A KPI Is a Question, Not a Verdict

What if the most dangerous number in your organization is the one everyone agrees to improve?

A key performance indicator appears to offer clarity. Choose something measurable, attach a target, watch it over time, and decisions become easier. Yet the same discipline that makes a KPI useful can also make it destructive. A number can focus attention so effectively that people stop noticing what the number leaves out.

This problem has an unexpected parallel in the way people discover ideas. When ordinary information no longer surprises us, we seek more complicated paths into the world. We follow interesting people, collect unusual signals, and allow those signals to route new material toward us. The result is not merely more information. It is a richer understanding produced by a deliberate interaction between structure and surprise.

These two practices seem unrelated: one belongs to management dashboards, the other to intellectual curiosity. But they share a deeper question:

How do we create a system that gives attention direction without making discovery impossible?

The answer is to stop treating measurement as a verdict. A good KPI is not a complete description of reality. It is a search query aimed at reality, issued repeatedly over time. Its value depends not only on whether the number rises, but on what the number helps us notice, investigate, and learn.

The hidden similarity between a dashboard and a reading habit

Any useful performance indicator contains three elements: a unit of measurement, a goal, and a time series. For example, a company might track monthly customer renewals against a target. A school might track student attendance by week against an expected rate. A hospital might track average waiting time by day against a service standard.

This structure is powerful because it converts vague concern into a navigable question. Instead of asking, “Are we serving customers well?” we ask, “How did renewal rates change this quarter relative to our goal?” Instead of asking, “Are students engaged?” we ask, “What happened to attendance over the last six weeks?”

But the structure is also a form of compression. It takes a sprawling world and reduces it to one measurable pathway. This is similar to writing a very long search query for finding interesting people and ideas. You cannot specify every possible connection you want to encounter. You construct a route through the world, then rely on the route to produce unexpected destinations.

A KPI does something similar. It tells a system where to look. It does not tell the system everything that matters.

Consider a customer support team whose primary KPI is average resolution time. The goal is to reduce it from thirty hours to twelve. Within weeks, the number improves. Agents close tickets faster, managers celebrate, and the dashboard turns green.

Then a secondary signal appears. Customers are reopening more tickets. The team has not solved problems more effectively. It has learned to make unresolved problems disappear from the first report.

The KPI was not false. Resolution time really did decline. The error was treating a search query as a verdict. The team asked, “How quickly are tickets closed?” and behaved as if it had asked, “How effectively are customer problems solved?” Those are different questions, even when they are statistically related.

The central distinction is this:

A metric reports what happened inside a chosen frame. Learning begins when we investigate what the frame excluded.

Why simple metrics become dangerous when they stop surprising us

At first, a metric is informative because it changes what we see. A sudden decline in sales prompts investigation. A rise in employee turnover reveals a problem that anecdotes had obscured. A steady improvement in student enrollment confirms that an intervention may be working.

Over time, however, the metric becomes familiar. Its movements no longer surprise us. We know the seasonal pattern, the reporting cycle, and the likely explanations. The number becomes part of the background. This is the organizational equivalent of intellectual boredom: simple things no longer generate new questions, so attention moves toward more complicated patterns.

That transition matters. A dashboard designed only to confirm familiar expectations gradually loses its ability to teach. People begin to optimize the visible measure instead of exploring the system that produced it.

This is why mature organizations need two kinds of attention:

1. Directed attention, which follows explicit goals and tracks whether essential outcomes are improving.
2. Exploratory attention, which follows anomalies, interesting people, weak signals, and unexpected relationships.

Directed attention protects execution. Exploratory attention protects intelligence.

A business that has only exploration may produce clever observations without reliable results. A business that has only measurement may become efficient at repeating yesterday’s assumptions. The first lacks coordination. The second lacks discovery.

Imagine a product team measuring weekly active users. That number is useful for understanding whether people return. But it cannot explain why some users become passionate advocates while others leave quietly. To discover that, the team might read customer interviews, follow a support specialist who notices unusual complaints, examine a small group of highly engaged users, or investigate a region whose behavior differs from the average.

These activities may not immediately improve the main KPI. Their purpose is different. They generate better questions, which eventually produce better measures and better decisions.

This suggests a more complete model of performance:

Performance is not just the distance between a current value and a target. It is also the quality of the questions generated by that measurement.

The two loop system: control and discovery

A useful way to combine metrics with curiosity is to think in terms of two connected loops.

The first is the control loop. It contains the familiar elements: define a measure, set a goal, observe the time series, and adjust behavior. This loop is appropriate when the desired outcome is clear and relatively stable. If a warehouse must ship orders within twenty four hours, tracking shipping time against a target creates necessary discipline.

The second is the discovery loop. It begins with an unexpected signal, a person with unusual experience, a contradiction, or a persistent question. You investigate, gather context, form a hypothesis, and then decide whether a new measure or experiment is needed.

The control loop asks: “Are we moving toward the goal?”

The discovery loop asks: “Is this still the right goal, and what are we failing to see?”

The loops should not be confused. Discovery without control can become endless browsing. Control without discovery can become polished blindness. The strongest systems alternate between them.

A practical example is a public library trying to increase use of its digital services. Its main KPI is monthly digital checkouts. The target is twelve thousand, compared across months.

The control loop reveals that usage is below target. The library might promote the service more aggressively, improve sign up instructions, or extend access to popular titles.

The discovery loop asks a different set of questions. Are older patrons struggling with device settings? Are certain neighborhoods underrepresented because internet access is limited? Are people borrowing fewer books because the catalog is difficult to search? Is a small group using the service in a completely unexpected way, such as for language learning or professional certification?

Each question may reveal a different intervention. The main KPI gives direction, but the surrounding inquiry supplies meaning.

This is also where interesting people become important. Front line workers, unusual customers, minority users, and specialists with cross domain knowledge often function as human sensors. They notice changes before the aggregate data does. Their observations are not substitutes for measurement. They are routes to measurements that the organization has not yet thought to collect.

Metrics as search queries

A search query is valuable not because it contains the whole answer, but because it retrieves a useful neighborhood of information. The same is true of a KPI.

If the query is too broad, the results are noisy. “Improve the business” is not operationally useful. If it is too narrow, the results are precise but impoverished. “Reduce average call duration by ten percent” might produce a clean improvement while damaging customer trust.

The best metrics have enough specificity to coordinate action and enough openness to invite investigation. They are precise about what they observe, modest about what they claim, and connected to a meaningful outcome.

One way to test a KPI is to ask three questions:

• What behavior will this number encourage?
• What important reality could improve while this number stays flat?
• What could make this number look better while the underlying outcome gets worse?

The third question is especially important because every target creates incentives. If a sales team is judged only by new contracts, it may pursue customers who churn quickly. If a school is judged only by exam scores, it may narrow learning to what appears on the exam. If a software company is judged only by time spent in the application, it may create friction that keeps people engaged without making them successful.

A KPI therefore has at least two lives. It is a descriptive instrument for leaders and a behavioral instruction for everyone being measured. The gap between those two lives is where unintended consequences emerge.

The solution is not to abandon KPIs. It is to design them as question generating instruments rather than isolated commandments.

For every primary KPI, add three supporting elements:

1. A guardrail measure, which detects damage elsewhere. If reducing support time is the goal, track repeat contacts or customer satisfaction as guardrails.
2. An observation channel, which captures information numbers miss. This might include interviews, open ended feedback, field notes, or direct conversations with staff.
3. An anomaly ritual, which gives the team permission to investigate surprising changes instead of explaining them away.

This design turns a dashboard into a living search system. It still tracks progress, but it also routes attention toward the edges of the model.

The discipline of following interesting signals

Curiosity is often described as the opposite of discipline. In practice, useful curiosity requires discipline of its own. Without a method, “interesting” becomes a justification for distraction. With a method, interesting signals become sources of strategic information.

A signal deserves investigation when it has at least one of four properties:

• It is surprising, meaning it contradicts an expectation.
• It is concentrated, meaning a small group or location behaves very differently from the average.
• It is persistent, meaning it keeps appearing despite attempts to explain it away.
• It is connected, meaning it links two areas that are usually treated separately.

Suppose a company’s overall retention rate is stable, but one customer segment renews at nearly twice the average. That segment is not merely a success story. It is a query. What do these customers understand, value, or experience that others do not?

Suppose employee turnover rises in a single office while compensation and workload appear normal. That anomaly could lead to a conversation with a local manager, an examination of promotion patterns, or the discovery of a cultural problem hidden by organization wide averages.

Suppose a writer notices that the most valuable ideas in their inbox come not from famous experts but from people who connect two distant fields. That pattern may change how they search, whom they follow, and what kinds of questions they ask. The discovery is not just an interesting fact. It is a redesign of the information system.

The same principle applies to organizations. The people and places that produce unusual signals may be more valuable than the signals themselves, because they teach you where your current model is incomplete.

This creates a practical habit: once a week, review not only the biggest changes in your metrics, but also the most interesting exceptions. Ask someone close to the work to explain one. Ask a person outside the usual decision circle to interpret another. Then record whether the inquiry produced a new hypothesis, a new measure, or a reason to revise an existing goal.

Over time, this creates an organizational memory of discovery. The company becomes better not only at hitting targets, but at finding targets worth hitting.

Designing a dashboard that keeps the world open

A strong performance system can be built in layers.

At the center is one outcome KPI. It should express the result that matters most, with a clear unit, goal, and time series. Around it sit a small number of guardrails that prevent local optimization from damaging the broader system.

Alongside the formal metrics, maintain an exception log. This is a short record of surprising cases, unusual customer stories, unexplained reversals, and observations from people doing the work. The log should not become a second bureaucracy. Its purpose is to preserve anomalies long enough for someone to learn from them.

Finally, establish a review question that no dashboard can answer by itself:

What has become more interesting since the last review?

This question sounds soft, but it can be made concrete. It might refer to a new customer behavior, an unexpected failure mode, an emerging subgroup, an unexplained correlation, or a person whose perspective changes the interpretation of the data.

The aim is not to replace targets with intuition. It is to prevent targets from becoming the only legitimate form of knowledge.

Key Takeaways

• Treat every KPI as a search query. It directs attention toward one slice of reality, but it cannot describe the whole system.
• Pair outcome measures with guardrails. For every target, identify what could deteriorate while the target improves.
• Protect exploratory attention. Review anomalies, unusual users, front line observations, and unexpected connections as deliberately as you review aggregate performance.
• Use people as sensors. Individuals close to customers, operations, or unfamiliar domains often detect changes before the average does.
• Revise the question, not just the target. If a metric becomes predictable or easy to manipulate, the next improvement may be a better measure rather than a more ambitious number.

The deepest mistake in measurement is believing that clarity means completeness. A number can make a goal visible while making everything outside that goal harder to see. The remedy is not less measurement, but a richer relationship with it.

Use metrics to coordinate action. Use surprise to challenge the frame. Let interesting people and exceptions route you toward questions your dashboard could not have generated on its own.

A mature organization does not merely ask whether it is winning according to the score. It keeps asking whether the score still describes the game. That is the point at which measurement stops being a cage for attention and becomes a way of discovering what deserves attention next.