You make a decision you probably shouldn't have made. Maybe you rushed it, ignored something important or took a risk you couldn't really justify.
And then it works.
Nothing immediately tells you that the reasoning was poor. Quite the opposite. The result gives you a reason to use the same reasoning again. Whatever doubts existed before the outcome now have to compete with a fairly persuasive piece of evidence: last time I did this, it worked.
Usually, that's a sensible way to learn. Successful outcomes often contain useful information about the actions that produced them. The complication begins when the result was good but your decision had less to do with producing it than you think.
A bad decision that fails comes with an obvious problem. Something went wrong. A bad decision that succeeds is harder to spot because the result gives you no immediate reason to investigate the process behind it.
So perhaps the hidden danger of a bad decision isn't always that it fails. Sometimes it might be that it works.
When Success Deserves the Credit
Researchers studying goal-directed behaviour use the idea of instrumental contingency to describe how much performing an action actually changes the probability of receiving a particular outcome [1]. Behind the technical language is a surprisingly simple question: would this have happened anyway?
Imagine two people recommend a supplier and both projects succeed. In one case, choosing that supplier substantially increased the project's chances of success. In the other, almost every available supplier would probably have produced the same result. Both people get to observe the same sequence: I made a decision and the project succeeded. But those outcomes contain very different amounts of information about the decisions that preceded them.
That's the first problem with being right. A successful outcome tells you what happened. It doesn't automatically tell you how much credit your decision deserves for making it happen.
The distinction is easy to describe after the fact. It becomes much harder when you're the person who just got the result you wanted.
What If the Reward Had Nothing to Do With What You Did?
There is a strange way to make this problem visible: remove the connection between behaviour and reward altogether.
In an experiment by Koichi Ono, university students sat in front of a counter and several levers. Points appeared according to schedules controlled by the experiment, regardless of what participants did. There was no behaviour they needed to discover that would reliably produce the reward [2].
Most participants did not develop a persistent pattern. But three of the 20 students did. They developed distinctive behaviours that continued even though those behaviours had no actual control over when points appeared. One participant, for example, developed a sequence involving a lever and movements around the room.
That limitation matters. The experiment does not show that humans inevitably invent strategies whenever random rewards appear. Seventeen participants did not develop the same kind of persistent behaviour. What it does show is more modest and more interesting: under some conditions, behaviour can become organised around rewards even when the behaviour did not cause them.
From the participant's perspective, the sequence still contains everything needed to look persuasive. They did something, something good happened, and eventually a relationship between the two could begin to emerge even though the experiment itself contained none.
The missing information is the counterfactual: would the reward have appeared if they'd done something completely different?
When Being Right Starts to Feel Informative
The problem becomes more interesting when people aren't simply receiving rewards, but actively making predictions.
In an experiment by Peter Ayton and Ilan Fischer, 32 participants repeatedly predicted one of two outcomes on a computerised roulette wheel. The outcomes were genuinely random: each colour had an equal 50% chance on every trial, and participants received feedback after each prediction [3].
The researchers found something curious. Runs of the same roulette outcome produced the familiar gambler's fallacy: after seeing the same colour repeatedly, participants increasingly expected the other colour. But the history of their own predictions produced a different pattern. Runs of correct predictions produced positive recency, consistent with participants treating their recent success as informative about their chances of being right again [3].
Nothing about a run of correct guesses had made the next guess more likely to succeed. What had changed was the participant's own history. They had been right.
That distinction matters. Accidental success doesn't necessarily need to produce an explicit belief that someone has discovered a reliable strategy before it can influence expectations. A sequence of correct decisions can begin to acquire credibility simply because it exists.
One success is an outcome. Repeated success starts to become a track record.
The reasoning hasn't necessarily improved. Its history has become more persuasive.
What Happens After a Bad Bet Wins?
Gambling makes this unusually easy to examine because the distance between a decision, an outcome and the next decision can be very short.
Someone can make a bet for a reason that has little or no predictive value, perhaps following an unsupported system or simply feeling unusually confident about an outcome, and still win. The reasoning may have contributed nothing to the result, but the reward is completely real.
Evidence from actual gambling behaviour suggests that winning histories can be associated with what players do next. Abe and colleagues analysed 7,935,566 baccarat games played by 3,986 casino customers and examined how betting changed during sequences of wins and losses [4]. As streaks became longer, players increased the amounts they wagered, with the increase being larger during winning streaks. Certain multiple bets, including longshot bets, also became more common following sequences of wins and less common during losing streaks.
Because this was observational casino data, it cannot tell us exactly what each player believed or establish that winning caused a particular belief about strategy. But it does show that histories of successful outcomes were associated with changes in subsequent betting behaviour.
That's enough to make gambling useful here. A win doesn't have to prove that the original reasoning was sound before it becomes part of the feedback carried into the next decision.
And if another win follows, the player no longer has just an outcome. They have a history.
When Success Lets the Process Survive
Repetition is only part of the problem. There is another possibility: a good result can leave less reason to examine the process that produced it.
Research on learning from experience draws an important distinction between simply receiving outcome feedback and systematic reflection. Shmuel Ellis and colleagues describe systematic reflection as examining the actions and decisions behind an outcome in order to understand what actually produced it [5]. Their review argues that this kind of reflection can help people learn from successes as well as failures.
That matters because success does not prevent scrutiny. You can make money and still ask whether your reasoning deserved the result. A team can deliver a successful project and still examine which decisions actually contributed to it.
The interesting possibility is what happens when nobody feels much need to ask.
When something fails, the process is likely to attract attention because there is a problem to explain. Success contains no equivalent demand. If the outcome already looks like confirmation, a weak process can survive without anyone establishing why it worked.
Repeated success can make that harder to notice. Each favourable outcome adds another entry to the apparent track record. Eventually, the process is no longer supported by one lucky result but by a history of things apparently going right.
The important word there is apparently. A history of success and evidence for a good process can overlap. They are not necessarily the same thing.
Failure Isn't Automatically the Better Teacher
It would be tempting to reverse the argument and conclude that failure must therefore be the more useful teacher.
The evidence doesn't support something that simple.
Experiments by Keith and colleagues found that, under some conditions, people actually learned less after failure feedback than after success feedback [6]. Failure can produce what researchers have described as a tune-out effect, where people disengage from negative feedback rather than examining what it can teach them.
The experiments also showed why the distinction matters. The learning disadvantage following failure disappeared under different incentive framing and when participants received corrective information alongside the failure feedback [6]. Knowing that something went wrong was not always enough. Learning improved when the feedback also helped people understand what needed to change.
So failure contains no automatic wisdom. It creates a problem that may need explaining, but that doesn't guarantee the right explanation will follow. Success isn't automatically deceptive either. If the relationship between action and outcome is strong and the process is examined carefully, a good result can be exactly the feedback it appears to be.
The harder question applies to both outcomes: what did this result actually teach me about the decision that produced it?
The Problem With Being Right
Being right once can mean many things.
Perhaps the reasoning was excellent. Perhaps the decision genuinely improved the odds. Perhaps the outcome would have happened anyway. Or perhaps chance temporarily made a weak process look much better than it was.
The evidence does not establish that every accidental success becomes a habit, or that a lucky win inevitably grows into a long-term strategy. What it does show is that rewards can sometimes shape behaviour without a genuine action-reward relationship, that runs of successful predictions can influence expectations even when the outcomes themselves are random, and that winning histories in gambling can be associated with changes in what players do next.
One lucky result may simply be luck. But if the same behaviour succeeds again, something changes. The process gains a history.
And histories are persuasive. They give us examples to remember, previous results to point to and reasons to do the same thing again. Eventually, what began as it worked can start to feel more like this works.
Sometimes that conclusion will be correct. The difficulty is that success alone cannot tell you which kind of history you're building.
That's what makes a bad decision succeeding different from a bad decision failing. Failure puts the process under suspicion. Success can give it another chance.
The obvious problem with a bad decision is that it might fail. The less obvious one is that it might work.
References
- Liljeholm, M. (2021). Agency and goal-directed choice. Current Opinion in Behavioral Sciences, 41, 78–84. Source
- Ono, K. (1987). Superstitious behavior in humans. Journal of the Experimental Analysis of Behavior, 47(3), 261–271. Source
- Ayton, P., & Fischer, I. (2004). The hot hand fallacy and the gambler's fallacy: Two faces of subjective randomness? Memory & Cognition, 32(8), 1369–1378. Source
- Abe, N., Nakai, R., Yanagisawa, K., Murai, T., & Yoshikawa, S. (2021). Effects of sequential winning vs. losing on subsequent gambling behavior: Analysis of empirical data from casino baccarat players. International Gambling Studies, 21(1), 103–118. Source
- Ellis, S., Carette, B., Anseel, F., & Lievens, F. (2014). Systematic reflection: Implications for learning from failures and successes. Current Directions in Psychological Science, 23(1), 67–72. Source
- Keith, N., Horvath, D., Klamar, A., & Frese, M. (2022). Failure to learn from failure is mitigated by loss-framing and corrective feedback: A replication and test of the boundary conditions of the tune-out effect. Journal of Experimental Psychology: General, 151(8), e19–e25. Source