Changing the Ending

Knowing how the movie ends for the CINO is an advantage for the CAIO, or for anyone leading in both areas, like many of you. In Part 1 we named the trap: the AI mandate, similar to the innovation mandate, is given without the authority to make the hard calls, and the decisions that have the greatest impact can be made by default, at the speed of AI.

The good news is that the trap is structural, and structures can be rebuilt. And the rebuild is simpler to start with than it looks. Out of everything AI touches in your company, a handful of decisions determine whether the work pays off. Get those few back into the hands of the people accountable for the outcome, with the authority and the structure to make them stick, and you are most of the way out.

So start by giving up on owning all of it. You cannot personally maintain accountability across everything AI touches, and you do not need to. Accountability drains from a decision in two ways: slowly, when a funding decision runs on for quarters because no one owns the call to stop it, and quickly, when an AI takes a consequential action before anyone can weigh in. The work is to maintain it at the few decisions that matter, at both speeds.

Here is how, in four moves.

Name the decision, and give it one owner. Take the handful of decisions that determine whether the work pays off, whether to move a pilot into production, whether to scale it, whether to stop it, and put a single named person on each who accepts that the call is theirs.

The reason this matters is that today most of these belong to “the process” or “the team,” so they get made by drift rather than by a person. An owner’s real job is stewardship: watching the decision as conditions change, making it at the moment it should be made rather than when it is finally forced, and leaving a record of who decided and on what basis, so the organization can learn from the call and answer for it later.

It is also the move the other three depend on, because criteria need someone to apply them, a stop needs someone to accept it, and a scoreboard needs someone accountable for the number.

Set the criteria before the decision. Agree in advance what a pilot has to show to earn its next round, and what would make you stop it.

This is where two ideas from Dan Toma, Cost of Failure and the gate, come together: the gate is the moment a project continues or stops, and the criteria are what make that moment a real decision.

Setting them early does two things. It keeps the call from becoming a personal verdict on whoever sponsored the work, because the test was agreed before anyone was attached to the answer. And it holds the bar still: without a line set in advance, every review turns into a negotiation, and the reasons to continue accumulate—”we only need one more quarter,” “this one is strategic,” “the sponsor has too much riding on it.”

That is how work that should have stopped keeps earning another round, and how the Cost of Failure climbs. Pre-set criteria turn the decision from a debate you can lose to your own best arguments into a test the work either passes or fails. (More on the fear behind those exceptions in the next piece.)

Build the stop into the structure itself. For the actions an AI takes on its own, the structure should stop the consequential action from executing until a named person has accepted it.

The reason goodwill is not enough is speed: at the pace AI runs, a rule that lives in someone’s head or in an agent’s instructions will not hold at the moment it counts, and by the time a person notices a bad action it has already executed, often with no way to take it back. Stopping execution at the decision point is the only place a human can still change the outcome, because everything downstream is cleanup.

It also closes the “the AI did it on its own” gap: the action cannot happen unless a named person accepted it, which is exactly what regulators and courts have started asking organizations to show.

Put the scoreboard on the decisions. Track what the decisions produce—the capital you redeploy by stopping work that is not earning it and the time it takes to move from a pilot to a real call—rather than counting pilots launched, tools adopted, and people trained.

The reason this matters is that teams optimize for whatever you measure, so counting pilots launched buys you more pilots, which is how Innovation Theater gets funded. Measuring the decisions changes what the organization gets good at: stopping well and deciding quickly, instead of simply starting things.

And measured over time, as ownership gets attached, the Cost of Failure stops reading as money already spent and starts reading as capital you can win back—the number that earns your function its budget.

Our portfolio management and Innovation Accounting platform, SATORI, gives companies the flexibility to define how frequently they review each initiative, map their governance processes, and equip decision-makers to make one of three decisions at every review meeting: progress, persevere, or stop.

None of this is easy, and the politics do not disappear. Naming an owner means someone has to accept the call. Setting criteria in advance means giving up some room to maneuver later. But it is a smaller and more winnable fight than owning the entire AI agenda, and it is the one that actually changes the outcome.

Underneath all four moves is a single idea: get human judgment back to the moment the decision is made, and give it the authority and the structure to hold. That is what went missing in the innovation wave, and it is what is going missing again, faster, in AI.

Next we will take on the hardest of these decisions, the stop, and the fear that keeps it from happening: who gets blamed when you kill the pilot.

For now, pick the single most consequential AI decision your team is responsible for, and ask who actually owns it, and against what criteria. If the answer is “the process,” you have found where to start.