ingridleiria OKR Planning
Goal systems fail in two ways, and both are visible from the artefact alone. The first is volume: fourteen objectives, each defensible, none prioritised, so every team optimises for whatever their manager last mentioned. The second is disguise: goals written as projects, so the quarter ends at 78 percent completion with every deliverable shipped and no number in the business different from where it started.
The cost compounds. A quarter is spent, the review produces a conversation about effort rather than outcome, and the organisation learns that goals are a reporting exercise. Once that is learned it is hard to unlearn, because the next honest attempt is read as more of the same. A third failure follows: the scoring meeting becomes an argument about the number itself, because nobody agreed in advance which system the number comes from. This skill produces a goal set where each key result could only be achieved if the outcome actually changed, and where scoring is arithmetic rather than negotiation.
When to use this, and when not to
Use it when a period's priorities have to be set and measured; when an existing goal set needs repairing because it is too long, unmeasurable, or unowned; when company priorities have to reach teams without being copied downward; when the quarter has to be scored and the carry-over decided; or when someone needs help converting a strategy into things that can be checked.
Do not use it to decide where the company is going in the first place. That is strategic-plan-and-action-plan, which sets the destination and the bets, and which must exist, at least in the leadership team's head, before objectives can be written. Objectives written without an agreed direction are a popularity contest between functions. Do not use it to build the budget or sequence hiring; that is annual-planning-and-headcount, which decides whether the goals are affordable and staffed. Do not use it to design the meetings where goals are reviewed; that is operating-cadence-design. Do not use it for weekly reporting against goals, which is weekly-status-update, or for individual performance and compensation, which OKRs should be kept away from, because tying scores to pay produces sandbagged targets within one cycle.
The boundary: strategic-plan-and-action-plan says what has to be true. This skill says how we will know, by when, and who is holding it.
What you need before starting
The direction, as bets or priorities already agreed. Missing: stop and get one sentence per priority from whoever decides, in writing. Half a day here saves a planning cycle, because objectives written against an unstated direction get relitigated at the review.
A baseline for every candidate metric. A target with no current value is a wish and makes the score meaningless. Missing: the first key result of that objective becomes measuring it, with a date inside the first three weeks.
The named system each number comes from, and who owns that system. Missing: agree a source before the objective is written. An unnamed source guarantees an argument at scoring, and the argument always arrives when the number is disappointing.
The period, with dates for cascade, mid-quarter review and scoring. Missing: propose them and put them in calendars in the same session.
Last period's scores and what happened to the carry-overs. They show whether targets are being sandbagged, whether anything ever gets closed, and which teams write projects instead of outcomes. Missing: ask what last quarter's goals were and what score they got; the absence of an answer is itself a finding.
A rough sense of capacity, and who decides when two teams want the same scarce person. Missing: state the capacity assumption in the document and name the likely decider provisionally. A goal set that assumes full availability of a team already 60 percent absorbed fails on arithmetic rather than ambition, and cascade stalls more often on unowned contention than on disagreement.
The method
Score and close the previous period first. Never write new goals over an unscored quarter. Score each key result on the measured number, write one line on what the score means, and decide explicitly whether each objective continues, changes, or closes. An objective that has carried twice should either be resourced properly or killed.
Draft two to four company objectives, each traceable to a bet. An objective is a qualitative statement of what must be true by the end of the period, in language a new joiner understands: "Enterprise customers renew because the product is embedded in their weekly workflow." Not a metric and not a project. If a fifth is proposed, ask which of the four it replaces; the answer reveals the real priority order.
Apply the situation test. If every key result under an objective scores 1.0, has the company's situation actually changed? If the honest answer is no, the objective is a to-do list wearing a costume. Rewrite it as the outcome those tasks were meant to produce.
Give each objective one named owner, a single person even when the work spans four teams. Two owners means nobody chases the dependency.
Write two to four key results per objective, each a number with a baseline, a target, a date, an owner, and a source. Rewrite every proposed key result through that filter and show the before and after; people learn the distinction from seeing their own sentence rewritten far faster than from a definition.
Separate outcomes from the bets that move them. "Launch the new onboarding flow" is a project. "Time to first value for new enterprise accounts falls from 21 days to 10 by 30 September" is a key result. Projects sit underneath the key result they serve. A key result with no listed project is unresourced; a project attached to no key result is either run-the-business or should stop.
Label each key result commit or stretch, and say what each label means for scoring. A commit is one the team is accountable for hitting. A stretch is one where 0.7 is a good outcome, which has to be said in advance and in writing, because saying it afterwards reads as excuse-making.
Run the cascade by question, not by copy. Publish the company objectives, then each team answers one question: what must my team make true for this company objective to happen? Never assign a company key result to a team that cannot move it. A team may hold one local objective serving no company objective where it keeps the lights on; two or more signals the company objectives missed something real.
Write cross-team dependencies into the key result and get them agreed. "Requires the data team to deliver the events pipeline by week three" belongs in the text, and that lead has to have said yes before publication. An unagreed dependency is a defence being prepared in advance.
Close the cascade within a week, then check the whole set on one page. Does the capacity exist. Do two objectives need the same person in the same weeks. Is any metric double counted so two teams claim the same win. Where a team is not ready, publish theirs late with a named date rather than delaying everyone. Then put the mid-quarter check and the scoring date in calendars: the mid-quarter check is the intervention point, and end-of-quarter scoring is only a record.
Scoring, and the conversation it should produce
Score each key result 0.0 to 1.0 on the measured number, never on effort. The objective score is the average of its key results carried with a sentence of commentary, because the average alone hides the case where one key result hit 1.0 and the two that mattered scored 0.3.
The mid-quarter check marks each key result on track, at risk, or off track, and requires the single action that would change the trajectory. That requirement is the whole value of the check: a list of amber statuses with no action attached is a status report, and by quarter end nothing can be done about any of it.
At the retrospective, ask three questions per objective and write down the answers. What do the scores mean about the business, not about the team. What did we learn that changes next quarter. Does this objective continue, change, or close. A goal system that has never closed an objective is being reported rather than reviewed.
Keep scores away from compensation. Where an organisation insists on linking them, expect targets to be negotiated downward within one cycle, and say so in advance.
Worked example
Situation. Meridian Pay, 340 people, business payments software. Going into the third quarter the goal document ran to eleven company objectives and 47 key results across seven teams. The second quarter had scored 0.73 on average, and revenue had missed plan by 9 percent in the same period. The chief executive's question was direct: how do we score 0.73 and miss.
Task. A goal set for the third quarter that would make that contradiction impossible, published within two weeks so teams had eleven weeks to work against it.
Action. The second quarter was scored properly first, which had not happened. Of the 47 key results, 29 were projects: migrate the reconciliation service, run the partner summit, hire three account executives. All 29 scored above 0.9 and averaged 0.91. The 18 genuine outcome metrics averaged 0.44. That is where the headline number came from: 29 at 0.91 plus 18 at 0.44 is 34.31 points across 47 key results, which is 0.73. That split answered the chief executive's question in one line and was the most useful output of the exercise.
Eleven objectives were cut to three, each traceable to a bet in the annual plan: that mid-market retention is the constraint on growth, that self-serve activation is the cheapest source of new revenue, and that gross margin has to hold while both happen.
The first cascade draft was a mistake and was abandoned in week one. Company key results had been assigned straight down: support was given "net revenue retention rises from 101 to 106", which they could not move, and platform was given "gross margin holds at 71 percent", which they influenced only through infrastructure cost. Both teams accepted without comment, which was the signal, because teams argue about goals they believe they own. The cascade was rerun by question. Support wrote "every mid-market account has a named second use case live within 45 days of signature". Platform wrote "compute cost per transaction falls from 0.31 to 0.24". Both were movable by the team that held them and laddered honestly upward.
One objective had no baseline. Self-serve activation was measured nowhere, three teams each had a different definition, and one counted an account as activated at creation. Rather than invent a target, the first key result became "a single activation definition agreed and instrumented by 18 July", owned by the analytics lead, with the outcome target written as a range to be fixed on that date. Publishing that was uncomfortable and was the right call; the alternative was a target nobody could verify.
Labels were applied explicitly: two commits, five stretches. Margin was a commit, because finance had already given the number to the board. The activation improvement was a stretch at a factor of two, with 0.6 stated in writing as a good outcome.
Result. Three objectives, nine key results, one page, published on day twelve. The cascade closed on day eighteen with six of seven teams complete; the seventh published four days late with a named date rather than holding up the rest. At the mid-quarter check five key results were on track, three at risk, one off track, and the off-track one got an intervention the previous format would never have surfaced: two support engineers moved off the ticket queue for six weeks. It finished at 0.7. The quarter scored 0.63 on average, well below the previous 0.73, and revenue landed within 2 percent of plan. The lower score with the better result was the point, and it was said out loud so nobody read the drop as decline.
A second scenario, where it goes differently
A 90-person industrial hardware business with a nine-month sales cycle and one product release a year. Quarterly outcome metrics barely move here, and forcing them produces either noise or dishonesty.
What changes: key results become leading indicators and verifiable milestones, labelled as such so nobody confuses the two. "Three named reference customers have completed on-site pilots by 30 November" is legitimate here in a way it would not be at a software company with a two-week sales cycle, because it is the only thing that moves inside the period. The annual objective carries the outcome number; the quarterly key results carry the evidence that the annual number is on track, and scoring follows that split.
What does not change: two to four objectives, a baseline for every number, a named source, one owner each, and a scoring date already in the calendar.
Output
One page, in this order. Anything that does not fit is an appendix.
The objective set:
| # | Objective | Owner | Bet or priority it serves |
|---|
The key result table, which is the working document for the quarter:
| Objective | Key result | Baseline | Target | Date | Owner | Measurement source | Commit or stretch |
|---|
Projects mapped to key results, so people know what to do on Monday, with columns for key result, project or bet, lead, start date and expected effect. Then the cascade instruction to teams in four lines: the company objectives, the one question each team answers, the template, and the deadline. Then the review calendar: cascade close, mid-quarter check, scoring date, retrospective.
At quarter end, the scorecard:
| Objective | Key result | Target | Actual | Score | What it means | Continue, change, or close |
|---|
Where progress is charted, keep it monochrome, distinguish series by marker shape and dash pattern rather than colour, put the legend outside the plot area, and use one accent colour only for the line that is off track.
Failure modes
Projects wearing a key result's clothes. Recognise it because the key result contains a verb like launch, migrate, run, hire, or complete. Rewrite it as the number that project was meant to move, and list the project underneath.
A target with no baseline. Recognise it because the phrasing is increase, improve, or reduce with no starting value. Make measurement the first key result, dated inside three weeks.
Fourteen objectives. Recognise it by count alone. Force the trade by asking which four you would keep if the others were deleted, then delete the others.
The cascade by copy. Recognise it when a team's key results are identical to the company's, or when a team accepts its goals without argument. Rerun by question.
Scores that rise while the business does not move. Recognise it by splitting the score into project-type and outcome-type key results and comparing the averages. The gap between them is the diagnosis.
Sandbagged targets. Recognise it when almost everything scores between 0.9 and 1.0 quarter after quarter. Check whether scores feed performance reviews, which is the usual cause, and separate them.
The eternal carry-over. Recognise it when the same objective appears for a third consecutive quarter with a similar score. Either resource it properly with something else stopping, or close it.
Edge cases
No baseline exists anywhere and cannot be built this quarter. Set a milestone key result for instrumentation and a qualitative objective, and say plainly that the outcome target starts next quarter. Do not invent a plausible baseline; a fabricated starting number produces a meaningless score and is discovered at the review.
A metric outside the team's control, such as market price or a regulatory date. Rewrite the key result as the controllable part, or track it as a risk in weekly-status-update rather than a goal.
A very small team, under about fifteen people. Skip the cascade. One set of three objectives for the whole company, with owners across it, is the right shape; two layers create ceremony without clarity.
Mid-quarter, the direction changes. Do not amend key results silently. Close the affected objective with its score to date and a written reason, and open the replacement with its own baseline and date.
The organisation has rejected OKRs by name. Use the structure and drop the vocabulary. Priorities and measures of success carry the same discipline, and arguing about the label wastes the goodwill needed for the substance.
Quality bar
- Two to four company objectives, each traceable to a stated bet or priority.
- Every key result has a baseline, a target, a date, one named owner, and a named measurement source.
- Nobody reading the set can find a project disguised as a key result.
- Every key result is labelled commit or stretch, and what each label means for scoring is written down.
- Team objectives answer what the team must make true; no team holds a company metric it cannot move.
- Every cross-team dependency is named in the key result and agreed by the team that owes it.
- The whole set fits on one page and a new joiner could explain the quarter from it.
- The mid-quarter check and the scoring date are in calendars before the set is published.
Adapting this to your context
The counts, the quarterly rhythm and the 0.0 to 1.0 scale come from venture-backed software companies of fifty to five hundred people. A starting point, not a standard.
- The quarter as the planning unit. Where the operating cycle is longer than the period, as in hardware, clinical or construction work, put the outcome number on an annual objective and keep quarterly key results as dated milestones.
- Two to four company objectives. That fits one leadership team. A group with genuinely separate divisions can hold two to four per division, provided no individual serves more than one set.
- One cascade hop. Below about fifteen people, drop the cascade. Above five hundred, add a division layer, but keep every person one hop from the objective they can actually move.
- A named measurement system. Where the number lives in a spreadsheet or a manual export, the source is a named person, a named file and a stated refresh day.
- What not to change. Every key result is a number with a baseline, a target, a date, one named owner and a named source, and the previous period is scored and closed before the next one is written.
Related skills
strategic-plan-and-action-plan sets the destination and the bets these objectives measure, and must come first. annual-planning-and-headcount decides whether the objectives are affordable and who is hired to deliver them. operating-cadence-design places the mid-quarter check, the scoring session and the retrospective into the company's rhythm. weekly-status-update reports progress against these key results between reviews, and meeting-to-decisions captures the decisions taken at the scoring and retrospective sessions. decision-memo handles a contested trade-off surfaced during the cascade, such as two teams needing the same scarce resource.