I keep seeing the same AI meeting. Twelve people are in the room. Everyone agrees the technology is important. Three vendors have shown impressive demos. There is a pilot somewhere, but nobody can say what happens if it succeeds.
That is not a technology problem. It is a decision problem wearing an AI badge.
A pilot needs an owner, not an audience
When an AI initiative belongs to “the organisation,” it belongs to nobody. The technology team can make it work. The innovation team can make it visible. A consultant can make it sound strategic. But only a business owner can decide whether the result matters enough to change a workflow.
Start with the person who owns the number. If the problem is slow sales response, that might be the commercial director. If it is invoice processing, it may be finance. If it is customer-service backlog, it is operations. That person must define success before anyone selects a model or platform.
Do not ask, “Where can we use AI?” Ask, “Which expensive delay are we no longer willing to accept?”
Give the experiment a business boundary
“Improve customer experience with AI” is too soft to build and impossible to judge. “Reduce the time from a qualified website enquiry to a useful human response from four hours to ten minutes” has edges. It tells us where the workflow starts, where it ends and what value looks like.
A useful first AI initiative has four boundaries:
- One workflow: not an entire department.
- One accountable owner: not a steering committee.
- One measurable result: time, cost, conversion, accuracy or risk.
- One decision date: scale it, change it or stop it.
The 90-day AI decision
I prefer ninety days because it is long enough to expose real operational friction and short enough to prevent a pilot becoming permanent theatre.
In the first two weeks, map the workflow as it actually happens—not as the SOP says it happens. Find the exceptions, handoffs and missing data. During the next month, build the narrowest useful version and put it in front of the people doing the job. Use the remaining time to measure, correct and document what changes when the system meets reality.
At day ninety, leadership should face a decision, not receive another presentation:
- Did the result move?
- What did it cost to move it?
- What operational change is required to scale it?
- What new risk did we introduce?
AI strategy is mostly subtraction
Good AI consulting in the UAE and Saudi Arabia is not about adding every possible capability. Regional businesses already carry complicated stacks, legacy systems, WhatsApp-led customer journeys and approval processes shaped around how work really moves. The job is to remove uncertainty and connect the smallest number of systems required to create value.
Sometimes the right answer is an AI agent. Sometimes it is a deterministic automation with one AI step. Sometimes the business needs a cleaner process before it needs any model at all. I am comfortable with all three answers.
The goal is not to prove that AI works. We already know it works. The goal is to decide where it is worth changing the business.