Why Productivity Metrics Fail Knowledge Teams

Productivity metrics are the measures a team uses to describe how much valuable work it produced over a period, and most of them are plain counts: tickets closed, commits pushed, calls handled, documents shipped, story points burned. Counts are cheap to collect, easy to chart, and comforting to report upward, which is why they spread. They also break in a specific and predictable way once people know they are being counted. In knowledge work, where the size and difficulty of each unit varies enormously, the break is bad enough that the metric ends up worse than no metric at all.

This article explains the mechanism, names the gaming behavior each common count produces, covers the developer and sales variants separately because they behave differently, and sets out what we would use instead.

What Goodhart's Law does to productivity metrics

Goodhart's Law states that when a measure becomes a target, it ceases to be a good measure. The original observation came from economics: a statistical regularity that holds while nobody acts on it tends to collapse once policy starts pushing against it. The same thing happens inside a team. A count that quietly described good work becomes a thing people manage directly, and the link between the number and the underlying quality comes apart. Which measures survive that pressure is the subject of employee performance tracking.

Why does the link break rather than simply weaken? Because there is nearly always a cheaper way to move a number than to do the work it represented. Splitting one ticket into four raises your count without helping a single customer. Nobody has to be dishonest for this to happen. People respond to what gets noticed, and a published count is the loudest signal in the room. Our complete guide to employee productivity works through the same distinction between activity and value in more depth.

The cost of getting this wrong is not confined to a spreadsheet. Gallup reported in April 2026 that global employee engagement declined for the second year in a row in 2025, reaching its lowest level since 2020, and estimated the cost of low engagement at roughly $10 trillion in lost productivity, or 9% of global GDP. Gallup also found the decline was most acute among managers, who are usually the people holding the dashboard.

Which productivity metrics get gamed, and how

Every common count has a well-worn workaround, and the workaround usually looks like compliance rather than cheating. Any manager who has run these numbers for a year will recognize most of the middle column.

Every consequence in the right-hand column arrives later than the metric improvement, and lands on someone other than the person being measured. That delay is exactly why output targets feel like they are working for a while.

Productivity metrics by function: engineering and sales

Two functions get asked for numbers more than any other, and they fail in opposite directions. Engineering work is highly variable and interdependent, so counts there are nearly meaningless. Sales work has a cleaner unit, so counts there carry real information and still need a counterweight.

Developer productivity metrics in software engineering

Individual developer counts measure the queue, not the engineer. Commits, pull requests, and story points all reward small, safe, visible work and punish the person who spends a week deleting a subsystem nobody wanted to touch. DORA is the sane alternative, holding deployment frequency and lead time for changes against change failure rate and time to restore service. SPACE adds the dimensions DORA leaves out, including satisfaction and collaboration. Both are explicitly team measures, and using either to rank individuals is a misuse of the framework.

Sales productivity metrics

Sales counts are more defensible and still need pairing. Calls dialed and emails sent measure motion. Qualified pipeline created measures something closer to the job, and it means far more reported next to retention of those accounts at 90 days. Without the second number, a quarter of aggressive qualifying looks identical to a quarter of good selling, and you find out which one it was three quarters later, when renewals come due.

Why counting individuals damages collaborative work

Individual counts tax the work that holds a team together. Reviewing a colleague's draft, pairing on a hard bug, writing the document that stops five people repeating the same mistake: none of it produces a unit with your name on it. When the scoreboard is personal, helping someone else is a cost you pay out of your own score. Reporting at the team level is what keeps this honest, and our guide to workplace analytics without surveillance covers the design rules that stop aggregate reporting collapsing back into individual scores.

There is a second problem, more structural and just as damaging. In interdependent work, an individual's throughput is largely set by the queue in front of them. A designer waiting three days for approvals looks slow, and no amount of personal effort fixes that. You end up measuring the system and blaming the person. When the work is distributed, the temptation gets stronger, which is why we wrote a separate guide on how to measure employee productivity when you cannot see the work.

What should replace raw productivity metrics

Replace single counts with pairs: a volume or speed measure reported next to a quality measure that moves the other way when the first is gamed. It is the most effective change available and it costs almost nothing to make.

Pair volume with a counterweight

Tickets closed pairs with repeat contact rate. Cycle time pairs with rework rate. Deals closed pairs with retention at 90 days. Speed alone is easy to fake. Speed with stability sitting beside it is not, and the pair turns a status update into an actual conversation about tradeoffs.

Measure outcomes rather than units

Ask what would be different if this work went well, then measure that. A support team's outcome is customers who stop having the problem, not tickets moved to closed. A content team's outcome is qualified inquiries or reduced support load, not pieces published. Outcomes are slower and noisier than counts. They are also much harder to game.

Watch the trend, not the value

A single period's count means very little. A three-month trend in the same measure, read alongside its counterweight, tells you whether something real changed. We would rather see six months of two paired measures than a weekly dashboard of twelve, and a short productivity audit run over one week is usually enough to work out which two are worth keeping.

The same discipline applies to how numbers are presented. A productivity report carrying four measures that constrain each other is far harder to game than one built around a single headline figure.

Which work genuinely cannot be counted

Some work has no honest unit, and pretending otherwise produces fiction. Research where the answer might be no. Mentoring. Architecture decisions that prevent a year of pain. Design exploration where four discarded directions were the point. Incident prevention, which by definition leaves no trace.

So how do you evaluate that work? Through review by someone competent to judge it, plus the accounts of the people it affected. Written narrative, work samples, and peer input are not soft substitutes for metrics. For this category they are the more accurate instrument. Say so plainly, rather than inventing a proxy that flatters the spreadsheet.

Pressure makes the fiction worse. Microsoft's 2025 Work Trend Index, drawn from anonymized Microsoft 365 telemetry plus a survey of 31,000 knowledge workers across 31 markets, found that 48% of employees and 52% of leaders describe their work as chaotic and fragmented, and that 1 in 3 employees say the pace of work makes it impossible to keep up. A team in that state will hit whatever number you set and quietly drop the work you forgot to count. The early warning signs are covered in signs of employee burnout hiding in workload data.

A caveat worth stating plainly: this argument is weakest for high-volume, standardized work. A team processing near-identical claims has a real unit of output, and counting it is reasonable and fair. Everything here is aimed at variable, interdependent knowledge work.

How do you move a team off output targets?

Move in one visible step rather than a quiet drift, and tell leadership what to expect before the first number moves.

1. List every count that currently feeds a goal, a review, or a bonus.
2. For each one, write down the gaming behavior you have already seen.
3. Pick a counterweight for the counts worth keeping, and demote the rest to diagnostics.
4. Move all reporting to team level, and say publicly that you have done it.
5. Warn leadership that throughput may dip for a quarter, and explain why in advance.
6. Review after a quarter using quality and predictability, not the counts you dropped.

Step five is the one that fails most often, because it needs a format leadership trusts. Our guide to reporting team capacity to leadership covers how to present paired productivity metrics without inviting a request for the old scoreboard back. If you are looking for what to do instead of counting, we cover how to improve employee productivity lever by lever.

Key takeaways

• Goodhart's Law is the mechanism: targets corrupt the measures they are built on.
• Every common count has a cheap workaround that looks exactly like compliance.
• Individual counts tax collaboration and mostly describe the queue, not the person.
• Developer counts fail hardest; sales counts survive, but only in pairs.
• Some work has no honest unit. Evaluate it by review rather than inventing a number.