DEEP READING · READ 915
Measure What Matters · John Doerr · 2018
The whole book is one sentence template — "I will ______ as measured by ______" — plus the discipline built around it: the Objective says where you're going, the Key Results say in numbers how you'll know you got there. But the real argument isn't "set goals." It's set few goals, set them in public, set them high enough to fail, and don't pay bonuses on them — miss any one of those four and OKRs decay into a quarterly essay assignment.
John Doerr was an engineer and salesman at Intel in the 1970s, where he watched Andy Grove run the goal system Grove had built; he joined the venture firm Kleiner Perkins in 1980, invested in a very young Google in 1999, and that same year gave the first OKR lecture to a company of about three dozen people. So this 2018 book isn't research — it's the testimony of a carrier: the man who moved Grove's method into Silicon Valley, backed by two decades of case studies from his own portfolio. It is the mirror image of Grove's High Output Management (already covered here as read905): Grove wrote the mechanism; Doerr wrote what happened after it escaped into the world.
Grove split goal-setting into two questions: where do I want to go? (the objective) and how will I know I'm getting there? (the key results). On paper: "I will ______ as measured by ______."
An Objective is qualitative, directional, ideally something people want to get out of bed for: "make our search the fastest in the category." Key Results — usually about three — are quantitative, dated, and adjudicable by a stranger: "median time-to-first-paint from 1.4s to 0.8s," "P95 latency under 2s." Marissa Mayer's test is the cheapest one available: "It's not a key result unless it has a number."
Here is the pit every beginner falls into: writing the work as if it were the result. "Re-architect the index service" is an activity. You can finish it and have improved nothing. A key result has to name the change the activity was supposed to produce.
| Written as activity (wrong) | Written as outcome (right) |
|---|---|
| Re-architect the index service | Search P95 latency from 2.6s to 1.2s |
| Run a customer satisfaction study | NPS from 22 to 35 on ≥ 500 responses |
| Improve onboarding for new hires | New hires shipping their first change solo by day 90: 40% → 80% |
| Ship 3 new features | 3 features shipped, each with < 2 severe defects in its first 30 days |
That last row demonstrates one of the book's sharpest devices: paired key results — one for quantity, one for quality. Score only "features shipped" and you'll get a pile of half-built ones; chain a defect count to it and speed has to drag quality along. Wherever a single metric stands alone there is a shortcut that games it; the paired metric closes the shortcut.
OKRs aren't new. Peter Drucker proposed Management by Objectives (MBO) in 1954: manage by agreed goals rather than instructions. It ran for two decades and then broadly stopped working, because in most companies it became annual paperwork — set once a year, issued downward, wired straight into the performance rating and the raise. So everyone learned the same craft: pitch this year's target slightly above last year's number and collect full marks.
Grove changed three things at Intel, each aimed at a load-bearing joint:
Why these three matter: goal-setting never failed because setting goals was wrong; it failed because the people setting goals were paid to set easy ones. Grove's redesign converts the system from an evaluation instrument into a navigation instrument — something that tells you mid-route that you're off course, rather than grading you at the finish.
This is the book's most persuasive case because Doerr lived it. In 1979 Motorola's new processor outperformed Intel's 8086 and Intel's salesforce was losing accounts. Within weeks Grove launched a company-wide campaign codenamed Operation Crush — and expressed the whole thing as OKRs:
What matters is how that fourth item reached individual people: the company objective was decomposed into division, region and finally each salesperson's quarterly OKRs. Within two weeks the slogan "beat Motorola" had become a number on every rep's desk: how many design wins I personally close this quarter. Intel held the line — and one of those design wins was the IBM PC choosing a derivative of the 8086, a single decision that shaped the next twenty years of the industry.
What the case actually proves isn't that OKRs make people try harder; it's that OKRs let a strategic turn penetrate to the edge of the organisation in two weeks. Big companies rarely fail because leadership doesn't know which way to turn. They fail because the turn signal attenuates on the way down and nobody's daily work changes. OKRs are the machine that forcibly translates that signal into a number in each person's hands.
Doerr organises the benefits into four "superpowers." The first is focus and commitment, and the rule is plain: at most three to five objectives per cycle, at most three to five key results each.
The cap is the whole point. Any organisation can produce twenty important things, and producing twenty is identical to producing none — the resources didn't grow, so twenty means "carry on as before." OKRs hurt precisely because they force you to write down, explicitly, that something genuinely important is not happening this quarter. Doerr's test: if nothing on the list stings, you haven't chosen yet.
How it changes the way you work: at the next goal review, don't ask "is this important?" — everything passes that. Ask "if we could keep only three, which three?" Priority is exclusionary; a priority list that excludes nothing is a wish list.
At Google, everyone's OKRs are visible company-wide by default, from CEO to intern. That isn't a cultural gesture; it does three specific jobs:
One word here is badly misunderstood: cascading. Many companies read OKRs as a strict tree — each of the CEO's key results split into division objectives, then team objectives, down to individuals. Doerr's position is that pure top-down cascading is slow and brittle: slow because everyone waits for the level above, brittle because one change at the top invalidates the whole tree — and nobody feels much about a goal they were assigned. The healthy version is mixed: leadership supplies direction and a few genuinely company-level objectives; teams propose how to attack them and align upward.
OKRs aren't written in week one and filed. The cadence Doerr wants is a short look at the progress bar every week or two, and at quarter's end a 0-to-1.0 score on each key result (0.7 means you got 70% of the way).
The more useful piece is his four mid-quarter options. For every key result running behind, make an explicit decision: continue (it's working, push harder), update (conditions changed — move the number or the date), start (the current approach isn't working; open a new objective mid-flight), or stop (this no longer matters; formally delete it).
"Stop" is the most underrated item in the entire system. In most organisations goals only accumulate, never disappear, so the list grows longer and less honest until everyone has learned to ignore it. A goal that can be formally deleted is the only kind that was ever taken seriously. As for the scores themselves, Doerr is unambiguous: they exist to start a conversation, not to hand out rewards — a low score earns its keep by forcing the question "where exactly was our judgment wrong?"
Google splits OKRs into two explicit classes, and the most common real-world wreck is treating them as one thing.
| Committed | Aspirational / moonshot | |
|---|---|---|
| Typical content | Ship on date, quarterly revenue, SLA uptime | Rethink the thing at ten times the scale |
| Expected score | 1.0 — anything less is a problem | ~0.7 on average counts as success |
| When missed | Requires a post-mortem: resources, judgment or execution? | Normal, provided you learned something real |
| Resourcing | Fully funded, top priority | Often scraped together |
| Cost of failure | A real loss to the business | Tuition |
Why deliberately keep a class of goals you expect to miss? Because the goal a person sets determines which solutions they're willing to consider. "Grow 10% this year" is always answered by tuning the current approach; "make it ten times bigger" declares the current approach dead and forces a search for a different road. That's the point of the odd 0.7 rule: if failure is punished, nobody will ever set the kind of goal that is worth failing at.
The book's best example is YouTube. The company had been steering by views and switched its top-level metric to watch time, then set a target that looked absurd at the time: one billion hours of watch time per day. That was roughly a tenfold jump; it took a bit over four years and landed in 2016. The real lesson isn't that they hit it — it's that choosing the metric was itself the strategic decision: once it became watch time, recommendations, creator incentives and product design were all rewritten around it. (That is also the bill the criticism section below has to settle.)
Intuitively, goals ought to drive bonuses — otherwise why take them seriously? Doerr's answer, after Grove's: the moment you connect them, you stop evaluating what people achieved and start evaluating what they were willing to write down. The mechanism is trivial. People estimate their own odds when they draft a target; if 0.7 means 70% of the bonus, the rational move is to write only what you're 90% sure of. A system designed to extract ambition gets tuned, in reverse, into a machine for manufacturing safe goals.
Doerr's fix is to separate the two jobs. OKRs align direction and expose problems. Pay and promotion come from a fuller judgment — the quarter's conditions, the difficulty of the assignment, contribution to others' work, peer input — in which OKRs are one input, not a conversion formula.
How it changes the way you read a company: to find out whether an organisation's OKRs are real, ask one question — does missing them affect my review? If yes, the goals are safe bets, however heroic the slide deck looks. Goals you dare to set and goals you dare to be judged on were never the same object.
The second half of the book turns to a different target: the annual performance review is obsolete. A once-a-year, backward-looking score wired to a raise can't correct anything in time, and it makes people perform for a single rating all year. Doerr's replacement is CFRs — Conversations, Feedback, Recognition: break evaluation apart into continuous one-on-ones, two-way feedback whenever it's relevant, and public, specific recognition (not "great work everyone," but naming who did what).
The relationship is easiest to see as a container and its contents: OKRs are the vessel, CFRs are the water. OKRs alone give you precise numbers and a workforce with no idea how they're doing; CFRs alone give you a warm team walking in different directions. Doerr's worked example is Adobe, which scrapped annual ratings around 2012 in favour of recurring "check-in" conversations, requiring managers to give feedback continuously rather than settle accounts in December.
Why this section isn't a bonus chapter: everything OKRs are good for happens after a gap is detected — and everything you have to do then (re-prioritise, admit the approach was wrong, move resources) happens in conversation. OKRs without CFRs are just a report that gets refreshed once a quarter.
The book is a short causal chain demonstrated by thirty-odd case studies:
So what does it establish? Strictly, it does not establish that OKRs make companies succeed (see below). What it does establish is humbler and real: when an organisation is forced to compress "what we are doing" into a handful of sentences that can be judged true or false, and then publish them, a great deal that survived on vagueness stops surviving — who is duplicating whom, whose goals have nothing to do with the company's, which strategy never actually landed on anybody's desk. All of it surfaces in one meeting. The main output of OKRs isn't the completion rate; it's that exposure.
1. The entire book is one template: "I will ______ as measured by ______" — the objective gives direction, the key results give evidence, and neither half works alone.
2. If it has no number it isn't a key result; and "re-architect the service" or "run a study" aren't results at all, just to-dos in a new outfit.
3. A lone metric will always be gamed, so write them in pairs — one for quantity, one for quality. "Ship 3 features" must stand next to "fewer than 2 severe defects."
4. Drucker's MBO died of being annual and tied to appraisal; Grove changed three things — quarterly cycle, half bottom-up, unhooked from pay — and it lived.
5. Three to five objectives per cycle: the function of OKRs is refusal, not coverage. A priority list that excludes nothing is a wish list.
6. Publishing them isn't a cultural gesture but three mechanisms: no duplicated work, dependencies visible early, private intentions converted into public commitments.
7. Every lagging key result needs an explicit mid-quarter decision — continue, update, start, stop — and a goal that can be formally deleted is the only kind anyone took seriously.
8. Committed OKRs are answered at 1.0, aspirational ones at 0.7; 0.7 has to be allowed because if failure is punished, nobody sets a goal worth failing at.
9. To test whether a company's OKRs are real, ask one question: does missing them affect my review? Goals attached to bonuses are always safe bets.
10. The warning worth keeping isn't in the book but on its reverse: OKRs work, and they reshape an organisation into the shape of the number you wrote down. Wells Fargo wrote "eight products per household"; YouTube wrote "one billion hours." Both delivered. So what you write down is a moral choice.