Before approving an AI pilot, leadership teams should be able to answer five questions clearly. If they cannot, the pilot is likely to become another proof of concept that never scales.
1. What decision or task will change?
Name the specific process, the people involved and what they will do differently. "Use AI in customer service" is not specific enough; "draft first responses to the 40% of enquiries about order status" is.
2. How will we measure success?
Agree the baseline and the target before you build: time saved, error rates, revenue, customer satisfaction.
3. Is the data available, and are we allowed to use it?
Check data quality, location and consent. Regulations such as Nigeria's NDPA, Saudi Arabia's PDPL and South Africa's POPIA all have implications for how AI may use personal data.
4. Who owns it after the pilot?
Every successful AI capability has a business owner, not just a technical one.
5. What would it take to scale?
Think early about integration, security, cost at volume and training. A pilot designed with scale in mind is far more likely to get there.