Should You Rebuild or Shut Down? How Notion Cut Its Team to Two and Started Again
Short answer: Rebuild only when evidence still supports the problem and product shape, the failure sits in a replaceable system, and the reduced team has enough cash for one complete test plus a closure reserve. Write the launch date, success threshold and stop rule before starting. Shut down when the next attempt depends mainly on hope, another broad reinvention or money you cannot afford to lose.
Founders often frame persistence as courage and closure as failure. That is emotionally convenient and financially dangerous.
A rebuild is not automatically persistence. It can be a disciplined decision to preserve validated insight while discarding technology, scope or cost. It can also be a way to postpone admitting that customers do not care.
Notion preserved the idea and replaced the system
Notion started in 2013. Ivan Zhao later wrote in a company retrospective that it saw signs of traction only with the 2018 Notion 2.0 launch. During those lost years, the team rebuilt the product four times. One version had what Zhao considered the right product but the wrong technical foundation.
In a later interview about the refounding, Zhao said the first version took four or five years to find product-market fit and the company was running out of money. He and co-founder Simon Last decided to lay off the rest of a group Zhao described as roughly five people and rebuild alone in Kyoto.
They rented out their San Francisco home and office, which Zhao said made them cash-flow positive during the period. The rebuild from scratch took another year and a half. A separate Notion account of its Japanese history dates the move to 2015 after the first iteration failed.
Notion's later success does not prove that long rebuilds are wise. Survivorship bias hides the teams that made similar sacrifices and still closed. What makes the case useful is the distinction: the founders remained committed to the product thesis, identified the technical stack as a constraint and radically lowered the cost of another attempt.
My view is that a rewrite should be presumed wrong until you can name evidence the existing system cannot exploit. Customers do not pay for cleaner architecture. They pay for an outcome that the architecture must enable.
Use the Rebuild Runway Decision Tree
The Rebuild Runway Decision Tree asks four questions in order.
| Question | Rebuild evidence | Shutdown evidence | |---|---|---| | Is the problem real? | Repeated use, payment or costly workaround | Praise without behaviour | | Is the proposed product shape clearer? | Specific job and retained behaviour | Another broad vision | | Is the failure replaceable? | Technical, scope or delivery constraint | Weak need or unreachable buyer | | Can one full test be funded? | Runway plus closure reserve | Rebuild consumes the last cash |
A “no” on the first two questions should normally stop the rebuild. A “no” on funding means reduce the test further, raise committed money on honest terms or close.
Diagnose the failure before changing code
Separate four failure types: problem, proposition, system and economics.
A problem failure means customers do not experience enough pain. A proposition failure means the problem exists but your promised outcome is weak or unclear. A system failure means customers want the outcome but the current product cannot deliver it reliably. An economic failure means delivery works but costs more than customers will support.
Only the third clearly points towards a rebuild. The second may need a narrower version. The first usually demands a different problem, not a rewrite. The fourth requires price, cost or model change alongside any technical work.
Use observed attempts, payment, retention, failures and interviews about real behaviour. Founder conviction is not customer evidence.
Define what survives
Write a one-page preservation decision. State which customer, job, behaviour, distribution asset, data and brand remain. Everything else must justify its place.
Notion retained its broad mission and what Zhao described as the right product shape while replacing its technical foundation. Your preserved insight should be more concrete than a mission. For example: “Recruitment teams repeatedly pay to search candidate conversations, but the present database cannot return results within the required time.”
If you cannot articulate what has been learned, the rebuild is simply another start with less money.
Price the entire attempt
Calculate monthly founder living costs, employee and contractor cost, infrastructure, customer obligations, tax, professional fees and the cost of closing. Include contingency for technical work taking longer than expected.
Keep a closure reserve outside the rebuild budget. It may need to fund final payroll, refunds, supplier bills, data handling and professional advice. Company insolvency, redundancy, director duties, tax and customer obligations vary by jurisdiction. If the company may be unable to pay debts, obtain qualified local legal and insolvency advice immediately.
The article on [how much money you need to start a business](/business-finance/how-much-money-do-you-need/) explains why the lowest bank balance matters more than a simple total budget.
Worked example: Workboard
Workboard sells project software to small architecture practices. It has £92,000 in cash and £6,000 monthly recurring revenue. These figures are illustrative.
The current eight-person team costs £48,000 monthly including employment on-costs and infrastructure. Gross contribution from customers is £4,800, so net burn is £48,000 - £4,800 = £43,200. Runway is only £92,000 divided by £43,200 = 2.1 months.
Forty of 55 paying firms use one workflow weekly, but exports fail on large projects and 18 firms have complained. Twelve have stated in writing that they will renew if performance is fixed. That is system evidence, not proof, but it is stronger than general enthusiasm.
The founders consider reducing to two founders plus one engineer. Monthly cost would be £17,000. Contribution remains £4,800 initially, so burn becomes £17,000 - £4,800 = £12,200.
They reserve £22,000 for redundancy, refunds, professional fees and closure. Rebuild cash is £92,000 - £22,000 = £70,000, giving £70,000 divided by £12,200 = 5.7 months.
| Decision element | Rule | |---|---| | Preserved job | Weekly project coordination for architecture practices | | Replaced constraint | Export and data architecture | | Launch deadline | End of month four | | Success by month five | 45 retained firms, 10 new paid firms, £7,500 monthly revenue | | Stop rule | Miss launch by 30 days or retain fewer than 38 firms |
At £7,500 revenue with an 80% contribution margin, contribution would be £6,000 and burn £17,000 - £6,000 = £11,000. The rebuild would still not be break-even. It would merely earn time for a subsequent growth decision. Workboard proceeds only because the reserve protects an orderly closure and the test has a fixed end.
Reduce scope and cost together
Cutting staff while preserving the same roadmap creates an impossible workload. Remove customer segments, features, integrations and channels that do not support the preserved job.
Communicate honestly with affected employees and customers. Do not imply certainty about the rebuild. Meet contractual and statutory obligations, protect access to data and avoid taking long-term customer money that the company may not be able to serve.
A smaller team can move faster only when the decision surface is also smaller.
Set a market gate, not a coding milestone
“New version shipped” is not success. The gate should contain customer behaviour and economics: retained paying accounts, completed core jobs, price, support burden and contribution.
Use a deadline and a grace rule. A short, predefined extension may be sensible for a clear technical delay. Repeated extensions convert the stop rule into theatre.
Track leading evidence weekly, but do not keep changing the final threshold. If the market gate is missed, close, sell the assets or choose another outcome allowed by your obligations.
Decide within ten working days
Days one to three: reconcile cash, liabilities, customer commitments and legal obligations. Freeze discretionary spending. Days four to six: interview retained and lost customers about real use, then classify the failure.
Days seven and eight: define what survives, price the smallest rebuild and protect a closure reserve. Day nine: write the market gate, deadline and stop rule. Day ten: decide and communicate.
If you proceed, review cash every week and customer evidence every fortnight. Do not wait for the rebuild to finish before discovering that the retained assumption was wrong.
Related guides
Frequently asked questions
How much runway is enough for a rebuild?
Enough to complete the work, run one meaningful customer test and still meet closure obligations if it fails. Add contingency rather than using the developers' best-case estimate. A six-month plan with six months of cash has no safety margin. Calculate runway from net monthly cash use after realistic retained revenue, not current revenue if customers may leave. Keep redundancy, refunds, tax and professional costs reserved. If solvency is uncertain, director duties and options depend on jurisdiction, so obtain qualified insolvency and legal advice promptly.
When is a complete rewrite justified?
When customers repeatedly seek and pay for the outcome, the current system creates a specific barrier that incremental repair cannot remove, and the proposed new architecture has a clearly bounded route to that outcome. A rewrite is not justified by developer frustration, fashionable technology or untidy code alone. Compare repair, replacement and managed reduction carefully. Ask exactly what customer behaviour changes when the rewrite ships. If the answer is merely “the product will be better”, the scope is not decision-ready.
Should I lay off staff to extend runway?
Only as part of a lawful, funded plan that reduces scope and has a credible market test. Reducing people while retaining all commitments transfers impossible work to those who remain. Price redundancy, notice, holiday, benefits, morale and knowledge loss before assuming savings begin immediately. Employment requirements vary by country and circumstances. Obtain qualified local employment and insolvency advice, communicate directly and meet every material obligation. A longer runway has little value if the remaining team cannot deliver the narrower outcome.
Can customer promises justify another attempt?
Promises help only when tied to behaviour: continued payment, a signed renewal, paid pilot, data access or a specific commitment conditional on the fix. General encouragement is weak because customers do not bear your rebuild cost. Record what each customer does now, the failure they encounter and what they will do when it is removed. Discount evidence from close relationships appropriately. Base the decision on a cohort, not one large prospect whose requirements could turn the rebuild into bespoke work.
What if the founders disagree about shutting down?
Return to written evidence, cash obligations and the pre-agreed decision rule. Each founder should state what fact would change their view and what personal risk they can accept. Review shareholder agreements, director responsibilities and employment positions with qualified advisers. Do not resolve disagreement by quietly spending until only one option remains. If one founder wants to continue alone, price asset ownership, liabilities, intellectual property and customer commitments formally. A clean disagreement handled early is safer than an informal handover during insolvency.
How do I close without abandoning customers?
Immediately stop taking commitments you may not fulfil, communicate dates plainly, provide lawful refunds or continuity options and preserve access to customer data for the required period. Pay staff, tax and suppliers according to legal priority and professional advice. Document system shutdown, privacy and security responsibilities carefully. Do not promise indefinite support from a company that is closing. The precise process depends on structure, contracts, solvency and jurisdiction. Qualified local legal, accounting and insolvency support is necessary for material decisions.
Comments (0)