A successful founder tells you what worked. You hear a rule. Somewhere in the middle, a story became a causal claim.
“Charge more.” “Never offer a free tier.” “Launch every week.” Any of these might be useful in a particular business. A founder’s success does not tell us whether the business succeeded because of that choice, despite it, or for reasons the founder barely mentions.
The founder is describing one path through a market. We need to know what happened to comparable businesses that took the same path.
An outcome can also change how a decision is judged. Baron and Hershey demonstrated outcome bias in experiments where otherwise relevant decision information was held available to evaluators. That is evidence for separating the quality of a decision from knowledge of how it turned out, not a finding that every founder narrative is dishonest. Outcome Bias in Decision Evaluation
The missing denominator
Jerker Denrell’s 2003 paper examines how undersampling failed organizations can distort beliefs about management. Practices can appear beneficial among survivors even when they have no positive relationship with performance in the full population. A compelling pattern in successful companies can be an artefact of which companies remain available to study. Vicarious Learning, Undersampling of Failure, and the Myths of Management
Here is a hypothetical example. Twenty products use aggressive pricing and twenty use conservative pricing. Two of the aggressively priced products survive and become spectacular businesses. Ten of the conservatively priced products survive with modest returns.
Interview the two spectacular winners and you will hear an excellent story about pricing ambition. Count the original groups and aggressive pricing has a 10% survival rate against 50%. Neither view, by itself, establishes which strategy has the better expected return. We still need the size of the wins, the losses, the time involved and whether the groups were comparable in the first place.
Those numbers are invented. The sampling problem is real.
Failure stories need the same skepticism
The correction is not to replace founder success stories with founder failure stories and believe those instead.
A shutdown announcement might cite market conditions. The company may also have had distribution problems, expensive support or a product nobody needed badly enough. We should distinguish the event we can document from the explanation someone offers for it.
A discontinued product is also not automatically a failed business. It may have been profitable, acquired, folded into another product or retired because its owner found something better to do. An old plugin can keep doing a small job perfectly well. A closed listing tells you its status, not its entire history.
The useful record separates observation, attribution and inference. “The vendor announced an end date” is an observation. “The founder says maintenance costs became unsustainable” is an attributed explanation. “Platform dependence caused the shutdown” needs additional evidence.
Research on organizational failure also has context. Madsen and Desai find stronger learning from failures than successes in the orbital launch vehicle industry. That is an informative comparison, not a universal return estimate for collecting startup postmortems. Failing to Learn
Compare businesses before borrowing their rules
Advice about enterprise sales carries assumptions about contract value, procurement, support and customer expectations. Moving it into a cheap self-service plugin business without checking those assumptions is not learning from experience. It is copying the sentence and discarding the context.
Before treating advice as useful, I want to know the market, period, customer, price, distribution channel and definition of success. I also want counterexamples. Did the supposedly fatal practice appear in successful products? Did the recommended practice appear in failed ones?
Even then, comparison does not automatically establish causality. Better-funded companies may choose one strategy more often. Companies may change pricing because they are already struggling. Products launched recently have had less time to fail. Missing records may cluster around the smallest and least visible businesses.
These are reasons to narrow a claim, not reasons to stop investigating.
Automation can make collecting and organizing public evidence cheaper. It can also duplicate a founder’s story across ten syndicated articles and make one account look like ten independent confirmations. Source lineage matters. So does preserving disagreement instead of asking a model to turn every case into a confident lesson.
Turn advice into a test
The useful output is a conditional hypothesis: for this kind of customer and product, this choice might improve this outcome, through this mechanism.
Then decide what would change your mind. A pricing experiment might measure retained contribution margin after support and refunds, rather than celebrate a higher checkout price. A launch experiment should include maintenance and support work after the launch. A surviving business should remain in the comparison even when it makes the story less tidy.
Founder interviews are valuable. They tell us where to look, which tradeoffs someone noticed and which questions we might otherwise miss. They do not exempt us from doing the comparison.
I want startup advice that can survive contact with the companies that tried it and disappeared.