Stop
Stop when evidence calls for a narrower bet
Stopping protects attention for the work that still has a credible path forward. It can mean narrowing, pivoting, or ending an investment.
Primary question
What evidence would tell you to narrow, pivot, or end this work rather than keep investing?
Why it matters
The decision behind the output
Teams rarely lack reasons to continue. Another feature, model, segment, or distribution attempt can always be imagined. Without a stopping rule, possibility becomes a shield against evidence and each weak result creates one more task. The cost is not only money; it is attention withheld from opportunities whose assumptions remain credible.
Stopping is broader than shutting a product down. It can mean removing a segment, rejecting an automation level, returning from product development to manual validation, or ending a feature whose maintenance cost exceeds what it teaches. A good stop decision preserves the learning while reducing the commitment that no longer deserves protection.
Watch for
Common mistakes
- Continuing because effort already spent feels too costly to abandon.
- Waiting for certainty instead of setting a decision threshold.
Use this sequence
A working framework
- Set the evidence threshold before more investment is made.
- Compare new evidence with the assumption that justified the bet.
- Narrow, pivot, or end the work when the threshold is not met.
From the field
A field example
An AI operations tool may repeatedly produce useful summaries while failing to enter the weekly planning decision it was built to support. If intended users read the output but still reconstruct the decision elsewhere, more summary quality may not solve the problem. The responsible move could be to narrow the product to evidence preparation, pivot toward the planning handoff, or stop if the team cannot access that workflow.
Use this threshold
The decision rule
Set the failure, narrowing, and continuation thresholds before the next investment. Stop or reduce the bet when repeated evidence contradicts the assumption that justified it, when the intended user cannot adopt the workflow under realistic conditions, or when every result requires a new explanation to keep the original plan alive.
Field Notes
Related essays
- When to Stop Adding AI Features and Narrow the Workflow
When does another AI capability strengthen the product loop, and when does it only expand scope?
- When to Stop Optimizing AI Generation Speed
When should an AI product stop optimizing generation latency and improve review, correction, or decision time instead?