How this track works
Ten practical sessions, one running company: Cartwheel, a mid-size direct-to-consumer home-and-lifestyle retailer. You will write real OKRs for its teams, choose metrics that are fair to grade people on, set a monitoring cadence that steers without smothering, run performance conversations that separate steering from judgment, use AI as a bias-checking helper (never a judge), and diagnose metrics that have gone wrong. Tonight is the foundation: why we measure, the difference between activity and outcome, and the checklist that decides whether a metric is fair. A running note repeated all track: an OKR score is for steering and learning, never a direct input to someone's pay - the moment it drives compensation, people stop being honest.
Three jobs people jam into one number 6 min live
Most broken performance systems fail for one reason: they force a single number to do three different jobs at once. Pull them apart and almost everything gets easier - and fairer.
LiveActivity, output, outcome, impact - reward the right rung3 min▶
The second foundation is knowing which rung of the ladder a metric sits on:
- Activity - what you did (calls made, hours logged). Easy to count, easy to fake, says nothing about value.
- Output - what you produced (tickets closed, features shipped). Still a task count.
- Outcome - the result that actually changed (customers retained, conversion lifted). This is what to grade.
- Impact - the long-run business effect (revenue, loyalty). Real, but shared and slow.
"Busy" is not "effective". A Cartwheel support agent who closes 60 tickets a day looks productive - until you see the reopens and the falling CSAT. Activity and output are seductive because they are easy to count. Fair measurement climbs the ladder to outcomes: did the customer actually get helped?
The six-point fair-metric checklist 5 min live
Before you put a person's name against a number, run it through six questions. If it fails the first three, do not grade anyone on it - at most, watch it as context.
Self-studyControllability and attribution - the two that get forgotten2 min read▶
Two checklist points quietly cause most unfairness:
- Controllability - do not grade a person on something they cannot move. A support agent does not control product bugs; punishing them for complaint volume during a botched release is measuring the weather, not the person.
- Attribution - many outcomes are shared. Cartwheel's conversion rate is moved by Marketing, Product, Engineering, and Merchandising together. Making it one person's grade is both unfair and gameable. Give each team a controllable slice; watch the shared outcome together.
You will meet the attribution problem head-on in b4, when you map the Cartwheel metric tree to the teams that actually own each branch.
Diagnose and fix a bad Cartwheel metric ★ 12 min · everyone tries
Take a metric a manager actually proposed, run it through the checklist, and rewrite it into something fair to grade on.
The proposal: grade Cartwheel support agents on tickets closed per day. Score it against the six points.
Find the failures: it is an output (activity), highly gameable (close-and-reopen), and narrow (ignores whether the customer was actually helped).
Rewrite it to an outcome that resists gaming: CSAT paired with first-contact resolution - two outcomes that keep each other honest.
Keep "tickets closed" only as a workload/context signal, never as the grade.
| Check | Tickets closed | CSAT + first-contact resolution |
|---|---|---|
| Outcome not activity | ✗ output | ✓ outcome |
| Controllable | ◐ partly | ✓ mostly |
| Hard to game | ✗ very gameable | ✓ two paired outcomes |
| Holistic | ✗ narrow | ✓ quality + resolution |
Sort your metrics into the three purposes ★ 8 min · do your own
Take the numbers your team already tracks and file each one under the job it is actually for. Most confusion comes from a metric being used for the wrong purpose.
List every metric your team tracks today.
For each, mark whether it is really for steering (are we on track), development (how do I coach), or evaluation (rating and pay).
Flag any metric doing double duty as both a steering signal and a pay input - that is where honesty leaks out.
The same number, three uses, three problems. Cartwheel's weekly conversion rate is a great steering signal (spot a broken checkout fast), a fine development lens (where should Product invest?), and a terrible individual pay metric (nobody controls it alone). The number is not the problem - using one number for all three purposes is. Decouple, and it becomes trustworthy again.
Try it yourself - this week ◐ 30-45 min total
- Take three metrics your team is graded on and score each against the six-point checklist. Which fail the first three?
- Rewrite one failing metric into a fairer outcome-based version, and note what you would keep it as context.
- Sort your team's metrics into steering / development / evaluation. Circle any doing double duty with pay.
- Find one metric your team is graded on that it does not actually control. Decide what to do about it.
- Optional: read the HBR piece "Use OKRs to Set Goals for Teams, Not Individuals" before b2.
Sources covered (cited)
Taught from the goal-setting and fair-measurement canon. This session covers:
Three questions before you go 🎯 ◐ 90 seconds
1 · What is the single most common cause of broken performance systems?
Separate the three purposes. When a steering or stretch metric also drives pay, people defend the number instead of doing the work, and the data stops being honest.
2 · Which rung of the ladder should you grade people on?
Activity and output are easy to count and easy to fake. Impact is real but shared and slow. Outcomes are the fair, gradable middle: the result that changed and the person influenced.
3 · "Tickets closed per agent" fails the checklist mainly because it is...
It is an output, highly gameable (close-and-reopen, rushing customers), and narrow. Pair CSAT with first-contact resolution to grade the outcome instead.