There is a point in almost every AI conversation when someone asks the wrong question. Which model should we use? Which vendor is ahead? How quickly can we get a pilot live?
Having worked with technology businesses and watched organisations assess what to buy, build and adopt, I have become increasingly convinced that the difficult part is rarely choosing the technology. It is deciding whether the organisation is ready to make the investment in the first place. Before approving an AI budget, these are five questions I would want answered.
1. What is the actual business problem we need AI to solve? Not “we need to be more innovative” or “the board wants an AI strategy”. A specific problem, with a measurable cost if it remains unsolved. RAND's research into AI project failures, based on interviews with 65 experienced data scientists and engineers, identified misunderstanding or miscommunicating the problem as the most common cause of failure. If the problem cannot be written clearly in one sentence, I would not start with a vendor shortlist.
2. Who owns the outcome by name? Not the project. The outcome. AI deployment is rarely the finish line: models change, processes change, people have to adopt new ways of working, and pilot economics can look very different at scale. IBM's 2025 CEO Study found that only 25% of AI initiatives had delivered expected ROI and only 16% had scaled across the enterprise. Someone needs to remain accountable for the business result, with the authority and budget to act.
3. What does success look like in numbers? “Improved efficiency” and “better customer experience” are not business outcomes. What changes, by how much and by when? S&P Global Market Intelligence's 2025 research found that 42% of companies had abandoned most of their AI initiatives, compared with 17% the previous year. A defined target gives an organisation not only a way to measure success, but a basis for deciding whether to continue.
4. What happens if the investment delivers nothing? What if the pilot underperforms, adoption remains low or the economics do not work at scale? A serious AI investment should have an exit strategy alongside its implementation plan: clear milestones, evidence required to proceed and an agreed point at which the investment can be reconsidered. Otherwise, a pilot can become a programme that is increasingly difficult to stop.
5. Has anyone independently challenged the decision? Vendors have products to sell and implementation partners have services to sell. That is how the market works. But the buyer still needs someone asking whether the problem genuinely requires AI, whether the proposed solution is proportionate and whether the economics are realistic. The expensive mistake is rarely choosing Vendor A instead of Vendor B. It is approving the wrong problem and discovering it after the money has been spent.
These questions will not guarantee a successful AI investment. They should make the decision itself considerably better. At ALTA, this is where we believe the conversation should start before the vendor shortlist, before the pilot and before the budget is committed.