Objectives and Key Results (OKRs)
Objectives and key results (OKRs) pair an objective, a qualitative intent, with three to five numeric key results that check the movement towards it. The set holds for one bounded cycle, in practice the quarter, then closes: the next cycle starts from the position reached and states new key results, whether the previous ones were met or missed. Andrew Grove built the practice at Intel; John Doerr took it to Google in 1999 and gave it the form that spread. The technique commits a team to a change for a set period, where a key performance indicator watches an activity period after period, with no end in view.
Goal
An OKR ties a strategic intent to a team's work over a short period and makes that link checkable. Grove built it around two questions asked in that order: "Where do I want to go?", whose answer is the objective, and "How will I pace myself to see if I am getting there?", whose answer is the set of key results. The second supplies the numbers that will settle the matter at the end of the cycle.
The deliverable is one page per team per cycle. It carries an objective, then three to five key results with their starting value, their target and the source that will read them.
Metrics and key performance indicators build a permanent measurement system that nothing closes. Value modelling puts a number on what a decision brings a named stakeholder, without committing anyone to move anything and without a cadence for coming back to it. A key result often looks like a KPI, a percentage or a threshold, without being one: it shows that a commitment made this quarter was honoured.
Usage
When to use it
- An annual strategy to turn into team work: give each team a direction for the quarter and the numbers that check it.
- Teams whose work depends on each other's: make visible how one team's key result depends on the other's delivery, then negotiate it before the cycle opens.
- An ambition nobody can report on: set the three numbers that will settle it at closing.
- Prioritisation that has no dated criterion: supply one, the key result the cycle has to move.
- An autonomous team to steer: the objective sets the direction, the team keeps the choice of means.
When not to use it
- A stable activity with no change in view: nothing to move over the period, so run a permanent indicator system.
- An effect measurable beyond the cycle: no key result will move in three months, so milestone by deliverable and follow the effect with indicators.
One objective, three to five key results
The objective is qualitative, turned towards action and bounded by the length of the cycle. It fits in one sentence the team can repeat without rereading it. It leaves out the name of a deliverable: "rebuild the invoicing portal" names a piece of work, "become the invoicing solution of choice for SMEs in French-speaking Switzerland" states a direction. Doerr gives the wording template: "I will (objective) as measured by (this set of key results)".
Every key result carries a number. The line Doerr reports from Marissa Mayer is the shortest test: it is not a key result unless it has a number. Three forms fill that role: a target value reached, the number of customers going from 340 to 500; a threshold crossed and held, service availability staying above 99.5%; a binary event, delivered or not delivered, when the delivery itself is the checkable unit of progress. Naming the source lets a third party verify the number without arguing with whoever owns it.
The ceiling of five keeps the commitment readable. Past it, the team can no longer tell what counts from what was added out of caution, and the page stops acting as arbiter when the cycle opens. A team likewise carries one to three objectives per cycle.
The cycle and its reset
The quarter is the reference cadence for Grove as for Google. A cycle has four moments. Planning opens the period: the team sets its objective, negotiates its key results with its management and with the neighbouring teams, then reads off the starting values. The check-ins, short and usually weekly, update the numbers and surface what is blocked. The review closes the period: each key result gets its score and the team says what the number means. The retrospective looks at the practice itself, at how the targets were set and at what the next planning will do differently.
A key result that was met is not carried over: the next quarter states what has to move from the new position. A missed key result is not dragged from one cycle to the next either; the review asks whether the objective still deserves a quarter, at what target and with what means. That closing is what allows an uncomfortable ambition, since its cost is bounded at three months.
Committed or aspirational
Doerr separates two kinds of objective. The team declares which one it is setting. A committed objective has to be reached: it addresses a near-term operational need, and the team puts nearly all its means behind it. A team that meets every committed objective each quarter with capacity to spare has set the bar too low. An aspirational objective aims at what the team does not yet know how to reach: it is set on the understanding that it will be partly missed. At Google an average of 0.6 to 0.7 on an aspirational objective counts as healthy, and a run of 1.0 across the whole portfolio signals targets that were too easy.
At closing each key result gets a score from 0.0 to 1.0, equal to the share of the distance covered between the starting value and the target; going past the target is still scored 1.0. The average of the key results gives the score of the objective. That score judges the target as much as the work done, and it feeds the decision of the next cycle.
Alignment is negotiated in both directions
OKRs run from one level to the next. The company's objectives frame those of the teams; the teams send up key results management would not have worded, since they alone know what is reachable with the means they have. Horizontal alignment weighs as much: two teams whose key results depend on each other write that down at planning. Without that line, the dependency surfaces in the second month.
Management by objectives, formalised by Drucker in the 1950s, set objectives top down, person by person, and tied them to pay. OKRs keep the idea of written objectives and reverse those three choices: negotiated, owned by a team, decoupled from reward.
Where OKRs come apart
They are used to grade people
Grove treated the practice as a management instrument and warned against using it to sanction; Doerr picks the warning up and turns it into a design rule. This is the failure mode both sources document most. Tying the attainment of key results to a bonus or to an annual rating moves what is at stake to the moment the target is set.
Targets are calibrated to be cleared
Sandbagging follows from the previous point: everyone sets what they know they can hold, and the organisation spends a quarter confirming what it already knew. The counter is structural: decouple OKRs from pay and promotion decisions, then score honestly, misses included. The second does not survive long without the first.
The key results are tasks in disguise
"Deliver the new checkout flow" describes work; "the checkout completion rate reaches 3.4%" describes a result. A deliverable is a binary key result when it is itself the checkable unit of progress towards the objective. The defect is in the set: five key results that make up a task list dressed in OKR vocabulary, where the question "how will we know" has gone. The test: if the team can tick the line without a single business number having moved, it was not a key result.
The cascade turns top-down again
Objectives handed to teams at the start of the quarter, with no negotiation, reproduce management by objectives under new vocabulary, passivity included. The most reliable sign is on the page itself: not one key result there was proposed by the team that owns it.
AI considerations
AI earns its place first in the drafting. Given a worded objective, a model proposes candidate key results, their numeric expression and the source that would verify them, material to argue against in the planning workshop. It is reliable on form checking: spotting a key result with no number, a statement that describes a task, two teams whose key results contradict each other or rest on the same unplanned delivery. At the check-ins it consolidates the readings of several teams and drafts the commentary that goes with the numbers.
Three decisions stay human. The level of the target depends on the starting value, on the team's capacity and on what the organisation will accept risking; it is negotiated with the people who will have to reach it. Whether an objective is committed or aspirational is a promise, and a team words that promise itself. Scoring means standing behind an unwelcome number in front of one's peers, which no tool supplies. Progress readings attached to named individuals fall under the revised Federal Act on Data Protection (nFADP) and stay in a service whose hosting is known.
Examples
The artefact is the OKR page of a Lausanne publisher of online invoicing software for SMEs, read at the close of the quarter.
| Line | Start | Target | Kind | End of quarter | Score |
|---|---|---|---|---|---|
| Objective 1: hold the service promise made to SME customers | – | – | committed | – | 0.97 |
| KR1: support first-response time | 40 h | 20 h | – | 21 h | 0.95 |
| KR2: invoices issued without a support ticket | 90% | 95% | – | 96% | 1.0 |
| KR3: annual subscriptions renewed | 100 | 120 | – | 119 | 0.95 |
| Objective 2: become the invoicing solution of choice for SMEs in French-speaking Switzerland | – | – | aspirational | – | 0.65 |
| KR1: active paying SMEs | 340 | 500 | – | 452 | 0.70 |
| KR2: monthly recurring revenue (MRR) | CHF 68'000 | CHF 95'000 | – | CHF 84'200 | 0.60 |
| KR3: monthly churn rate | 4.2% | 2.5% | – | 3.1% | 0.65 |
A fourth key result was dropped from the first objective at planning, the go-live of the automatic reminder module: its completion ticks a box without saying anything about movement towards the objective, and the work behind it stayed in the backlog. The next quarter carries over none of the six lines kept: the counter of active SMEs restarts at 452, and planning asks what movement deserves the commitment of the new cycle.
Visualisations
The cycle ring reads in one direction only and has no exit: the retrospective leads into the next planning, which starts again from the position reached. The cadence written on the ring, the quarter, bounds how long the key results commit the team.
The alignment diagram reads through its three arrows, each of them a negotiation. Removing the upward arrow is enough to bring the practice back to a hand-down of objectives.
In the OKR page, the Score column says what the value columns leave unsaid: the same score is a success under an aspirational objective and a failure under a committed one.
Cost
| Phase | Level | Justification |
|---|---|---|
| Preparation | Medium | Wording an objective and putting numbers on its key results means finding the data that will verify them and negotiating the dependencies with the neighbouring teams. The first cycle costs markedly more than the ones that follow. |
| Execution | Low | A weekly reading of a few numbers, grafted onto an existing meeting, and a one-hour closing review per team. |
| Documentation | Low | One page per team per cycle. The cycle closes and nothing is kept up beyond it, apart from the history of the scores. |
Tooling
A shared page or a spreadsheet is enough for the OKRs of a few teams, as long as dependencies are settled face to face. Dedicated tools (Perdoo, Quantive, Weekdone, Workboard) and the objective modules built into work-management suites take over when the number of teams makes crossed dependencies invisible: they hold the objective tree, the weekly reading and the score history from one cycle to the next.
The number behind a key result comes from a source system, invoicing, CRM, product usage data or accounting, and the OKR tool does well to pull it in automatically: a value retyped by hand every week ends up rounded in the convenient direction. No tool repairs a badly set key result; it displays it faster.
Sources
- Andrew S. Grove, High Output Management, Random House, 1983, chapter 6 Planning: Today's Actions for Tomorrow's Output: the original practice, the two questions that give the objective and the key results, the warning against using it to sanction.
- John Doerr, Measure What Matters: How Google, Bono, and the Gates Foundation Rock the World with OKRs, Portfolio/Penguin, 2018: the wording template, the distinction between committed and aspirational objectives, the decoupling from pay, the quarterly cadence and the scoring observed at Google.
- What Matters, an OKR resource published by John Doerr's foundation: worked examples of OKR pages and answers on the calibration of targets.