What this course is really about
Six leader sessions on how to measure people and teams well. Not the mechanics of a review form - the design of the whole system: what a measurement is for, how goals (OKRs) differ from health metrics (KPIs), how to grade what a person actually controls, how bias corrupts ratings, where AI helps and where it becomes surveillance, and the cadence and culture that make monitoring build trust instead of fear. Everything is illustrated on Cartwheel, a mid-size direct-to-consumer online retailer. Tonight is the foundation the other five sessions lean on: measurement can align and motivate people, or it can teach them to game the number and hide - and the difference is a matter of design, not good intentions.
Measurement has two faces 7 min live
"What gets measured gets managed" is a management adage, often attributed to Peter Drucker though he never quite said it. It is half a truth. Measure the right thing well and people align and pull in one direction. Measure the wrong thing, or measure it to judge and punish, and the same people quietly optimise for the number instead of the mission.
LiveMeasure to steer and grow, not only to judge3 min▶
There are three honest reasons to measure a person or a team, and they pull in different directions:
- To steer - are we on track this week, and where do we adjust? Fast, cheap, low-stakes signals.
- To grow - what does this person need to get better? Forward-looking and developmental.
- To judge - the rating, the pay, the promotion. Rare, backward-looking, high-stakes.
Most measurement damage comes from letting the third purpose contaminate the first two. The instant a steering signal also decides someone's bonus, it stops being an honest signal and becomes a number to defend. We will keep pulling these apart all course - and in a4 we make it a rule.
Cartwheel's weekly conversion dashboard. Read as a steering signal, it catches a broken checkout within hours - genuinely useful. Wired to an individual's pay, it becomes a number people spin in the stand-up rather than a truth they surface. The measurement did not change; the intent behind it did, and that is enough to flip its whole effect.
Goodhart's Law and the cobra effect 6 min live
This is not a people problem you can hire around. It is a law. Goodhart's Law: "When a measure becomes a target, it ceases to be a good measure." Campbell's Law (Donald Campbell, 1979) says the same about high-stakes social metrics: the more a number is used for consequential decisions, the more it will be corrupted and the more it will distort what it was meant to track.
LiveThe cobra effect - the story you will not forget3 min▶
Colonial Delhi had a cobra problem, so the administration offered a bounty for every dead cobra brought in. It worked - briefly. Then enterprising residents began breeding cobras to collect the bounty. When officials realised and cancelled the scheme, the breeders released their now-worthless snakes, and the wild cobra population ended up higher than before. The measure ("dead cobras handed in") became the target, people optimised the measure rather than the goal, and the goal moved backwards.
Cartwheel's own cobra. Reward the support team purely on "tickets closed" and you breed cobras: agents close tickets fast, reopen counts climb, and customer satisfaction quietly falls. The number went up. The thing you actually cared about went down. Every metric you attach a reward to is a bounty - always ask what someone would breed to collect it.
Spot the cobra in a Cartwheel target ★ 10 min · think it through
A thought exercise, not a form to fill. Take a target a leadership team might set and predict, out loud, exactly how a rational person would game it - before it ever ships.
The target on the table: Cartwheel's VP of Growth proposes rewarding the marketing team on "number of new sign-ups this quarter".
Run the Goodhart test: if this number alone earns the reward, what is the cheapest way to move it without moving the real goal?
Name the cobras: buy low-quality traffic, incentivise throwaway accounts, run a sweepstakes that attracts non-buyers. Sign-ups soar; paying customers do not.
Design the defence: pair the quantity ("sign-ups") with a quality guardrail ("sign-ups that place a first order within 30 days"), and watch the two together - never one alone.
| If you reward only... | The cobra people breed | Pair it with |
|---|---|---|
| Sign-ups | junk accounts, sweepstakes traffic | activated sign-ups (first order in 30 days) |
| Tickets closed | close-and-reopen, rushing customers | CSAT + first-contact resolution |
| Revenue this month | deep discounts that wreck margin | revenue at healthy margin |
Sort measures into steer, grow, or judge ★ 8 min · think it through
Most measurement pain is a purpose mix-up: a good steering signal doing a judging job it was never built for. Practise the sort on Cartwheel before you do it on your own org.
Take three Cartwheel numbers: weekly conversion rate, an engineer's code-review turnaround, and the annual performance rating.
For each, decide its honest primary job - steer (on track?), grow (how to coach?), or judge (rating and pay).
Flag any number being used for a job it is bad at: conversion rate as an individual's pay input is the classic trap - nobody controls it alone.
Decide the fix: keep conversion as a shared steering signal, move judgment to things the person genuinely controls, and never let a stretch target set a bonus.
The same number, three uses, three fates. Cartwheel's conversion rate is an excellent steering signal, a useful development lens for Product, and a terrible individual pay metric. The number is not the problem - asking one number to steer, grow, and judge all at once is. Separate the jobs and the number becomes trustworthy again. This is the single move that survives the whole course, and a4 turns it into a hard rail.
An OKR score is for steering, never for pay read this
Before we have even written an OKR, plant the linchpin of the whole course: an OKR score exists to help a team steer and learn - to answer "are we on track and what did we learn?" It must never be wired directly to compensation or a rating. The moment it is, people set safe goals, sandbag, and defend the number instead of chasing the mission. We build this out properly in a4 (fair-measurement principles); for now, just hold it as a promise: measurement that builds trust instead of gaming.
Try it yourself - this week ◐ 25-40 min total
- Name three things your organisation measures people on. For each, write down the single cheapest way a rational person could game it. If the cheapest game is easy, you have found a cobra.
- Pick the one you would least want gamed. What quality guardrail could you pair it with so gaming shows up immediately?
- Take one number your org treats as high-stakes (it drives pay or promotion). Ask honestly: is it steering, growing, or judging - and is it any good at the job it is actually used for?
- Optional: read a short summary of Goodhart's Law and Campbell's Law before a2, and keep the cobra story in your pocket for the next planning meeting.
Sources covered (cited)
Taught from the measurement-and-incentives canon. This session covers:
Three questions before you go 🎯 ◐ 90 seconds
1 · "What gets measured gets managed" is best described as...
It is a widely-repeated adage (often mis-attributed to Drucker), not a law. Measure the right thing well and it aligns people; measure the wrong thing, or measure to punish, and people game the number instead.
2 · Goodhart's Law says that when a measure becomes a target...
Once a measure is rewarded, people optimise the measure rather than the underlying goal, and the two diverge - the measured number climbs while the true outcome can fall.
3 · The cobra effect is a vivid example of...
Bounties on dead cobras led to cobra breeding; cancelling the scheme freed the snakes and left more cobras than before. Rewarding the measure ("cobras handed in") moved the real goal backwards.