total users ever" (only increases)
UnambiguousDoes everyone agree on the definition?"Active users" without defined actions
TimelyIs it available quickly enough for decisions?Customer lifetime value (requires years)
RobustIs it resistant to gaming?"Time on page" (can be inflated with loading delays)

Metric Categories

North Star Metric

The single metric that captures whether the product delivers value to users.

CompanyNorth Star MetricRationale
NetflixHours watchedIndicates value delivered
AirbnbNights bookedRepresents completed transactions
SlackMessages sentMeasures active usage
UberRides completedCore value delivery
SpotifyListening timeEngagement depth

The north star metric provides organizational alignment. When priorities conflict, teams evaluate which approach has greater impact on the north star.

Primary Metrics

For specific experiments or features, one metric determines success.

ContextPrimary Metric
Checkout flow redesignConversion rate
Onboarding experimentTime to first action
Pricing testRevenue per user

Multiple primary metrics create ambiguity when results conflict.

Guardrail Metrics

Metrics that should not degrade while optimizing the primary metric.

CategoryExample Guardrails
PerformanceLoad time, error rate
RevenueRevenue (when optimizing engagement)
User satisfactionSupport tickets, complaint rate
RetentionReturn rate, churn

A 10% conversion lift that doubles page load time may represent a net negative.

Counter Metrics

Metrics that specifically prevent gaming of the primary metric.

Primary MetricCounter MetricRationale
Engagement (time spent)Content quality, satisfactionPrevents low-quality addictive content
RevenueChurn rate, complaintsPrevents aggressive monetization
Conversion rateAverage order value, return ratePrevents misleading conversions
Click-through rateActual conversions, bounce ratePrevents clickbait

HEART Framework

Google's framework for product metrics:

CategoryMeasurement FocusExamples
HappinessUser satisfactionNPS, survey scores, app ratings
EngagementUsage intensityDAU/MAU ratio, session length, actions per session
AdoptionNew user uptakeSign-ups, feature activation, first-time use
RetentionUser persistenceD7/D30 retention, churn rate, repeat usage
Task SuccessCompletion efficiencyCompletion rate, time to complete, error rate

Not all categories apply to every feature. Select relevant categories based on the feature's purpose.

Common Interview Topics

Feature Success Measurement

Framework:

  1. Identify the feature goal
  2. Select primary metric (directly measures goal)
  3. Define guardrail metrics (should not degrade)
  4. Specify data sources and calculation methods

New Product Metrics

Apply HEART framework or similar structured approach. Demonstrate consideration of multiple measurement dimensions, not just a single number.

Metric Gaming Prevention

Use counter metrics, diverse metric sets, and qualitative verification. Reference Goodhart's Law: "When a measure becomes a target, it ceases to be a good measure."

Leading vs Lagging Indicators

TypeCharacteristicsExample
Leading indicatorsPredict future outcomes; enable early actionEngagement (predicts revenue)
Lagging indicatorsConfirm outcomes after occurrence; higher accuracyRevenue

Both types serve different purposes: leading indicators provide rapid feedback; lagging indicators confirm actual business impact.

Metric Trade-offs

ChoiceAdvantageDisadvantage
Simple metricEasy to understand and align onMay miss nuance
Composite metricCaptures multiple factorsHarder to interpret and debug
Short-term metricFast feedback cycleMay not reflect true impact
Long-term metricAccurate outcome signalSlow learning
Proxy metricAvailable immediatelyMay not correlate with true goal
Direct metricMeasures actual outcomeMay take too long to observe

Each metric choice involves trade-offs. These trade-offs should be explicit and documented.