Choosing a Performance Metric That Won't Corrupt in a Decade

Ten years is a long time for any metric to stay honest. In HR, the half-life of a good performance indicator is often shorter—three, maybe four years—before people start figuring out how to beat it. You see it all the time: a team hits every target on paper while the actual work degrades. The sales rep who closes easy deals early in the quarter, then coasts. The engineer who churns out low-impact pull requests to meet a count goal. The manager who avoids assigning hard projects because they'd tank the team's satisfaction score.

This isn't malice. It's what happens when a metric becomes the goal instead of a signal. Goodhart's Law says it plainly: 'When a measure becomes a target, it ceases to be a good measure.' So how do you pick a metric that resists corruption for a decade? You start by understanding who needs this and what breaks when you get it wrong.

Who Needs This and What Goes Wrong Without It

HR leaders building new performance systems

You're about to pick a metric that will shape every promotion, every bonus, and every quiet Friday resignation for the next ten years. That sounds dramatic until you watch a company that chose 'tickets closed per week' in 2019—and by 2023 had zero senior engineers left. The wrong number doesn't just mislead; it actively breeds the behavior you least want. I have seen teams optimize so hard for 'response time' that they stopped picking up the phone at all. The metric became the mission. And the mission became a lie.

What breaks first is trust. Not just in the data—in management itself. When people realize the number they chase has nothing to do with real work, they do one of two things: they game it, or they leave. Both hurt. The gaming is subtle at first—a manager reclassifies a late deliverable as 'in progress' to preserve the on-time rate. A developer splits one bug fix into five tickets to boost velocity. By year two, the metric is a fiction everyone pretends is true. By year five, the best people are gone because they can't stomach acting out the pantomime.

The catch is that most HR leaders pick a metric that worked last quarter and assume it will work forever. That's how you end up measuring 'lines of code' in a generation that runs AI-generated patches. Or tracking 'hours logged' when your best contributor works 11 p.m. to 4 a.m. and never touches a time sheet. The decade-proof metric isn't about precision today—it's about resilience to tomorrow's distortions.

Startups scaling past 50 people

Fifty people is the danger zone. At twenty, you can smell bad work. At fifty, you need numbers. And founders, desperate for something trackable, grab the first metric that doesn't require a calculator. Usually revenue per employee. Or NPS. Or a composite score that nobody can explain to the new hire in onboarding. These numbers look clean on a board deck. In practice, they create a culture of polite camouflage. Nobody argues with the metric because nobody remembers how it was built.

I fixed this once by throwing away the entire dashboard and starting with a single question: 'What would a great week look like for one person?' Not the company. One person. We built the metric from that answer. It was ugly. It was specific. And it held up for four years before we needed to adjust. That's the test: can your metric survive a new CEO, a remote-work mandate, and a competitor that copies your product? Most can't. They shatter under the first real shift because they were designed for the conditions of last Tuesday.

That sounds fine until the conditions change. Which they will.

Companies that have already been burned by a bad metric

You know who this article is for? The team that spent six months chasing 'engagement score' only to realize it correlated with free pizza, not retention. Or the HR director who watched 'utilization rate' climb while customer satisfaction plummeted—because people were billing more hours and delivering less value. The scar tissue is real. Once you have watched a metric corrupt your best team, you get paranoid. Good. Paranoia beats naivety.

What usually breaks first is the incentive loop. Someone in finance says 'let's tie comp to this number.' Suddenly the number is sacred. Nobody questions it. Nobody dares improve it honestly because improvement would lower the payout. The metric ossifies. I have seen a company keep measuring 'first response time' for three years after they switched to asynchronous chat—where response time meant nothing. The data was pristine. The decisions were garbage. Worth flagging—the metric had not changed. The work had. That's the silent killer: a static metric in a dynamic system.

'The worst metric is not the one that measures the wrong thing. It's the one that used to measure the right thing and nobody noticed when it stopped.'

— conversation with a VP of People, post-mortem on a failed OKR cycle

So who needs this chapter? Anyone who has ever suspected their scorecard is lying. Anyone who has felt the tension between what the system rewards and what the customer praises. Anyone who wants a metric that will outlast their own tenure. Because the metric you choose today will be gamed, worshipped, distorted, and defended long after you leave the room. Choose something that can survive that kind of love.

Prerequisites: What You Need Before You Choose a Metric

Data Hygiene and Historical Accuracy

Most teams skip this: they pick a metric and then try to backfill three years of data. Wrong order. If your attendance records have four different spellings for the same shift, if your CRM logs timestamps from two time zones, if manager notes live in siloed Excel sheets—nothing you build will hold. I have watched a quarterly bonus plan implode because the HRIS counted a 9:02 AM clock-in as "late" while the payroll system called it "on time." That's not a metric problem. That's a garbage-in problem. Clean the pipes before you connect the gauge.

What does "clean" actually mean? Every row needs a single source of truth for dates, names, and departments. Run a union of your three most-used reports. Where they disagree, flag it. Fix it. Then lock the schema. You will lose a week doing this. Keep the week.

Historical accuracy matters differently. A metric that looks back five years must account for reorganizations, role renames, and software migrations. Did the sales team's "closed-won" definition shift in 2021? Did your engineering promotions include a coding test before 2023 but not after? If you flatten those differences, the trend line lies. One client of ours tracked "time-to-hire" across a merger—quietly merging two separate ATS definitions. The resulting 40% improvement was an artifact, not an achievement. — Anonymous HR analyst, 2024

Clear Definitions of 'Performance' for Your Context

Performance means revenue for sales. It means deployed code for engineering. It means resolved tickets for support. But inside one company, these definitions fight each other. A customer success rep who retains accounts but never upsells—high performance or low? A developer who ships fast but breaks production twice a month—where do they land? Settle this before you pick a number.

The catch is that shared definitions require trade-offs. If you define performance as "output per hour," you incentivize speed over quality. If you define it as "peer reviews score," you encourage politeness over challenge. Neither is wrong. But whichever you choose, write it down—and write down what it explicitly excludes. "We care about revenue attainment, not activity volume." "We rate managers on retention, not on firing speed." That clarity is the anchor. Without it, the metric will drift toward whatever the loudest stakeholder shouts about in Q3.

Odd bit about resources: the dull step fails first.

Odd bit about resources: the dull step fails first.

One rhetorical question here—does your definition survive a debate with a tenured manager who disagrees? If not, the metric won't either.

Leadership Alignment on What Success Looks Like

Definitions land on paper. Alignment lands in rooms. You need the CEO, the VP of Ops, and the head of your biggest department to agree—verbally and in writing—that the chosen metric reflects what they actually want. Not what they say they want. What they fund, promote, and defend in all-hands meetings. A metric that contradicts the CEO's private incentives will be corrupted before the first review cycle.

I have seen a company adopt "net promoter score" as their north star while the same CEO fired two VPs for missing quarterly revenue targets. That contradiction ripped the culture. Managers learned to game NPS surveys—coaching customers to give 9s and 10s—while revenue targets remained the real lever. The metric became a decoration, not a decision tool. That hurts.

So hold an alignment session. Three questions: 1) What behavior should this metric reward? 2) What behavior would we fire someone for—even if the metric looks good? 3) What would we tolerate looking bad on the metric, as long as the underlying result is healthy? Get the answers on slides. Send them out. If someone changes their mind in a month, the metric changes too. But now you know why.

Step-by-Step: How to Select a Decade-Proof Metric

Map the desired behavior, not the output

Start by ignoring the number you want to report. Sounds backward, I know. Most teams begin with a target—say, “10% revenue growth per rep”—and then reverse-engineer a way to measure it. That order is the corruption seed. Instead, sit with the actual human actions that, if done consistently, would produce the result you want over years. I watched a sales team chase “calls per day” until the product broke. Reps dialed numbers, rushed scripts, irritated clients—high volume, zero value. The behavior they wanted? Thoughtful discovery conversations. That metric looked completely different: ratio of follow-up meetings booked per initial call. Slower, harder to game, but stable across a decade. Define the repeatable motion, not the quarterly outcome.

List every behavior that matters. Then kill the three easiest to manipulate.

Stress-test the metric against known gaming scenarios

Here is where you play saboteur. Imagine you're the worst employee who still wants a bonus—how would you inflate this number? If you can think of three ways in ten seconds, the metric is too weak. A classic: “customer satisfaction score” where the survey goes out immediately after payment. So reps beg for ratings before the product is even delivered. That metric rots in ninety days. The fix? Randomize survey timing and exclude responses within 24 hours of any transaction. We did exactly that at a logistics firm—their NPS dropped 14 points overnight, then stabilized, then became a real predictor. The catch is most teams skip the stress test because they trust their own creation. Don’t. Publish the metric, invite critique, let a rival team try to break it. What usually breaks first is the lag—metrics that look good for six months then collapse when incentives reset. That’s your signal to redesign.

“A metric that can be gamed within a quarter isn’t a metric. It’s an invitation to cheat.”

— Operations director, after watching his team double “tickets closed” by creating fake bugs

That quote sticks because it’s true. You won’t notice the rot until the quarterly bonus cycle hits.

Build in a review cadence

No metric survives a decade untouched. The environment shifts—new tools, new regulations, new ways to game the system. So schedule checkpoints: every six months, review the metric against three questions. Is the behavior still the right one? Are people finding shortcuts? Has the business model changed? I write these as recurring calendar events, not optional memos. One team I consulted for ignored this step; their “projects completed” metric was fine until the company switched to agile sprints. Suddenly the metric rewarded tiny, safe tasks over valuable, risky ones. They lost three quarters before catching it. A half-year review would have caught the drift in two weeks. The cadence itself forces honesty—you admit up front that the metric is a living thing, not a monument. That prevents the slow decay that kills most performance systems by year three.

Wrong order kills metrics. Right order? Behavior first, stress test second, scheduled renewal third. Get there and the number will hold.

Tools and Setup: What You'll Need to Track It

HRIS and performance management platforms

Your metric lives inside a system. Pick that system before you pick the tool. Most teams already run an HRIS—Workday, BambooHR, Rippling—but here’s the catch: none of those platforms were built to track a custom metric for a decade. They ship quarterly feature updates, pivot product strategy, and sometimes deprecate the exact field you rely on. I have watched a CHRO lose a six-metric dashboard overnight because the vendor folded a data point into a new module and broke the API mapping.

Anchor yourself to raw-data exports instead of platform-specific charts. Pull timestamps, employee IDs, and the metric’s raw value into a separate table—one you control. That way when the HRIS changes its schema, your historical trend line doesn’t snap. Worth flagging—never store the metric’s *interpretation* (color codes, percentile buckets) in the primary table. Store the number. The interpretation will age; the number won’t.

What about performance management platforms? Lattice, 15Five, Culture Amp—they excel at pulse surveys and review cycles. Use them for the *qualitative* half of your metric (manager sentiment, peer feedback). But don't let them own the quantitative spine. That spine belongs in a warehouse.

Simple dashboards vs. complex BI tools

Most teams over-invest upfront. They buy Tableau or Power BI licenses, hire a dashboard engineer, and produce a dense 18-tile cockpit that nobody opens after month two. That hurts. The truth is—for a single metric that works across a decade—you need three views: a weekly trend, a cohort comparison, and a raw-data audit table. That’s it.

Google Sheets or Airtable can handle the first year. Not sexy. But a spreadsheet forces you to understand every cell. I fixed a client’s broken retention metric by rebuilding it in a shared sheet—we spotted a date-offset bug inside 40 minutes. Their Power BI model had hidden that bug behind a month of row-level security rules. The trade-off is clear: simplicity costs you automation but buys you transparency.

If you must go heavy, buy Looker or Metabase—not Tableau. Reason: Looker’s modeling layer (LookML) lets you define the metric once and reuse it across every report. Tableau forces you to rebuild the calculation per workbook, which invites drift. Three years in, two workbooks will disagree on the same KPI. You don't want that argument in a board meeting.

A rhetorical question worth asking: can an intern replicate your dashboard from scratch in a day? If no, the setup is too complex.

Not every human checklist earns its ink.

Not every human checklist earns its ink.

Pilot groups and A/B testing

You don't roll a decade-proof metric to 10,000 people on day one. You pilot. Pick one department—maybe Customer Support or Engineering—where the metric’s logic is easiest to verify. Run it for 90 days. Then compare the metric’s behavior against actual outcomes: did it flag the low performer who eventually left? Did it stay flat for the steady team that hit every target? If the signal feels noisy, tweak the formula and re-run, don’t scale.

“Our pilot showed the metric predicted turnover at 70% accuracy. We tripled the data window and hit 88%. That extra month saved us a bad org-wide rollout.”

— VP of People Operations, mid-market SaaS company

A/B testing here means running the new metric alongside the old one—not replacing it. Keep both visible for at least two review cycles. Why? Because the old metric corrupts slowly (remember section six?), and the new one might look better simply because people are paying attention (Hawthorne effect). When the lines converge and diverge naturally, you know the new metric is reading reality, not attention.

What usually breaks first is the data pipeline. Your pilot group’s time-tracking tool pushes an integration update, the field name changes from `hours_logged` to `billable_hours`, and suddenly the metric shows a 40% drop. Catch that in a pilot. Fix it with a field alias. Then sleep well before the full launch.

Variations for Different Constraints

Small teams (<50 people): lightweight, qualitative metrics

If your entire company fits in one room, you don't need a multidimensional scorecard. I have watched startups spend two weeks designing a balanced-scorecard system, then abandon it by month three. The constraint here is attention—every hour on metric design is an hour not building product. So keep it simple: one outcome metric (e.g., ‘client tickets resolved within 24 hours’) and one health metric (e.g., ‘team satisfaction score, anonymous, every sprint’). That's enough.

The trap is forcing quantitative rigor too early. A 12-person agency I worked with tried to track billable hour ratios per employee. Within two months, people inflated their time logs, trust frayed, and the metric became a weapon. We fixed this by switching to a binary check: “Did you finish your top three priorities this week? Yes or no.” That sounds soft. It's not. It surfaces blockers faster than any dashboard can. The trade-off: you lose granular comparability between people. But with a small team, comparability is often a mirage anyway—roles blur, context shifts, and one bad week can tank a quarterly average.

What about growth-mode startups? Here the variation is frequency. Small teams in hypergrowth should review their metric every two weeks, not monthly. Why? Because a month is four sprint cycles, and the business model may flip in that window. Pick one metric you won't change for six months—but check its pulse every fourteen days. That rhythm catches corruption early.

‘The worst metric for a team of ten is the one designed for a team of a thousand.’

— Founder of a now-80-person company, reflecting on his first year

Mid-size companies (50-500): hybrid quantitative-qualitative

Now you have layers: managers, departments, maybe a People Ops function. The constraint shifts from attention to alignment—different teams will interpret the same metric differently unless you force a common language. Most mid-size firms default to a stack of individual KPIs (revenue per hire, time-to-fill, satisfaction score). That works until it doesn't. What usually breaks first is the qualitative half—the “how” behind the “what.”

I have seen a 200-person SaaS company where every department hit its numbers for six straight quarters, yet customer churn climbed. The quantitative metrics looked pristine; the qualitative signals (ad-hoc peer feedback, exit interview themes, skip-level meeting notes) were rotting underneath. The fix: a hybrid score where each quantitative KPI has a mandatory one-paragraph narrative field attached. “Why did we hit this? What broke? What was surprising?” The narrative is not optional—if it's blank, the metric is flagged yellow. That forces managers to interpret, not just report. The catch is volume—500 employees generating weekly narratives can drown an HR team. So sample: rotate who submits narratives each week, or require them only for metrics that deviate more than 15% from target.

Another variation: industry matters here. A mid-size construction firm cares about safety incident lag time (quantitative) and supervisor trust scores (qualitative). A mid-size tech firm needs sprint velocity (quantitative) and code review sentiment (qualitative). Don't copy-paste a generic template. The right hybrid for you depends on which metric, if corrupted, would cause the most damage. For construction, it's underreporting of near-misses. For tech, it's gaming of velocity. Identify your single highest-corruption-risk metric, then add a qualitative check that specifically guards that risk.

Large enterprises (500+): multi-metric composites

Scale changes everything. A workforce of thousands generates so much data that single metrics become statistically meaningless or politically impossible to change. The constraint here is inertia—once a metric is embedded in annual bonus plans, firing it takes two quarters and a steering committee vote. So you need composites: bundles of 3–5 sub-metrics that balance each other, so gaming one pulls down another.

For example, a retail chain with 12,000 employees used to reward stores on “sales per employee hour.” Result: managers scheduled fewer people at peak times, service tanked, and long-term revenue dropped. The composite fix: “Sales per employee hour” × “Customer satisfaction score (mystery shopper)” × “Employee turnover rate (voluntary, under 90 days).” No single number can be inflated without damaging the others. That design is not elegant, but it's durable—the composite survived three reorganizations and a CEO change.

The pitfall: composites can become spaghetti. A global manufacturer I advised had a 14-metric composite for plant managers. Nobody understood it, so nobody trusted it. They pruned to five, but weighted two of them at 40% each—which recreated the single-metric gaming problem. The lesson: keep the composite to five sub-metrics maximum, and cap any single weight at 25%. If you need more nuance, create separate composites for different job families (e.g., sales versus operations) rather than trying to fit everyone under one umbrella. That said, large enterprises often resist this differentiation because “fairness” demands identical metrics across roles. Worth flagging—that fairness argument is itself a corruption risk. Identical metrics across fundamentally different roles produce misaligned behavior in every department. Fight for differentiation. Your composite will last a decade only if it reflects the actual work, not the org chart.

Pitfalls: What to Check When the Metric Starts to Fail

Early signs of gaming: tail distribution shifts

Most teams watch the average. That's their mistake. The metric looks healthy—mean numbers flat, targets hit, dashboards green. Meanwhile the tail distribution has quietly twisted into something unrecognizable. I have seen this happen with a call-center resolution metric: average handle time stayed stable for eighteen months, but the 95th percentile crept up by 40 seconds while the 5th percentile dropped by 12 seconds. The spread told the truth. What caused it? Agents learned to dump short, incomplete calls into a “callback later” bucket to keep their personal averages clean, and those callbacks never closed within the tracked window. The average smiled at you. The tail screamed.

You catch this by plotting the decile spread every sprint—not just the median. If one end moves faster than the other, somebody is squeezing the system. A second check: compare the metric’s variance week-over-week. When variance suddenly drops while the mean holds, that’s often a sign of coordinated gaming—teams aligning on the minimum viable output to hit the number, nothing more. Worth flagging: a low-variance, stable-mean metric is not always healthy. Sometimes it just means people have learned exactly how little they can do without triggering alarms.

Unintended consequences: retention drop, silos, short-termism

A metric that survives a decade won’t just resist cheating—it won’t make your best people quit. That sounds fine until you measure individual ticket-closure rates in engineering. Suddenly no senior dev wants to touch the gnarly legacy bug that takes three weeks to untangle; they cherry-pick easy UI tweaks. Closure rate goes up. Quality tanks. And the people who actually fix the hard problems? They burn out watching their “performance” score flatline. I fixed this once by swapping the metric from tickets closed to “weighted complexity-adjusted completions” with a two-month lag on recognition. Retention recovered inside a quarter.

Reality check: name the resources owner or stop.

Reality check: name the resources owner or stop.

The catch is that silo formation happens slower than gaming, so you miss it. Watch for cross-team collaboration metrics dropping off—shared code commits, joint project tags, peer-mentoring hours. When those fall, the metric has started corrupting the culture, not measuring it. Short-termism is easier to spot: look at the ratio of quick wins (under two weeks) to long-cycle work (over eight weeks). If that ratio doubles while the primary metric holds, your decade-proof candidate just became a quarterly-optimization trap.

“A metric that survives a decade won’t just resist cheating—it won’t make your best people quit.”

— HR operations lead, after a failed NPS-based bonus system

Data integrity issues: missing or manipulated inputs

What usually breaks first is the data pipeline, not the metric itself. Someone changes a CRM field label, the ETL job silently drops 12% of records, and your “complete” metric is now measuring a ghost cohort. You need a daily row-count check against a trusted source—HRIS headcount, for example—with an alert if the gap exceeds 2%. We set this up after discovering six months of quarterly bonus data was missing everyone who transferred departments mid-cycle. The metric looked fine. The spreadsheet was lying.

Manipulated inputs are harder. Sales teams stuffing the pipeline with fake leads to inflate conversion rates. Engineers creating empty GitHub repos to pad commit frequency. The fix is audit-log sampling: every quarter, pull 50 raw input records for the metric and verify them against the original source (ticket system, timesheet, what have you). If more than 5% show signs of fabrication—timestamps outside work hours, identical boilerplate text, duplicate customer IDs—the metric is already compromised. Don't patch it. Replace the input layer entirely. A metric corrupted at the source can't be salvaged with weighting or normalization; you're just polishing a forged document.

FAQ: Common Questions About Performance Metrics in HR

How often should I review the metric?

Quarterly works. Not monthly — that breeds noise-panicking. Not annually — that lets rot settle. I have seen teams review a perfectly good headcount-efficiency ratio every month, only to abandon it because one spike from a seasonal project made the number look ugly. The catch is this: a metric meant to last a decade must be allowed to breathe across seasons. Review it every three months, but resist the urge to tweak the formula unless the business model itself shifts. Add a six-month check-in with a junior analyst who has no history with the metric — fresh eyes catch the calcification that veterans miss.

What if a quarter shows a sudden drop? Don't rewrite the metric. Instead, build a short log — one paragraph — explaining what outside force caused the dip. That log becomes your decade-long memory. Most teams skip this; they see a blip, change the formula, and lose comparability forever.

What if my metric conflicts with another team's goal?

Good. Conflict is not bug — it's signal. A retention metric that fights a performance metric usually reveals a real trade-off your company refuses to name. I once watched a product division optimize for "time-to-promotion" while the engineering side pushed "code-review depth." The two metrics screamed at each other. Instead of merging them into one bland average, we surfaced the conflict in a monthly ops review. The teams negotiated: engineers would accept slower promotion rates in exchange for a separate "learning velocity" measure. That worked. The disaster scenario is hiding the conflict; that guarantees the metric with the louder sponsor wins, and the quieter one silently corrupts your culture.

Worth flagging—a conflicting metric pair can be healthy if both are transparent. The danger comes when one metric is secret or unofficially weighted higher. Open the numbers. Let the tension live in a shared dashboard, not a whispered complaint.

How do I balance retention and performance in one metric?

You don't. Combining them into a single index is like blending speed and fuel economy into one dial — the result tells you nothing actionable. Instead, hold both as separate dashboard items and set a constraint: keep retention above 85% while performance scores stay in the top quartile. That forces explicit, honest choices. A single blended "Q-score" masks whether you're retaining low performers or pushing high achievers out the door. The trick is not fusion — it's a rule-of-two that creates tension you can actually manage.

“We tried a composite retention–performance index for two years. It always looked fine until someone asked: ‘Which part of this number is keeping the wrong people?’”

— VP of People Ops, midsize tech firm

That hurts. The composite hid the dark truth until it was too late. Your next step: pull the two measures apart tomorrow. Run them side-by-side for one quarter before deciding whether the constraint approach works for your team. Not yet ready for a dashboard? A spreadsheet with two columns and a sticky note on the monitor will do. Start there.

What to Do Next: Implementation and Review Schedule

Roll out in stages: pilot, refine, scale

Don't announce your chosen metric on Monday and expect adoption by Friday. That path burns trust. Instead, pick a single team—one whose manager actually wants data, not the one you suspect needs it most. Run a ten-week pilot. Track what the metric tells you, but also track what it doesn't catch. The first version will leak. Maybe it rewards volume over quality. Maybe a quarterly target encourages people to sandbag numbers in month one. I watched a sales team game a "customer satisfaction" score by simply excluding the worst responses from their sample—it took exactly one review cycle to spot the pattern. You catch those exploits in the pilot, not at scale.

Refine the definition. Then scale to two teams. Then four. Ramp slowly enough that your HR ops team can update the tracking logic between rounds. That’s the real gate, not enthusiasm.

Set calendar milestones: quarterly review, annual overhaul

Schedule two distinct meetings. First, a quarterly pulse—thirty minutes, no slides. Ask three questions: Is the metric still measuring what we intended? Are people inventing workarounds? What changed in the business that might shift the goal? I have seen a perfectly good "time-to-productivity" metric rot in six months because the onboarding process was restructured and nobody told HR. The quarterly pulse catches that drift.

Second, an annual overhaul. A full afternoon. You will kill at least one sub-metric. You will adjust the weight of another. That feels wasteful until you realize the alternative is a three-year-old metric that now rewards the wrong behavior and everyone knows it except the C-suite. Write the overhaul date on the calendar now—before you even launch the pilot.

“A metric that sits unchanged for twelve months is a metric that has already started to lie to you.”

— overheard during an HR ops post-mortem, after a retention metric accidentally encouraged managers to avoid firing low performers

Create a feedback loop with employees

Your metric will never survive contact with the workforce if it lives inside a spreadsheet only HR sees. Publish the definition. Show the trend line. Ask the people being measured: “Is this fair?”—and mean it. The easiest way to catch corruption early is to let the people who are most affected name it. They will. One operations lead told me, “I can hit that target every quarter, but only if I stop training my new hires for the first two weeks.” That feedback saved us from a metric that would have maximized short-term numbers while hollowing out long-term capability.

Make the loop simple: an anonymous monthly form, one field. “What is this metric getting wrong?” Read every response. Reply publicly when you adjust. Do that three cycles in a row and you build the one thing no metric can replace—credibility.