Every KPI is also a behaviour contract. You publish a number; the team optimises for the number; what gets measured gets done — and sometimes what gets measured gets done at the expense of the thing you actually wanted. The economist Charles Goodhart put it best: “when a measure becomes a target, it ceases to be a good measure.” Half of operations leadership is choosing measures that survive becoming targets.
This isn't an argument against KPIs. It's an argument for designing them with the same care you'd apply to a compensation plan — because that's effectively what they are. Here's how we approach it with clients, with examples of how good KPIs quietly become bad ones and what to do about it.
Step 1: define the outcome, not the activity
The most common KPI failure is measuring effort as a proxy for outcome. “Number of customer calls made” is an activity. “Customer retention rate” is an outcome. The activity is easier to count and almost always worse to incentivise.
The fix is a discipline question, not a technical one. For every proposed KPI, ask: if the team maxed this out tomorrow, would the business actually be better off?If the answer is “not necessarily,” you have an activity metric. Find the outcome it's meant to drive and measure that instead.
A working example we've seen go wrong: a support team incentivised on “average handle time.” Times dropped 20%. Customer satisfaction dropped further, and repeat-contact rate rose — because agents were closing tickets quickly without solving the underlying problem. The fix was measuring first-contact resolution and CSAT in combination. Handle time stopped mattering on its own.
Step 2: pair every KPI with its counterweight
The most reliable way to game a metric is to optimise for it without regard to the things it's implicitly trading off. The most reliable way to prevent that is to pair the metric with one that captures the trade-off.
- Speed needs to be paired with quality. A team rewarded only for speed will cut corners; one rewarded only for quality will miss deadlines.
- Volume needs to be paired with conversionor satisfaction. More leads, sold worse, is not a win.
- Cost needs to be paired with service level. The cheapest operation is one that doesn't serve anyone.
A useful design rule: no individual or team should be measured on a single number. Two paired metrics, watched together, are almost always more honest than one.
Step 3: pick metrics the team can actually move
A KPI the team can't influence is just a weather report. People stop reading it. Worse, they stop trusting that any of the metrics are real.
Before publishing a metric, walk through the chain: what specific actions, by which specific people, drive this number? If you can't name two or three, the metric is too high-level for that team. Keep it at the leadership-team layer and find a controllable proxy for the line.
For a finance team, “company revenue” is a weather report. “Days from invoice issued to paid” is something they can move with the right policies. The first one belongs in a board pack; the second belongs on a team dashboard.
Step 4: design against the obvious cheats
Before you publish a KPI, spend 20 minutes asking: if I were rewarded purely on this number, what's the laziest way I could move it? Then design the metric so that path is closed.
- Definition gaps.“Active customers” is meaningless without a definition of active. Tighten it before the team optimises around the ambiguity.
- Easy-segment harvesting. A sales metric that counts deals regardless of size invites a flood of $500 deals. Weight by value, or set a floor.
- Timing tricks. Quarter-end push-and-pull is the oldest trick in the book. Trailing 90-day metrics resist this better than calendar-quarter snapshots.
- Definitional drift.“Resolved” tickets quietly become “closed” tickets when nobody watches definitions. Lock them down in writing.
Step 5: review the metric, not just the number
The job isn't done when the dashboard goes live. A quarterly ritual that asks the right questions is what keeps KPIs honest over time.
- Is the number moving for the reasons we expected?
- Has any other number gotten worse since we started watching this one?
- Has the team found a way to move the metric that we didn't anticipate? Is it a healthy way or an unhealthy way?
- Is the metric still measuring the outcome we cared about, or has the business changed underneath it?
A KPI that survives that review for a year is worth keeping. Most don't. Retiring a metric is healthy; clinging to one because it's been on the wall for three years is how organisations ossify.
A few KPIs we've seen go badly wrong
- Net Promoter Score as the only voice-of-customer metric. Useful at a high level, easy to game when individual teams are paid on it. Sales reps coaching customers on how to score is a real thing.
- Story points completed. Engineering teams figure out within a quarter that inflating estimates makes the chart look better. The metric stops measuring throughput and starts measuring estimation inflation.
- Project on-time delivery. Teams pad deadlines to guarantee green. The metric is hit; the business gets the work slower than it had to.
- Tickets closed per agent. Quality drops, repeat contacts rise, customers leave.
None of these are bad metrics in isolation — they're bad targets. The fix is rarely to drop the metric; it's to pair it, contextualise it, and stop treating it as the single number that determines a team's worth.
The honest bottom line
Good KPIs are designed with the same scepticism you'd apply to a contract clause. You assume someone will read them adversarially, because someone will — usually a well-meaning person trying to do their job and look good doing it. Design metrics that survive that reading, pair them with their counterweights, and review them honestly every quarter. The number on the dashboard then starts to mean what you thought it meant.