Treat AI adoption as a portfolio, not a project

Most organizations approach AI one project at a time. A pilot is approved, a team is assigned, and a year later there is a demo and a slide. Whether it paid for itself is a question nobody was made to answer.

I find it more useful to treat AI adoption the way an investor treats capital. Any single position can lose. The portfolio has to return. That changes what you ask of each initiative up front: what it costs, what it is expected to earn, how likely that is, when the return arrives, who owns the outcome, and what would make you stop.

The last two are the ones most often missing. An initiative without an owner produces activity. An initiative without a stop condition produces a permanent pilot. Neither is a bet. Both are a subscription to hope.

The discipline is not about caution. A portfolio is how you afford to take risk: several bets, sized, with a known exit, so that the failures are survivable and the winners get more capital. An organization that cannot say which of its AI efforts it would shut down has not decided anything yet.

The person who authorizes the spend still carries the consequence after the analysis is done. That is true of buildings and funds. It is true of models.