Sandhya Indurkar

Math, Applied

Only the Winners Stay Visible: Survivorship Bias in Real Decisions

Survivorship bias: visible winners vs full cohort

The idea

Survivorship bias is selection bias after the fact. The dataset keeps the winners and drops the failures. Startup headlines, five-star reviews, and top-quartile fund tables all describe the survivors, not everyone who started.

Selection bias asks who never entered the sample. Survivorship bias asks who left before you measured. Both distort the rate. This one is especially common when success stories are the only stories you can still see.

Survivorship bias answers: Who failed or disappeared, and would the headline still hold if they were counted?

Example: who is still in the dataset?

Compare the visible survivors with the full cohort that started. Failures that dropped out can make success look far more common than it was.

Are most new startups still thriving?

Survivors

88%

n = 120

Full cohort

34%

n = 1,000

Gap

+54 pts

survivors vs everyone

88%Survivors34%Full cohort

106 successes visible vs 340 in the full cohort at the true rate

Who dropped out: Startups that shut down or never got press coverage. You only hear from survivors. The visible cohort looks healthy while most launches quietly fail. The survivor slice overstates success by a wide margin. Ask what dropped out before you copy the playbook.

The math

What the headline shows

survivor rate = successes among visible cases ÷ visible cases

If 88% of still-operating startups look healthy, that rate ignores the ones that shut down. The numerator and denominator both exclude failures.

What actually happened

cohort rate = successes in full cohort ÷ everyone who started

Include launches that closed, funds that stopped reporting, and users who churned before leaving a review. The true rate is often far lower than the survivor rate.

Size of the distortion

survivorship gap = survivor rate − cohort rate

A +54 point gap means visible winners overstate success by that much. Copying their playbook without the failure data is copying a filtered slice of history.

A simple application: portfolio stories

Investors study companies that made it. Product teams read reviews from users still active enough to post. Leaders benchmark against peers still in the league table. In each case, failures left the dataset before the analysis began.

Portfolio stories: who left the dataset

Raise dropout. Visible success rate rises because failures already exited.

59% among survivors — true cohort ~45%

Cohort flow

Success rate (%)

True: 45% · Survivors only: 59%

Survivor rate

59%

True rate

45%

Dropout

40%

Optimize (move here)

  • Report started vs remained on every success metric
  • Pair reviews with churn/exit data

Hold (do not over-react)

  • Benchmarking from league tables of survivors only

Escalate if

  • Exit reasons cluster in one product area

Ask how many started, how many remain, and what happened to the rest before strategy from success stories alone.

Ask for the full cohort: how many started, how many remain, and what happened to the rest. Pair visible success rates with exit data, churn, or closed-book records before you set strategy from survivor stories alone.