Every Founder Thinks They're the Exception

Entrepreneurs are highly motivated and dedicated people, but it is very interesting to look at what goes on inside their minds during the whole process of idea to market. When a person first comes across an idea they are highly motivated and certain of the success of their product, and what they miss out on is real-time research, surveys and idea validation.

The problem is that unlike professionals, a founder has just found their third, fourth or fifth love in an idea that arrived while walking down the street. Ask them a hard question “do you believe this idea will work?” and they tend to fall back on heuristic thinking and quietly answer a much simpler one instead: do I like the idea? And this isn't just a guess. Studies show that we humans tend not to use our algorithmic thinking unless we are specifically prompted to do the math. Econs, the perfectly rational creatures who live in economics textbooks, would work differently. We are not Econs.

I am not saying the professionals are always right either. Take an actuary, someone whose entire job is putting a number on uncertainty. If someone had predicted the success of an idea you just came across and the uncertain world decided to put forward a major event such as an earthquake or a flood, the scenario changes. That is why you see even the biggest stock market professionals using ifs and buts when predicting the price of a stock, or the success of the President of the United States. The idea of predicting the future is always controversial, because at some point even the best professional will be biased, and their intuition will play a part in a decision that is bound to be wrong.

The evidence for this is older than most people assume. In 1954, the University of Minnesota psychologist Paul Meehl published Clinical versus Statistical Prediction: A Theoretical Analysis and a Review of the Evidence, a slim review of about twenty studies that pitted trained clinicians against simple statistical formulas. The clinicians were asked to predict things like which patients would respond to treatment, which parolees would reoffend, which students would succeed. The formulas were given the same information, often less of it, and combined it by arithmetic. In nearly every comparison the formula matched or beat the expert.

Part of the reason is that human judges are noisy. Show the same expert the same case twice and you often get two different answers. Daniel Kahneman, in Thinking, Fast and Slow, cites experienced radiologists who, asked to classify chest X-rays as normal or abnormal, contradicted their own earlier verdicts on twenty percent of the images. The same instability turns up in auditors, pathologists and managers. A formula has no such problem: identical inputs produce identical outputs every time. This is also why a statistical model built to imitate a particular expert tends to outperform that expert. The model captures his rules and discards his fluctuations.

So an entrepreneur like me is bound to overestimate the chances of my own product. If a trained radiologist can disagree with himself about an X-ray, what hope does a founder have of judging the one idea he is in love with? Roughly half of all new businesses in the US are gone within five years, and among venture-backed startups the odds are considerably worse. Those are the base rates every founder has heard and every founder assumes they are an exception to.

Here is the part I keep coming back to. The fix is not to feel less certain. Certainty is most of the fuel. Nobody quits a stable job and spends two years building something on the strength of a balanced assessment of the evidence. The conviction is doing necessary work, and stripping it out would leave you with a well-calibrated person who never starts anything.

The fix is to stop letting the conviction answer the question. Kahneman's own solution, when he redesigned officer selection for the Israeli army, was not to find better judges. It was to break the judgement into fixed pieces, score each one before moving to the next, and add them up at the end. The interviewers complained they were being turned into robots. Accuracy went up anyway. He let them record a gut impression at the end, and that turned out to be useful too, as one more input, never as the method.

That is the version a founder can actually use. Write down, before you start, what would have to be true for this to work. Go and check those things with people who have no reason to be kind to you. Decide in advance what result would make you walk away, and then actually walk away when you see it.

So I am left with something awkward instead of a solution. The certainty that makes a founder start is the same certainty that makes him a bad judge of whether he should have. You don't get to keep one and delete the other. You just get to decide how much evidence you'll require before the feeling gets to win.