A lucky break leaves the method unproven
A successful result can mix good decisions with circumstances that will not recur. Research on streaks and scientific careers helps define what it can prove.

A project wins a major customer after an introduction at a conference. The founder remembers the conversation, the sales team remembers the proposal, and the public story becomes a lesson about the company's method. The introduction itself may have depended on someone changing tables at lunch.
Closing the deal still required work. The contingent introduction limits what the result can prove about the process that produced it. A successful outcome can contain both effective decisions and circumstances the team could not repeat on demand.
The problem is deciding which parts deserve to become policy after the celebration.
A winning result leaves several explanations open
Suppose a team changes its outreach message just before a prospect receives a new budget. The improved message may have helped. The timing may also have mattered. One completed sale cannot isolate their contributions.
Keep the claim close to the observation. The revised message preceded a sale. Showing that the revision improves the chance of a sale requires a comparison that takes other relevant differences into account. Even a careful comparison will carry uncertainty when the number of opportunities is small.
This affects how a company evaluates someone on the strength of one famous success. Ask what they chose before the outcome was known. Then ask how the same approach performed across their other attempts. A detailed explanation of the winning case can leave the unsuccessful cases completely out of view.
A decision record made at the time helps. It can show which assumptions were available to the team, the alternatives it considered, and what result would have counted as failure. Later storytelling is easier to judge against that record.
Streaks need an appropriate comparison
The usual warning about randomness can itself become too confident. A sequence of successful shots does not automatically prove a hot hand. Declaring every streak meaningless goes further than the evidence permits.
In their 2018 paper, Joshua Miller and Adam Sanjurjo showed a selection bias in a commonly used statistic for outcomes after streaks. Correcting it reversed the longstanding conclusion in the canonical study they reanalysed. This is a finding about a particular method and evidence, with consequences for how streaks are tested.
It provides no general rule that a successful founder is more likely to succeed at the next venture. A business history needs its own comparison. Market conditions can change between attempts, while the person may also gain experience or access to resources.
The question to ask is what performance would look like under the proposed explanation. If pure chance is the comparison, define that chance process. If experience is the explanation, look for evidence that distinguishes experience from the easier circumstances success may have created.
A model's stable parameter has a limited meaning
Roberta Sinatra and colleagues' 2016 study examined scientific careers and the timing of high-impact work. Their model separated productivity from a stable individual parameter, Q, and a random component affecting a paper's impact.
The stable parameter belongs to that model of scientific impact. Turning it into a claim that every person's ability stays fixed throughout life changes both the population and the meaning of the measurement. Citation impact also differs from the value of someone's work in settings where citations are irrelevant.
The study offers a useful reason to be cautious about judging a career from the timing of one celebrated result. Applying that caution to a startup is an analogy. It should be presented as an interpretation, with room for differences between publishing research and building a business.
Costs change the result being explained
A lucky success can be expensive to pursue. If a team runs many campaigns, the revenue attributed to the best one needs to be read alongside the spending on the others. Selecting only the winner changes the question from whether the overall method worked to whether it produced at least one attractive example.
The distinction has a well-known research counterpart. Eugene Fama and Kenneth French's 2010 mutual-fund study found that few funds in their analysis generated benchmark-adjusted expected returns sufficient to cover their costs. Their results also distinguished performance before costs from performance after them.
That is a dated study of a defined investment setting. It does not settle the explanation for every later winning fund. For an operating business, the transferable accounting question is narrower: which costs disappeared when someone described the success?
Include the abandoned attempts when reviewing the method. A profitable customer acquired after months of unsuccessful experiments can still be worthwhile. The experiments belong in the assessment of what it took to find that customer.
Keep the next experiment small enough to interpret
After an unexpected win, choose a specific part of the method to test again. Reusing the revised outreach message with a comparable audience can produce evidence about the message. Rebuilding the entire conference strategy around one chance introduction makes the explanation harder to test.
Set a review point before the new results arrive. Record what would justify continuing the experiment and what would lead the team to change it. Keep the cost within a limit the business can absorb if the original explanation proves wrong.
When the next outcome appears, open the earlier decision record before writing the case study. Compare what the team expected with what actually happened. The difference is where a useful lesson may still be hiding.