Real options thinking, applied to AI bets

When a board funds an AI initiative to completion up front, it pays the full cost regardless of what it learns along the way. If the bet fails, and RAND puts the AI project failure rate above 80%, the entire investment is lost. That is the opposite of how you should invest under deep uncertainty.

Real options theory says that under uncertainty, the right to wait and learn has value. You pay a little to keep the option open, then commit fully only when the evidence justifies it. AI, where value is genuinely unknown until tested, is the textbook case for this.

A pretotype is the cheapest possible option on an AI bet. For a small, fixed cost you run a behavioural test that tells you whether to invest further. If the signal is strong, you exercise the option and fund the build. If it is weak, you let the option expire and keep your capital.

This reframes the spend on a pretotype. It is not overhead before the real work. It is the purchase of decision rights. You are paying to make the next, much larger decision under far less uncertainty.

Across a portfolio, the implication is powerful. Instead of a handful of large, all-in AI bets, you hold many cheap options and exercise only the few that prove out:

Managing AI as a portfolio of real options lets a board move fast and stay disciplined at once. It funds learning cheaply, concentrates capital on proven bets, and treats a stop decision as a feature, not a failure.

That is the financial logic underneath Exponentially’s Rapid Experimentation Operating Model: test broadly, invest selectively, and keep every next dollar conditional on evidence.