Where we are
In a1 we saw the promise and the peril of measuring people, and in a2 we separated OKRs (change) from KPIs (health) and drew the e-commerce metric tree: Revenue = Sessions x Conversion Rate x AOV. Today we take that tree and do the leadership work: decide who owns which branch. The trap you are about to design around is that the most tempting metrics - revenue, conversion rate, retention - are the ones no single team controls. Point a person's rating at a number four teams move and you have built something both unfair and easy to game. The fix is not a better number; it is a better decomposition. A note we carry through the whole course: an OKR score steers and teaches, it never sets pay - decoupling that is what keeps this honest.
Push the tree down until every branch has a name 6 min live
A goal you cannot assign is a goal you cannot fairly grade. The discipline is to keep splitting a metric into its drivers until each leaf is small enough that one team can genuinely move it - and then write that team's name against it.
LiveThe rule: keep splitting until each leaf has an owner3 min▶
Start at the top-line number and ask, at each level, "what drives this, and who could actually move that driver?" Stop when the answer is a single team:
- Revenue is owned by no one - it is the whole company's result.
- Split it: Sessions x Conversion Rate x AOV. Still shared, but closer.
- Split conversion further: qualified traffic, add-to-cart rate, checkout completion, page speed. Now each driver has a plausible owner - Marketing, Product, Engineering.
- Split AOV: items per order (Merchandising), price and bundle mix (Merchandising), upsell (Product).
HBR, "Use OKRs to Set Goals for Teams, Not Individuals" (2020), makes the leadership point directly: OKRs work at the team level because outcomes are produced by teams, not lone individuals - and they should not be wired into individual appraisal. Your job as a leader is to decompose the company goal into team-sized OKRs each team can own, not to hang one shared number around one person's neck.
Self-studyBalance the branches - do not let one metric distort2 min read▶
Decomposition tells you who owns what; the Balanced Scorecard (Kaplan & Norton, HBR 1992) reminds you not to over-index on any single branch. Their argument: a company steered by financial numbers alone will quietly starve the things that produce those numbers later. Balance four lenses so no one metric distorts behaviour:
- Financial - revenue, margin (the lagging result).
- Customer - satisfaction, retention (why they buy again).
- Internal process - speed, quality, reliability (how the work runs).
- Learning & growth - skills, capability (what keeps it going).
Applied to the tree: give each owner a metric, but make sure the set of owned metrics spans more than one lens. If every team is graded only on a financial leaf, you have rebuilt the single-number problem in disguise.
Why a shared metric has no fair owner 7 min live
Conversion rate is moved by Marketing, Product, Engineering, and Merchandising at the same time. That is the attribution problem: when a result has many causes, you cannot cleanly credit or blame any one team - so grading one team on it is both unfair and gameable. The fix is to give each team a controllable sub-lever and keep conversion as a shared North Star.
Self-studyWhy shared-metric grading gets gamed2 min read▶
When a team is graded on a number it only partly controls, two bad things happen. First, luck dominates: a great quarter for the team can be erased by another team's outage, and a lazy quarter can be rescued by someone else's win - so the grade stops measuring the team. Second, people optimise their visible slice at the expense of the whole: Marketing floods the funnel with low-intent traffic to "help conversion inputs," which actually drops conversion rate. Give each team a controllable sub-lever and the incentive lines up with the real goal instead of fighting it.
Split Cartwheel's conversion goal into owned levers ★ Worked example (Cartwheel) · 10 min
Cartwheel's leadership wants "lift conversion rate from 2.4% to 3.0%." Watch how that one shared goal becomes four fairly-gradable team OKRs plus one shared North Star.
Name the shared outcome and refuse to grade any one team on it: conversion rate 2.4% - 3.0% becomes the jointly-watched North Star.
Decompose it into drivers and match each to the team that can actually move it: qualified traffic (Marketing), add-to-cart (Product), page speed (Engineering), sell-through and in-stock (Merchandising).
Write each team a key result on its own controllable lever - the thing it can move without waiting on anyone else.
Review the North Star together weekly as shared context, and grade each team only on its lever at the quarter's end.
| Team | Controllable lever (graded) | Shared outcome (watched) |
|---|---|---|
| Marketing | Qualified sessions +15% | Conversion rate |
| Product | Add-to-cart rate 9% - 11% | Conversion rate |
| Engineering | Page load under 2.0s on 95% of views | Conversion rate |
| Merchandising | Sell-through +8%, in-stock over 97% | Conversion rate |
Decide: grade it, or watch it together? ★ Thought exercise · 8 min
The core leadership judgment call on every metric is a single fork: is this something one team controls (grade it) or something many teams share (watch it together)? Get the fork right and fairness mostly takes care of itself.
The retention trap. Cartwheel nearly graded its CX team on 90-day customer retention. But retention is moved by product quality, pricing, shipping speed, and merchandising - CX owns only the support experience. The fair fix: grade CX on a lever it controls (first-contact resolution and CSAT) and keep retention as a shared North Star the whole leadership team watches. Same ambition, honest grade.
Try it yourself - this week ◐ 30-45 min total
- Pick one shared top-line metric in your org (conversion, retention, NPS, revenue per employee) and list every team that moves it.
- Assign each contributing team one controllable sub-lever it could genuinely be graded on - the thing it can move without waiting on another team.
- Declare the top-line number a jointly-watched North Star, and write down who reviews it together and how often.
- Run each sub-lever past the Balanced Scorecard lenses - do your owned metrics span more than just the financial one?
- Optional: read HBR "Use OKRs to Set Goals for Teams, Not Individuals" (2020) before a4.
Sources covered (cited)
Taught from the goal-setting and metric-design canon. This session covers:
Three questions before you go 🎯 ◐ 90 seconds
1 · What is "the attribution problem"?
Conversion rate is moved by Marketing, Product, Engineering, and Merchandising together. Grading one team on it punishes them for others' work and invites gaming.
2 · A metric is fair to grade a team on when...
Grade the controllable lever. If you cannot name one team that fully controls it, do not grade anyone on it - watch it together as a shared North Star.
3 · What should happen to a shared outcome like conversion rate?
The shared outcome stays as context everyone watches together; each contributing team is graded on a sub-lever it controls (add-to-cart, page speed, sell-through).