पाठशाला Pathshala · संचालन Sanchālan, Operations · Lesson 06 · Start

OKRs for a team of ten

Three objectives for the company, three measurable results under each, one owner per result, and a Monday and a Friday that keep them alive. The version of OKRs that works at ten people, before the system becomes the work.

Pathshala, The Founder Library · 11 October 2026 · 8 min read

Every founder who has read Measure What Matters comes back from the weekend with a spreadsheet of fourteen objectives, forty key results and a conviction that the company now has a system. By week six nobody has opened the sheet. OKRs do not fail at ten people because the idea is wrong. They fail because the version borrowed from a company of ten thousand has a hundred times too much of everything.

This lesson is the ten-person version: what an OKR is and is not, how many to set and how to write results that are actually numbers, how to score them so the score teaches something, the two short meetings that keep them alive, and a worked quarter for a company of ten.

What an OKR is, and is not

The method comes from Andy Grove at Intel, who turned Peter Drucker’s management by objectives into something quarterly, collaborative and separate from compensation; John Doerr learned it there and brought it to Google in 1999, and his book Measure What Matters is the account most founders have read. The form is a sentence. First Round’s glossary gives it as: I will achieve this objective, as measured by these key results. The objective is qualitative and ambitious, a change worth making this quarter. The key results are the three or so numbers that would prove it happened. Doerr’s own definition puts three to five key results under each objective and a cycle of a quarter, with results graded at the end.

Two confusions kill the system early. The first is between key results and KPIs. A KPI is a health metric the company watches every week, such as the eight numbers on the [metrics page](/library/weekly-metrics-review-one-page-one-hour); it should be stable and boring. A key result is a change the company is trying to cause this quarter. Churn at two per cent is a KPI. Churn from six per cent to three per cent by 31 December is a key result. The second confusion is between results and tasks. Ship the new onboarding flow is a task; it can be done perfectly and change nothing. Seventy per cent of new accounts live within fourteen days of signing is a result, and it is the reason the task exists.

Three objectives, three results, one owner

Google’s re:Work guide says to pick just three to five objectives with around three key results each, and that is for a company with thousands of teams. At ten people the right number is three company objectives and nothing else: no team OKRs, no individual OKRs, no cascade. The company is the team. Each objective should be traceable to a branch of the [north star tree](/library/north-star-metric-and-its-input-tree): one objective usually sits on the input with the most room, one on the money, one on the thing that will break if it is ignored, whether that is hiring, infrastructure or a compliance deadline.

Each key result has exactly one owner, named next to it. Heather Doshay, who has run OKRs at several companies and describes the practice in a Heavybit interview, is specific about this and about the tooling: one owner per result, and a shared spreadsheet is enough until well past ten people. Her other advice for the first quarter is the one most founders need: the first time might be a mess, and that is fine; set a best-guess baseline, expect to be wrong about ambition in both directions, and reset the baseline at the quarter’s end.

Writing results that are numbers

A key result has four parts: the metric, the baseline, the target and the date. If any is missing it is not a key result. First Round’s rule is blunt: if it is not measurable it is not a key result. Christina Wodtke’s guide to writing them adds the test of difficulty: typically three results, set so that they are difficult but not impossible, and a useful sign is a flicker of doubt that it would take a miracle to hit all three. She also suggests spreading them across kinds of change, growth, engagement, revenue, performance and quality, so an objective cannot be met by moving one number at the expense of another.

Before the quarter starts each result gets a confidence rating of five out of ten, which Wodtke recommends as the honest starting point for a goal with a fifty-fifty chance; the rating is adjusted every single week and the movement, not the level, is what the Monday meeting discusses. A result whose confidence is nine in week two was set too low. One that is two in week four is already telling the team something the end-of-quarter grade would otherwise tell them eight weeks late.

Scoring, and what a 0.7 means

Google grades each key result on a scale of 0.0 to 1.0, with 1.0 meaning fully achieved, and the objective’s grade is roughly the average of its results. The re:Work guidance is that the sweet spot for OKRs is somewhere in the 60 to 70 per cent range: a team averaging 0.6 to 0.7 is setting goals that stretch, consistently hitting 1.0 means the goals should be raised, and consistently landing below 0.4 means either the goal or the quarter was wrong. Doerr’s FAQ distinguishes two kinds. A committed objective is a commitment and is expected to have a passing grade in full; an aspirational one is a stretch, where 0.7 is a good quarter. Label each of the three at the start, because reading an aspirational 0.7 as a failure and a committed 0.7 as a success are both ways of learning nothing.

Score by distance travelled, not by whether the target was touched. A result that moved churn from six per cent to four and a half scored 0.5, and that is information: half the problem was solvable with the quarter’s methods and half was not. Scoring is for learning, not judgement, in First Round’s phrase, and the surest way to destroy that is the mistake every guide names and most companies make within a year: tying the grade to pay. Grove separated OKRs from compensation on purpose. The moment a 0.7 costs someone their bonus every target in the company becomes a 0.6 dressed as a stretch.

An OKR on the wall is a poster. An OKR with a confidence score that changed last Monday is a management system.

The check-ins that keep them off the poster

Monday, thirty minutes, after the metrics review. Wodtke’s Monday commitments uses one document with four squares: this week’s three or four most important things towards the objectives; the month ahead, so others can prepare; confidence in each key result out of ten and whether it moved; and two health metrics the team is protecting. The document is a conversation tool. Updates take five minutes; the argument about whether this week’s priorities are the right ones takes the rest.

Friday, thirty minutes. Friday wins: every team shows something it made or moved. At ten people that is every person, two minutes each, and it is where a key result that moved gets noticed by the people who did not move it. Week seven, one hour. The mid-quarter check Google recommends, in which each result’s confidence is read against the calendar and one objective may be rewritten if the world changed; a result that is clearly dead is killed here, not carried to the grading as a 0.1. Week thirteen, ninety minutes. Grading and retrospective. Each result scored, each objective averaged, and three questions asked: what did the scores teach about the company’s ability to estimate, which result was wrong rather than missed, and what is the baseline for next quarter. Then the next quarter’s three objectives are written, in the same sitting, by people who have just been reminded what a quarter can hold.

A worked quarter

A B2B software company in Hyderabad, ten people, ₹14 lakh of MRR, 40 paying accounts, 6 per cent monthly logo churn. The founders set three objectives for October to December. One, aspirational: customers stay because the product works. Churn from 6 per cent to 3 per cent; accounts live within fourteen days of signing from 50 to 90 per cent; support tickets per account per month from 2.1 to 1.0. Two, aspirational: the second channel exists. Paying accounts from 40 to 70; accounts acquired through partners from 0 to 12; CAC on the partner channel under ₹25,000. Three, committed: the books survive diligence. Monthly close by the fifth working day for all three months; GSTR-2B reconciled before every 3B; data room complete by 15 December. Nine results, nine owners, one sheet.

In week thirteen the grades are 0.63, 0.47 and 1.0. Objective one is in the sweet spot. Objective two is short, and the retrospective finds the partner result scored 0.25 because two of the three partners signed in December and had no time to sell; the goal was right and the quarter was wrong, so it rolls forward with a revised baseline. Objective three delivered, as a committed objective must, and is retired because the close is now a habit and no longer a change. Next quarter’s three are written in the same ninety minutes. Total time spent on the system across the quarter: about thirteen hours of Mondays, six of Fridays and four at the ends. Less than a day a month, for a company that knew every week whether it was on course.

The quarterly ritual, in dates

Week zero: three objectives, nine results, each with metric, baseline, target, date, owner and kind, in one sheet everyone can edit. Every Monday: the four-square, confidence out of ten re-rated, thirty minutes. Every Friday: wins, thirty minutes. Week seven: the mid-quarter check, one hour. Week thirteen: grading and retrospective, ninety minutes, and next quarter’s week zero in the same room. Put all of it in the calendar now as part of the [operating cadence](/library/founders-operating-cadence). Then check one thing at the end of the first quarter: did the confidence scores actually change from week to week? If they sat at five for thirteen weeks the system was a poster, and the second quarter starts with that conversation.

The figures in the worked quarter are illustrative. The sources are below; Google’s re:Work guide and Wodtke’s two short posts between them give the whole method in twenty minutes.