Every enterprise we meet has already seen an impressive demonstration. Fewer have a workflow that is quietly better than it was a year ago. The difference is rarely the model. It is the choice of first project.
A good first project has four properties. It is repetitive, so the gain compounds. It is measurable, so nobody has to argue about whether it worked. It has a clear owner, who decides what “good” means and signs off on changes. And it touches information the organisation is allowed to use, so the pilot does not stall in a permissions review.
Boring projects pass all four tests. Claims triage. Supplier document checks. Matter intake. Tier-one support cases. They are unglamorous and they are where the time goes. An agent that assembles a complete, sourced summary for a human decision-maker saves hours a day without changing who is responsible for the outcome.
What a showcase costs
The opposite choice, a visible, cross-functional, end-to-end transformation, usually fails for reasons that have nothing to do with AI. It needs data from five systems with five owners. Its success criteria are negotiated after launch. And when something goes wrong, nobody can say which part was at fault.
We see the same pattern across Taiwan, Malaysia, Australia and New Zealand: the organisations that moved fastest in year two are the ones that shipped something small and real in year one, learned how to evaluate it, and then extended.
How we choose
During the Diagnose stage of an AKE deployment we rank candidate workflows on value, feasibility, operational risk and readiness. The recommended pilot is almost never the one with the biggest headline number. It is the one the team can own, measure and extend. That is the project that earns the right to do the ambitious one next.