Your organization tracks 47 metrics. Every department has a dashboard. Monthly reports present dozens of charts. Everyone feels data-driven. Yet when it's time to make a decision, no one knows which metrics actually matter. Teams debate endlessly about what to measure, argue over metric definitions, and spend more time collecting data than using it.
The real question isn't "Can we measure this?" (you probably can measure almost anything), but "Should we measure this?" and more importantly, "Why should we measure this?" Most measurement efforts fail not from lack of data, but from measuring the wrong things for the wrong reasons.
As Douglas Hubbard argued, "If something is worth managing, it is worth measuring. If it's worth measuring, you can measure it." The challenge, then, is not capability but selection.
Good measurement starts with clarity about purpose: What decision does this metric inform? What outcome does it predict? What action becomes clearer with this data? Without clear answers, measurement becomes ritual—data collection for its own sake, creating noise instead of signal.
This guide provides a framework for deciding what deserves measurement, why certain things matter more than others, and how to design a measurement system that actually improves decisions and outcomes.
Meaningful measurement is the practice of selecting and tracking only those metrics that are demonstrably linked to decisions, outcomes, or causal theories worth testing — as opposed to measuring what is easy, available, or impressive-sounding. The criteria for what should be measured are: alignment with a specific goal, actionability (a clear response exists for changes in the metric), reliability (consistent collection is possible), predictive validity (the metric either causes or predicts an outcome that matters), and non-redundancy with other metrics already tracked. Without meeting these criteria, a metric adds noise to decision-making rather than reducing uncertainty.
Start with Goals, Not Metrics
"What gets measured gets managed." — Peter Drucker, The Practice of Management
The Backward Approach (Wrong)
Common pattern:
- List everything you can measure
- Track all of it
- Hope some of it is useful
- Drown in data without insight
Why it fails:
- No connection between metrics and decisions
- Too many metrics dilute focus
- Measures activities, not outcomes
- Leads to "metric theater" (reporting without impact)
The Forward Approach (Right)
Effective pattern:
- Define what success looks like (goals)
- Identify what drives success (drivers)
- Measure the drivers (metrics)
- Validate metrics predict success (testing)
- Act on metrics (decision-making)
Why it works:
- Every metric has clear purpose
- Limited set of vital metrics
- Measures outcomes and their drivers
- Enables action
Example: SaaS Company
Wrong approach:
- Track: signups, page views, followers, downloads, features shipped, support tickets, blog posts, email sends...
- No clear connection to goals
- 30+ metrics, none clearly actionable
Right approach:
Goal: Sustainable profitable growth
What drives sustainable growth?
- Acquire customers efficiently
- Retain them (low churn)
- Expand revenue from existing customers
What should we measure?
| Driver | Metric | Why It Matters |
|---|---|---|
| Efficient acquisition | CAC payback period | Shows months to recover acquisition cost |
| Activation | % completing first value action | Predicts retention |
| Retention | Net revenue retention (NRR) | Captures churn + expansion in one number |
| Product value | Weekly active users / Monthly actives | Shows engagement depth |
Four metrics. Each tied to goal. Each actionable.
The Measurement Hierarchy
Level 1: Outcomes (What You Care About)
Definition: The ultimate results you want to achieve
Examples:
- Revenue, profit, market share
- Customer satisfaction, retention
- Mission impact (for nonprofits)
- Health outcomes (for healthcare)
Characteristics:
- What you actually care about
- Lagging indicators (show what already happened)
- Slow to change
- Directly tied to success
Limitation: By the time outcomes change, it's often too late to adjust course
Level 2: Outputs (What You Produce)
Definition: The direct results of your activities
Examples:
- Features shipped
- Sales calls made
- Content published
- Services delivered
Characteristics:
- Under your control
- Leading indicators (happen before outcomes change)
- Faster to change
- Activities, not results
Limitation: Outputs don't guarantee outcomes (can ship features no one uses)
Level 3: Drivers (What Causes Outcomes)
Definition: The measurable factors that predict and cause outcomes
Examples:
- Conversion rates at each funnel stage
- Customer engagement scores
- Net Promoter Score (if validated)
- Time-to-value for new customers
Characteristics:
- Predictive of outcomes
- Actionable (can influence them)
- Leading indicators
- Require validation (ensure they actually predict outcomes)
This is where most meaningful measurement happens.
"In God we trust; all others bring data." — W. Edwards Deming, statistician and quality pioneer
The catch is that data without a theory of what causes what is just noise. Deming was equally insistent that measurement divorced from causal understanding produces neither insight nor improvement.
The Hierarchy in Practice
Example: E-commerce Company
| Level | Metric | Why It Matters | Limitation |
|---|---|---|---|
| Outcome | Monthly revenue | Ultimate goal | Lagging, slow to change |
| Driver | Conversion rate | Predicts revenue, faster to impact | Requires traffic |
| Driver | Average order value | Predicts revenue, faster to change | Can manipulate short-term |
| Output | Products listed | Activity, not result | Doesn't predict revenue |
| Output | Marketing emails sent | Activity | Doesn't mean engagement or sales |
Focus: Measure outcome (revenue) to know if you're succeeding. Measure drivers (conversion, AOV) to understand and influence success. Don't mistake outputs for drivers.
Criteria for Measurement
Should You Measure It? The Five Tests
Before adding a metric, it should pass all five tests:
Test 1: Aligned with Goals
Question: Does this metric relate to a goal we actually care about?
Example:
- Goal: Increase customer lifetime value
- Metric: Email open rate
- Test: Does email open rate predict or influence LTV? If not, don't prioritize it.
Red flag: Metric interesting but disconnected from strategy
Test 2: Actionable
Question: Can we take meaningful action based on changes in this metric?
Test: If metric goes up/down, what would we do differently?
Example:
| Metric | If Increases | If Decreases | Actionable? |
|---|---|---|---|
| Churn rate | Investigate quality issues, survey churned users | Document retention drivers, scale | Yes |
| Industry news mentions | Celebrate | Unclear what to do | No |
If you can't articulate clear actions for metric changes, don't measure it.
Test 3: Measurable Reliably
Question: Can we collect this data consistently and accurately?
Problems that kill reliability:
- Data not consistently available
- Requires manual collection (error-prone)
- Definition ambiguous (people measure differently)
- Measurement changes behavior (Hawthorne effect)
Example:
- Reliable: Conversion rate (automated tracking, clear definition)
- Unreliable: "Employee happiness" (subjective, hard to define, measurement affects result)
If you can't measure it reliably, either improve measurement method or choose different metric.
Test 4: Predictive or Outcome
Question: Does this metric either:
- Predict an outcome we care about (leading indicator), OR
- Measure an outcome we care about (lagging indicator)?
Leading indicators predict:
- Trial starts predict paid conversions
- Engagement predicts retention
- NPS predicts growth (if validated in your context)
Lagging indicators measure:
- Revenue
- Customer retention
- Profit
Neither:
- Page views (doesn't predict outcomes in most contexts)
- Social media followers (weak predictor)
If metric neither predicts nor measures outcomes, it's noise.
Test 5: Non-Redundant
Question: Is this captured by another metric we're already tracking?
Redundancy waste:
- Clutters dashboards
- Dilutes focus
- Creates confusion ("Which metric do we optimize?")
Example:
- Tracking: Monthly recurring revenue (MRR), annual run rate (ARR = MRR × 12)
- Problem: Redundant—both show same information
- Fix: Pick one
If metric is redundant, eliminate it or consolidate.
What to Measure: Domain Examples
Product Development
| What to Measure | Why | What to Avoid |
|---|---|---|
| Activation rate (% completing first key action) | Predicts retention | Total signups (many never activate) |
| Feature adoption rate | Shows value of features | Features shipped (doesn't mean usage) |
| Time-to-value (days to first success) | Predicts retention, satisfaction | Time to ship features (output, not outcome) |
| Weekly active users / Monthly actives | Shows engagement depth | Total user count (includes inactive) |
| Retention cohorts (% active after 30/60/90 days) | Core product health | Vanity metrics (downloads, views) |
Focus: Measure whether users get value, not just whether they show up.
Marketing
| What to Measure | Why | What to Avoid |
|---|---|---|
| Customer acquisition cost (CAC) | Determines profitability | Total ad spend (no context) |
| CAC payback period | Shows time to recover investment | Impressions (doesn't mean engagement) |
| Conversion rate by channel | Identifies effective channels | Traffic (doesn't mean quality) |
| Marketing-attributed revenue | Shows ROI | Activity metrics (emails sent, posts published) |
| Lead-to-customer rate | Shows pipeline efficiency | MQLs without conversion context |
Focus: Measure cost-effectiveness and revenue impact, not activity.
Customer Success / Support
| What to Measure | Why | What to Avoid |
|---|---|---|
| Customer churn rate | Core retention metric | Total support tickets (could indicate engagement) |
| Net revenue retention (expansion - churn) | Shows growth from existing customers | Customer satisfaction alone (doesn't predict behavior) |
| Time-to-resolution | Affects satisfaction | First response time (doesn't mean problem solved) |
| Customer health score (engagement, usage, satisfaction) | Predicts churn | Tickets closed (doesn't mean quality) |
| Expansion revenue rate | Shows upsell success | Support team size (input, not outcome) |
Focus: Measure retention and expansion, not just support activity.
Sales
| What to Measure | Why | What to Avoid |
|---|---|---|
| Win rate (deals closed / total opportunities) | Shows close effectiveness | Sales calls made (activity) |
| Sales cycle length | Affects capital efficiency | Opportunities created (doesn't mean quality) |
| Average contract value | Revenue per deal | Pipeline value (doesn't account for close rate) |
| Customer acquisition cost | Profitability per customer | Demos given (activity) |
| Lead-to-close rate | End-to-end efficiency | Meetings booked (doesn't predict revenue) |
Focus: Measure conversion efficiency and deal quality, not activity volume.
Content / Media
| What to Measure | Why | What to Avoid |
|---|---|---|
| Engaged time (active reading/viewing) | Shows actual consumption | Page views (doesn't mean reading) |
| Conversion rate (content → email/trial/purchase) | Shows business impact | Social shares (doesn't predict behavior) |
| Return visitor rate | Shows value delivered | Bounce rate (often misleading) |
| Content-attributed revenue | Shows ROI | Articles published (output) |
| Subscriber growth rate (from high-engagement sources) | Shows audience building | Total followers (many inactive) |
Focus: Measure engagement depth and business impact, not vanity metrics.
How Many Metrics to Track
The 3-7 Rule
For any given goal, focus on 3-7 key metrics.
Why this range:
- Less than 3: Incomplete picture, miss important drivers
- More than 7: Diluted focus, too many metrics to act on
Example: Product Team's Key Metrics
- Weekly active users (engagement)
- Activation rate (new user success)
- 60-day retention (long-term stickiness)
- Net Promoter Score (satisfaction)
- Feature adoption rate (value realization)
Five metrics. Manageable. Each actionable.
Organize by Layer
Andy Grove, who pioneered output-based management at Intel, framed the core tension well: "The key result has to be measurable. But at the end you can look, and without any argument: Did I do that or did I not do it? Yes? No? Simple. No judgments in it."
Create measurement hierarchy:
| Layer | Metric Count | Purpose | Review Frequency |
|---|---|---|---|
| North Star | 1 | Captures core value | Weekly |
| Primary | 3-5 | Key drivers of North Star | Weekly |
| Secondary | 5-10 | Supporting metrics, diagnostics | Monthly |
| Operational | 10-20 | Detailed tracking | As needed |
Example: SaaS Company
- North Star: Net Revenue Retention (captures retention + expansion)
- Primary: Activation rate, engagement score, churn rate, expansion rate
- Secondary: CAC, LTV, feature adoption, NPS, support satisfaction
- Operational: Funnel conversion rates, A/B test results, traffic sources
Focus daily on North Star and Primary. Check Secondary monthly. Review Operational when diagnosing issues.
Common Measurement Mistakes
Mistake 1: Measuring Everything
Problem: "If we track everything, we'll understand everything"
Reality: Too many metrics create noise, not signal
Russell Ackoff put it directly: "The manager's most important decisions are about what data to collect and what to ignore." Collecting everything sidesteps that decision rather than making it.
Symptoms:
- 30+ metric dashboards
- No one knows which metrics matter
- Decisions still made on gut feel
- Analysis paralysis
Fix: Radical prioritization—identify vital few, ignore rest
Mistake 2: Measuring What's Easy
Problem: Tools auto-generate metrics, so you track them
Reality: Easy-to-measure ≠ important to measure
Example:
- Easy: Page views, session duration, bounce rate (default analytics)
- Hard but important: Activation rate (requires defining "activated"), cohort retention, customer lifetime value
Fix: Measure what matters, not what's automatic
Mistake 3: Outputs Instead of Outcomes
Problem: Tracking activities, not results
Example:
| Team | Output (What You Do) | Outcome (What It Achieves) |
|---|---|---|
| Marketing | Blog posts published | Content-attributed revenue |
| Sales | Calls made | Win rate, revenue |
| Product | Features shipped | Feature adoption, retention |
| Support | Tickets closed | Customer satisfaction, retention |
Fix: Measure outcomes first, use outputs to understand drivers
Mistake 4: Metrics Without Context
Problem: Absolute numbers without comparison
Example:
- "We have 100K users" (Is that good? Growing? Engaged?)
- Better: "We have 100K users, up 15% MoM, with 40% weekly active"
Fix: Always provide context:
- Comparison (vs. last period, vs. goal)
- Rates and ratios (not just absolutes)
- Segmentation (averages hide patterns)
Mistake 5: No Validation
Problem: Assuming metric predicts outcomes without testing
Example:
- Assume: High NPS → Growth
- Reality: In some businesses, NPS doesn't correlate with revenue or retention
Fix: Validate predictive metrics:
- Track both metric and ultimate outcome
- Analyze correlation over time
- If metric doesn't predict outcome, drop or replace it
Mistake 6: Gaming and Goodhart's Law
"When a measure becomes a target, it ceases to be a good measure." — Charles Goodhart, economist, Bank of England adviser
Example:
- Metric: Support tickets closed
- Gaming: Close tickets quickly without solving problems
- Result: Metric looks good, customer satisfaction terrible
Fix:
- Use complementary metrics (tickets closed + satisfaction score)
- Mix outputs and outcomes
- Rotate metrics periodically
- Include qualitative feedback
Designing Your Measurement System
Step 1: Define Success
Clarify what you're trying to achieve.
Questions:
- What does success look like in 1 year? 3 years?
- What outcomes actually matter?
- How will we know if we're succeeding?
Output: 1-3 core goals
Step 2: Identify Drivers
What causes goal achievement?
Method: Work backward
Example Goal: Increase revenue
Ask: What drives revenue?
- More customers (acquisition)
- Higher value per customer (expansion)
- Longer customer relationships (retention)
Ask again: What drives each of those?
- Acquisition: Traffic quality × conversion rate
- Expansion: Product value × upsell process
- Retention: Product satisfaction × customer success
Output: Map of causal drivers
Step 3: Select Metrics
For each driver, choose 1-2 metrics.
Apply five tests:
- Aligned with goals?
- Actionable?
- Measurable reliably?
- Predictive or outcome?
- Non-redundant?
Output: 3-7 key metrics per goal
Step 4: Define Metrics Precisely
Avoid ambiguity.
For each metric, document:
- Name: What it's called
- Definition: Exact calculation
- Data source: Where numbers come from
- Frequency: How often measured
- Owner: Who's responsible
- Target: Goal value
- Action triggers: What changes trigger what actions
Example: Activation Rate
- Definition: % of signups who complete [specific action] within 7 days
- Data source: Product analytics (Mixpanel event: "First Value Action")
- Frequency: Weekly
- Owner: Product team
- Target: 40% (current: 32%)
- Action triggers: <30% = investigate onboarding; >45% = document what's working
Step 5: Validate Predictive Power
Test whether metrics actually predict outcomes.
Method:
- Track metric and outcome for 3-6 months
- Analyze correlation
- Look for leading relationship (metric changes before outcome)
Example:
- Hypothesis: Activation rate predicts 60-day retention
- Test: Track both for 6 months
- Result: 0.82 correlation, activation changes precede retention changes by 3-4 weeks
- Conclusion: Valid leading indicator
If metric doesn't predict outcomes, replace it.
Step 6: Create Dashboards and Rhythms
Make metrics visible and actionable.
Dashboard principles:
- Focus: North Star + Primary metrics on main view
- Context: Always show trends, targets, comparisons
- Segmentation: Enable drilling into segments
- Action-oriented: Link metrics to action items
Review rhythms:
- Daily: North Star (for critical products)
- Weekly: Primary metrics
- Monthly: Secondary metrics, deep dives
- Quarterly: Metric system review (add/remove/refine)
Step 7: Act on Metrics
Metrics only matter if they drive action.
For each metric:
- Green (on track): Document what's working, scale
- Yellow (warning): Investigate, test improvements
- Red (off track): Root cause analysis, action plan
If a metric doesn't trigger action within 90 days, eliminate it.
Conclusion: Measure What Matters
The temptation: Measure everything possible
The reality: More metrics = more noise
The solution: Ruthless focus on vital few
Principles:
- Start with goals (not available metrics)
- Measure drivers (not just activities)
- Validate predictions (test whether metrics correlate with outcomes)
- Limit quantity (3-7 metrics per goal)
- Act on metrics (if not actionable, don't measure)
- Review regularly (eliminate metrics that don't drive decisions)
Good measurement:
- Clarifies what matters
- Informs decisions
- Predicts outcomes
- Enables action
Bad measurement:
- Drowns teams in data
- Measures wrong things
- Creates illusion of understanding
- Leads nowhere
Measure less. Measure better. Act more.
Your decisions—and outcomes—will improve.
What Research Shows About Measurement Selection
The question of what should be measured has attracted systematic research attention from multiple disciplines. The convergent findings are more definitive than practitioners typically recognize.
Douglas Hubbard's work on measurement economics provides the most rigorous framework for deciding what to measure. In How to Measure Anything (2014), Hubbard argues that measurement decisions should be made on the basis of expected value of information: a measurement is worth making when the reduction in decision uncertainty it produces is worth more than the cost of making the measurement. This reframes the question entirely. Organizations typically ask "can we measure this?" or "should we track this?" Hubbard's question is: "What decision does this measurement inform, and by how much does knowing the answer change what we would do?" If the answer is "not much," the measurement provides little value regardless of how accurate or inexpensive it is.
Hubbard documented that organizations consistently overestimate how often they need measurement and underestimate how much uncertainty they have already reduced. In his consulting work, he found that most high-stakes decisions are made with less uncertainty than decision-makers believe, because prior measurements, domain expertise, and base rate data already constrain the range of plausible outcomes significantly. The implication is that measurement selection should begin with explicit uncertainty quantification: what are the key assumptions this decision depends on, and how uncertain are we about each?
Robert Kaplan and David Norton approached measurement selection through the lens of strategic causality. Their development of strategy maps (documented in Strategy Maps, 2004) was a response to an observed failure: organizations using the Balanced Scorecard would select metrics for each perspective without specifying the causal relationships between them. The result was four disconnected measurement sets rather than an integrated theory of how to achieve strategy.
Strategy maps required organizations to specify: if we improve employee learning in dimension X, we hypothesize this will improve internal process quality in dimension Y, which will drive customer outcome Z, which will produce financial result W. Each arrow in the map is a causal hypothesis that can be tested against data. This approach to measurement selection -- choosing metrics that test specific causal theories about how strategy works -- produces fundamentally different metric systems than those selected by looking at available data and choosing what seems important.
W. Edwards Deming contributed the operational counterpoint: not everything worth managing can or should be measured quantitatively. Deming's critique was directed at the American management practice of the 1970s and 1980s, which he observed had become obsessed with financial metrics while ignoring the process quality factors that actually drove financial results. His famous statement -- "it is wrong to suppose that if you can't measure it, you can't manage it" -- was a direct challenge to the assumption that unmeasured factors were unimportant. Customer satisfaction, employee morale, supplier relationships, and process improvement capacity are all things that matter enormously and cannot be fully captured in a number.
Deming's practical guidance was to focus measurement where it reduces decision uncertainty, and to rely on direct observation, qualitative judgment, and process knowledge for factors that resist valid quantification. Organizations that try to force everything into a quantitative metric framework typically end up measuring surrogates for the things that matter rather than the things themselves.
Andy Grove's output management framework at Intel addressed the measurement selection problem at the team level. Grove's principle, from High Output Management (1983), was that a manager's output is the output of their team -- not their activities, their inputs, or their intentions. This implied a specific discipline for measurement selection: identify the deliverable outputs of each role, then measure whether those outputs are achieved. Not "did the engineer write code" but "did the feature ship and work." Not "did the salesperson make calls" but "did the salesperson close deals." Grove was explicit that activity measurement (calls made, hours worked, meetings attended) is a proxy for output measurement, and that proxies degrade over time as people learn to optimize the proxy rather than the output.
Real-World Case Studies in Measurement Selection
Intel's OKR system. When Andy Grove implemented the OKR (Objectives and Key Results) framework at Intel in the 1970s, he was solving a specific measurement selection problem: how to identify, in a company with thousands of employees and dozens of competing priorities, which things were actually critical to measure and which were noise. Grove's solution was to require that every key result be something that would clearly and unambiguously tell you whether the objective was achieved. This forced measurement selection to be tied to specific decisions: if this result is achieved, we know the objective succeeded; if it is not, we know it failed and must adjust.
When John Doerr brought OKRs to Google in 1999, the framework forced the same discipline on a startup that had abundant data but unclear priorities. The OKR requirement that key results be quantifiable and binary (achieved or not) eliminated a class of measurements that sound important but do not actually inform decisions: "improve user satisfaction," "become more innovative," "grow our team capability." These are goals, not measurable key results. The OKR discipline of requiring specific, measurable, time-bound key results is essentially a system for measurement selection: it forces organizations to specify exactly what evidence would demonstrate that a goal has been achieved.
The Enron measurement failure. Enron's failure is partly a story about selecting the wrong things to measure. The company's financial reporting focused on revenue growth, earnings per share, and credit ratings -- all of which were managed through financial engineering rather than genuine business improvement. What Enron did not measure, or measured inadequately, were the actual cash flows from its trading operations, the quality of counterparty risk in its energy contracts, and the sustainability of profit margins in its broadband and water businesses. When Mark-to-market accounting was applied to these operations, the accounting revenue metric bore little relationship to cash generation. Had the measurement system included cash flow from operations alongside revenue, the divergence would have been apparent years before the collapse.
NHS balanced measurement. The British National Health Service's experience with measurement evolution illustrates how measurement selection should change in response to learning. The initial focus on waiting time metrics drove gaming behavior (documented in multiple Audit Commission reports). The response was not to abandon measurement but to add measurement dimensions: clinical outcome metrics (mortality rates, complication rates), patient experience metrics (from the NHS Patient Survey), and safety metrics (adverse events, near-misses). Each additional measurement dimension made it harder to game any single metric because improvement on one at the expense of another would be visible. The NHS Seven Steps to Patient Safety framework, introduced in 2004, embedded this multi-dimensional measurement approach in institutional practice.
Google's North Star evolution. Google's experience selecting its North Star metric illustrates the difficulty of measurement selection at platform scale. Early in Google's history, the dominant metric was search queries per day -- a measure of how often people used the product. This drove improvements in search quality (more queries meant the product was valuable) but also misaligned incentives: the metric improved when people needed to search multiple times to find an answer, which indicated poor search quality. Google shifted to a metric more directly tied to user value: the number of users who found what they needed on the first search result page. This was harder to measure but more directly aligned with the actual goal. The lesson is that measurement selection requires constant attention to whether the metric measures the goal or just correlates with the goal -- and whether that correlation holds as the organization optimizes toward the metric.
Evidence-Based Principles for Measurement Selection
Principle 1: Begin with decisions, not with data. The most consistent finding across Hubbard, Grove, Kaplan, and Norton is that effective measurement selection starts by identifying the decisions that measurement is supposed to inform. What would you do differently if you knew X versus Y? If the answer is the same action regardless of the measurement result, the measurement provides no decision value. Organizations that build measurement systems by surveying available data and selecting what seems important typically end up measuring what is easy rather than what is decision-relevant.
Principle 2: Specify causal theories before selecting metrics. Kaplan and Norton's strategy map requirement -- that organizations specify the causal chains linking activities to outcomes before selecting metrics for each link -- produces better measurement selection than either top-down metric mandates or bottom-up data availability surveys. The causal theory specifies what the metric is supposed to measure (a specific link in a specific chain) and makes explicit what evidence would confirm or refute the theory. This gives measurement selection a scientific structure: hypothesize, measure, test, revise.
Principle 3: Measure at the level closest to the actual outcome. Grove's output management principle applies generally: measurement should be as close to actual value delivery as possible. Activity metrics (inputs and processes) are furthest from value and most susceptible to gaming. Output metrics (delivered products and services) are closer. Outcome metrics (actual customer or organizational results) are closest. Each step toward the outcome increases measurement difficulty but decreases gaming susceptibility and increases decision relevance. Organizations should invest in outcome measurement even when it is more expensive and less precise than activity measurement, because the informational value is systematically higher.
Principle 4: Validate that metrics actually predict the outcomes they are supposed to predict. Deming's insistence on statistical thinking includes the requirement that measurement hypotheses be tested. An organization may believe that employee engagement scores predict retention, that NPS predicts revenue growth, or that feature adoption rates predict churn. These are hypotheses, not facts. They should be tested by tracking both the metric and the outcome over time and examining whether the predicted relationship holds. When it does not, the metric should be revised or replaced. Measurement selection is not a one-time design problem; it is an ongoing empirical inquiry into what actually predicts what.
About This Series: This article is part of a larger exploration of measurement, metrics, and evaluation. For related concepts, see [Vanity Metrics vs Meaningful Metrics], [KPIs Explained Without Buzzwords], [Designing Useful Measurement Systems], and [Why Metrics Often Mislead].
References
Kaplan, R. S., & Norton, D. P. (1996). The Balanced Scorecard: Translating Strategy into Action. Harvard Business School Press.
Hubbard, D. W. (2014). How to Measure Anything: Finding the Value of "Intangibles" in Business (3rd ed.). John Wiley & Sons.
Croll, A., & Yoskovitz, B. (2013). Lean Analytics: Use Data to Build a Better Startup Faster. O'Reilly Media.
Marr, B. (2012). Key Performance Indicators (KPI): The 75 Measures Every Manager Needs to Know. Financial Times/Prentice Hall.
Parmenter, D. (2015). Key Performance Indicators: Developing, Implementing, and Using Winning KPIs (3rd ed.). John Wiley & Sons.
Goodhart, C. (1975). "Problems of Monetary Management: The U.K. Experience." Papers in Monetary Economics (Reserve Bank of Australia).
Kerr, S. (1975). "On the Folly of Rewarding A, While Hoping for B." Academy of Management Journal, 18(4), 769–783.
Davenport, T. H., & Harris, J. G. (2007). Competing on Analytics: The New Science of Winning. Harvard Business School Press.
Ries, E. (2011). The Lean Startup: How Today's Entrepreneurs Use Continuous Innovation to Create Radically Successful Businesses. Crown Business.
Hope, J., & Fraser, R. (2003). Beyond Budgeting: How Managers Can Break Free from the Annual Performance Trap. Harvard Business School Press.
Neely, A., Adams, C., & Kennerley, M. (2002). The Performance Prism: The Scorecard for Measuring and Managing Business Success. Financial Times/Prentice Hall.
Eckerson, W. W. (2010). Performance Dashboards: Measuring, Monitoring, and Managing Your Business (2nd ed.). John Wiley & Sons.
Skok, D. (2015). "SaaS Metrics 2.0 – A Guide to Measuring and Improving What Matters." For Entrepreneurs (blog).
Ellis, S., & Brown, M. (2017). Hacking Growth: How Today's Fastest-Growing Companies Drive Breakout Success. Crown Business.
Maurya, A. (2012). Running Lean: Iterate from Plan A to a Plan That Works. O'Reilly Media.
Frequently Asked Questions
How do you decide what to measure?
Start with your goals, identify what drives those outcomes, focus on actionable metrics, and measure what you can actually influence.
Should you measure everything you can?
No. Too many metrics create noise, dilute focus, and waste resources. Measure what matters and drives decisions.
What makes a metric worth measuring?
It's actionable, aligned with goals, measurable reliably, influences decisions, and provides insight not available through other means.
How many metrics should you track?
Focus on 3-7 core metrics for any given goal. More creates confusion and dilutes attention from what truly matters.
What's the difference between outputs and outcomes?
Outputs are activities you do; outcomes are results those activities create. Focus on outcomes, use outputs to understand drivers.
Should you measure leading or lagging indicators?
Both. Lagging indicators show results; leading indicators predict future performance. You need both for complete insight.
What's the danger of measuring the wrong things?
You optimize for metrics instead of goals, miss what actually matters, and create perverse incentives that harm real performance.
How do you know if you're measuring the right things?
Your metrics inform decisions, changes in metrics correspond to real performance, and achieving metric targets advances actual goals.