When Dr. Jonah Tebaa evaluates whether an AI initiative will succeed, he pays almost no attention to the launch. In his experience, the day an initiative is approved — the deck, the vendor, the ambition — tells you very little about whether it will survive. What he watches instead is the first ninety days.
It is a deceptively simple lens, and it reflects a conviction that runs through his work: the model was never the hard part. The hard part is whether an organization builds the conditions for the technology to compound. That is where most AI initiatives quietly fail — not at launch, but in the weeks afterward, when the novelty fades and no one is watching closely.
Why ninety days is the deciding window
Dr. Tebaa argues that ninety days is the moment of truth because it is long enough for novelty to wear off and short enough that the original sponsors are still paying attention. Inside that window, an initiative either becomes part of how people actually work or becomes something they politely work around. Once that pattern sets, he notes, it is very hard to reverse. The initiative becomes either load-bearing or decorative.
The five questions he asks
Rather than reviewing technology, Dr. Tebaa tests for operational gravity. He looks for clear, named answers to five questions:
- Who owns the daily output? A specific person who notices when quality drops on the day it drops — not at the quarterly review.
- Is there a failure protocol? A decision, made in advance, about what happens the first time the system is confidently wrong.
- Are real users touching it weekly? In their actual work and deadlines, not in a hand-held pilot engineered to succeed.
- Is the input being refreshed? Because the data, context, and instructions every AI system depends on go stale on a schedule if no one owns them.
- Has anyone said no to scope? The initiatives that ship, he observes, got narrow on purpose.
When most of these have owners with names attached, Dr. Tebaa is confident the initiative will ship. When they are vague, he believes no roadmap can save it.
The first of those questions is becoming a legal requirement, not only good practice. For high-risk systems in Europe, Article 26 of the EU AI Act states that "Deployers shall assign human oversight to natural persons who have the necessary competence, training and authority, as well as the necessary support." A named owner with the authority to act is the same thing Dr. Tebaa looks for on day forty-five.
What good looks like at ninety days
Notably, Dr. Tebaa is not looking for scale early. He looks for evidence that the initiative has begun to pull real work toward it — one workflow that genuinely depends on it, one team that would complain if it vanished, one signal someone checks without being reminded. The initiatives that eventually scale, he points out, almost always look modest at ninety days, because they did one thing completely rather than ten things partially.
Discipline over spectacle
What makes the framework rare, in his telling, is not its complexity but its unglamorous nature. Clear ownership, honest failure handling, real adoption, fresh inputs, and the discipline to narrow scope — all of it has to happen after the exciting part is over. This operational layer, turning an AI decision into a durable capability, is the work Dr. Tebaa cares about most, and the work his team delivers for organizations across the region. His advice to leaders is direct: do not wait ninety days to apply the test. Ask the five questions before launch. If they do not yet have owners with names, what you have is not an initiative — it is an intention.
