पाठशाला Pathshala · विचार Vichār, The idea · Lesson 15 · Build
The pivot decision: evidence, not exhaustion
Most pivots are decided late at night by tired founders or never decided at all. Write the metric and the date that would make you change course now, while you can still judge.
Pathshala, The Founder Library · 11 October 2026 · 7 min read
There are two ways to get the pivot decision wrong. One is to keep going for two years past the point where the numbers stopped moving. The other is to change direction in a bad week because the founders are tired. Both are the same mistake: deciding at the moment you are least able to judge.
This lesson moves the decision to a calmer day. It shows how to write a kill criterion, a metric and a date fixed in advance, how to measure progress against it every week and what to do when the date arrives. The figure draws the criterion as a line and the business as another, so the gap is visible long before the date.
Why the decision comes too late, or too early
Paul Graham’s How Not to Die of August 2007 makes the case for persistence. When startups die, he wrote, the official cause is running out of money or a founder leaving, but the underlying cause is usually that the founders became demoralised. Startups rarely die in mid keystroke. Many companies that would have worked were abandoned by tired people a few months short of the turn.
Annie Duke, the former professional poker player who now writes on decision-making, makes the opposite case with equal force. In a 2022 interview on her book Quit she lists the forces that keep people going past the point of sense: sunk costs, the habit of treating goals as pass or fail, the tendency to escalate commitment to a failing cause and, hardest of all, identity, because the hardest thing to quit is who you are. She quotes Daniel Kahneman’s observation that the worst time to make a decision is when you are in it.
The two arguments do not conflict. Graham is warning against quitting from mood. Duke is warning against continuing from mood. The remedy for both is to take mood out of the decision by deciding what the evidence would have to show before you see it.
Kill criteria: a state and a date
Duke’s form is simple. Write down, in advance, a state you must be in and a date by which you must be in it, and pre-commit to the action you will take if you are not. Her model is the turnaround time on Everest, fixed before the climb starts, because nobody near the summit can be trusted to judge how much daylight is left. The founder version: if by 31 March we do not have 150 paying clinics with at least 70 per cent of February’s customers still paying in March, we will change the customer, the problem or the product.
Three properties make a criterion work. It is a number, so it cannot be argued with at two in the morning. It has a date, so progress cannot be deferred. And somebody other than the founders knows it, so it cannot be quietly moved. A co-founder, an investor or a founder friend who will ask on the date is enough.
Choosing the metric and the date
Pick one metric customers control. Paying customers, retained accounts, repeat orders, revenue. Not sign-ups, demos booked or features shipped, which measure the founders’ effort rather than the customers’ verdict. The metric should test the riskiest assumption of the business. If the risk is that nobody stays, the criterion is about retention; the [retention curves lesson](/library/retention-curves-and-flattening-test) shows what flattening looks like.
Set the state from what the next stage needs. Ask what level of the metric would make the business fundable, or self-funding, at the next step. If a seed round in your category needs about ₹1 crore of annual revenue and the price is ₹30,000 a year, the state is around 330 paying customers. Work backwards to the state the date requires, not forwards from where you happen to be.
Set the date from the runway. Leave at least six months of [runway](/library/runway-how-many-months-you-really-have) after the date, because a pivot needs money and time to test the new direction. A date that falls in the last month of cash is not a decision point. It is a funeral.
Match the horizon to the stage. Lenny Rachitsky studied forty pivots in May 2024. Pivots of an idea before launch generally happened within three months. Hard pivots, of companies with a live product and users, generally happened within two years of launch and most around the one-year mark. A first criterion of three to six months for an idea and six to twelve for a launched product follows the same rhythm.
A worked example: forty pharmacies in Kochi
A team in Kochi sells inventory software to independent pharmacies at ₹2,500 a month. Nine months after launch they have 40 paying stores. They write their criterion on 1 October: 150 paying stores, with at least 70 per cent of each month’s stores still paying the next month, by the end of March, 26 weeks away. They tell their angel investor and agree to meet on 31 March whatever happens.
The arithmetic is in the figure. To go from 40 to 150 in 26 weeks the business needs about 5.2 per cent growth a week, compounded. Over the past eight weeks it has grown about 3 per cent a week. At that slope it lands near 86 stores on the date, a little under 60 per cent of the target. The founders know this on 1 October, not on 31 March, and they have 26 weeks to change the slope or to prepare the pivot.
Decide what would make you pivot while you are rested. Then let the date make the call.
Look at what moves the line. At 5 per cent a week the team clears the target. At 3 per cent they need roughly 45 weeks, which is more runway than they have left after the date. The figure does not tell them what to do. It tells them, every Monday, how far the business is from the promise they made to themselves.
What happens on the date
The date triggers one of three actions, and only three.
Persevere. The state is met, or nearly met with a slope that will clear it within weeks. Write the next criterion for the next stage and carry on.
Pivot one element. Change the customer, the problem or the product, and keep what the evidence supports. The pharmacy team might find that chains with four or more stores renew at twice the rate of single stores, and pivot the customer while keeping the product. PostHog, whose founders pivoted several times before finding their product, puts it bluntly in its founder newsletter: pivots should be large and should feel scary and uncomfortable, and if the new idea resembles the old one too closely the change of direction should be sharper. It also advises a time limit on the decision itself.
Stop. If no element has evidence behind it, return what money is left, look after the team and write down what was learnt. Stopping on a date you chose is a decision. Stopping when the bank balance chooses for you is not.
A fourth option tempts every team: extend the date. Allow it once, with a new number written before the old date passes and the same outsider told. A criterion that moves twice is no longer a criterion.
Telling evidence from exhaustion
Between dates the founders will still feel the pull to pivot. Use four questions to tell a signal from a bad week.
Has the number changed, or has the mood? If the weekly slope is the same as last month and the urge to pivot is new, the urge is about the founders. Sleep, take two days off and look again.
Is it true across cohorts? One bad month of sign-ups is noise. Two successive cohorts that retain worse than the ones before is a signal.
What are customers saying when they leave? The signals in Rachitsky’s study were persistent lukewarm interest and the realisation that the idea was never going to be as big as the founders thought. Lukewarm is what you hear from churned customers when the problem is real but mild. Ask the last ten who left; the [twenty interviews lesson](/library/twenty-interviews-to-one-decision) gives the method.
Would a stranger with the same numbers keep going? Describe the business to a founder friend without telling them it is yours. Their answer is usually the one the evidence supports.
A monthly ritual: the criterion review
Keep the criterion in a single line at the top of the weekly metrics sheet: the metric, the state, the date and the slope it needs. Every Monday enter the actual number and the eight-week slope beside it. On the first Monday of each month, spend thirty minutes with your co-founder on three questions. Is the slope above or below what the criterion needs? Has anything in the evidence, not the mood, changed? And if the date were today, which of the three actions would you take?
Write the answer down. When the real date comes the decision will already have been made four or five times, in daylight, by people who had slept. That is the whole point.
The worked example is illustrative. Nothing here is investment advice. Sources were checked in October 2026.