OKRs and Goal-Setting
OKRs (Objectives and Key Results) are the most adopted and most cargo-culted goal-setting system in tech. Popularized by Andy Grove at Intel and spread through Google (and John Doerr's Measure What Matters), they're a genuinely good tool that most companies implement as a quarterly busywork ritual. This note is OKRs done right versus OKRs done as theater, and why the difference is entirely about outcomes versus output.
#The structure
An OKR has two parts:
- Objective: a qualitative, inspirational statement of what you want to achieve. It should be memorable and a little ambitious. ("Become the default way fans check live scores.")
- Key Results: 2 to 4 quantitative, measurable statements of how you'll know you got there. These must be outcomes, not tasks. ("Grow daily active score-checkers from 1.2M to 2M"; "lift D7 retention from 38% to 45%.")
The Objective is the destination; the Key Results are the instruments on the dashboard that tell you you've arrived.
A Key Result is correct only if it's a result (a change in user or business behavior), not a thing you did. "Ship the new onboarding flow" is not a Key Result; it's a task masquerading as one. "Increase activation rate from 60% to 75%" is a Key Result. If your team could complete the KR by working hard rather than by succeeding, it's an output. Rewrite it.
#OKRs done right vs cargo-culted
Here's the table I keep in my head, because the gap between these two columns is where almost all OKR disappointment lives.
| Dimension | Done right | Cargo-culted |
|---|---|---|
| Key Results | Outcomes (behavior/metric change) | Output (features shipped, tasks done) |
| Origin | Team co-authors them; tied to strategy | Handed down; disconnected from a bet |
| Count | One clear objective, 2 to 4 KRs | Ten objectives, no focus |
| Ambition | Stretch, ~70% is a "win" | Sandbagged to be 100% hittable |
| Cadence | Reviewed continuously, used to decide | Set in Jan, ignored till the Dec review |
| Link to comp | Deliberately decoupled | Bonus-linked → everyone sandbags |
| The roadmap | KRs are the goal; roadmap is how | Roadmap items relabeled as "KRs" |
The single most common failure is the first row: teams write their roadmap, then relabel each feature as a "Key Result." Now you have a feature factory wearing an OKR costume. Real Key Results state the outcome you're betting the feature will produce, which means you can ship the feature, miss the outcome, and have learned something. That gap between output and result is the entire point.
#My beliefs about goal-setting
Outcomes, not output. Always. This is the hill. A goal that measures activity ("ran 5 experiments," "shipped 3 features") rewards motion over progress. A goal that measures the outcome ("moved retention 7 points") rewards the only thing that matters. The whole reason Business Outcome Ownership exists as a competency is to enforce this.
Stretch, with permission to miss. Grove's original insight: if you hit 100% of your OKRs, you set them too low. I aim for ~70% as a genuine success. This only works if OKRs are decoupled from compensation. The moment a bonus rides on the number, every smart person sandbags, and you've engineered the timidity you were trying to avoid. Doerr is explicit on this and most companies ignore it.
Fewer, sharper. A team with one objective and three key results has focus. A team with five objectives has none. OKRs are a prioritization device disguised as a goals device, and the discipline is in what you leave out.
Tying OKRs to performance reviews and bonuses is the most common and most destructive mistake. It converts an honest goal-setting and learning instrument into a negotiation, where people lowball targets to protect their comp. Goal-setting and performance evaluation are different systems for a reason; fuse them and you corrupt both.
#How AI is changing it
AI is quietly improving the quality of goal-setting in a way I didn't expect. Drafting OKRs, sanity-checking that Key Results are outcomes rather than disguised tasks, and pressure-testing whether a KR actually ladders to the strategy are things a model does well. Point it at a draft and ask "which of these are output dressed as outcomes" and it'll catch the relabeled-roadmap trap better than most managers. The judgment AI doesn't replace is which outcomes are worth committing to and how much ambition to set, because that's a strategic bet tied to conviction about the business. Same pattern as everywhere in strategy: the model sharpens the artifact; the choice of what to chase stays human.
#Continue Reading
- Outcomes Over Outputs is the core belief that makes OKRs work or fail.
- The North Star Framework is an alternative and complement: one durable value metric plus its inputs.
- Business Outcome Ownership is the competency that holds you accountable to the Key Results.
- Prioritization Frameworks for OKRs as a focusing device and how they shape the roadmap.
- Calibration and Promotion for why goal-setting must stay separate from performance evaluation.