How to pick your first AI project: a five-question filter

AI use is widespread, but results are not. McKinsey's 2025 global survey found that 88% of respondents' organizations use AI in at least one business function, yet only 7% say it is fully scaled and just 39% report any enterprise-level EBIT impact. The gap is rarely the technology. It is usually the choice of first project.

The five-question filter

• Does the work recur at least weekly? Rare tasks do not pay back the setup effort.
• Can you state the outcome as a number? For example, hours per week or days to first response.
• Can the system reach the inputs? If the data is trapped in someone's head or a paper file, fix that first.
• Is there a named person who owns the result and reviews the output?
• What is the worst case if it is wrong, and can you recover from it? Start where the answer is "a person catches it before it matters".

Examples that usually pass the filter

• Routing and first response for inbound leads and enquiries.
• Drafting quotes or proposals for a person to approve.
• Entering data from invoices, purchase orders or forms.
• Sorting and drafting replies for a shared customer-service inbox.
• Searching your own procedures and documents.

How to run it

• Weeks 1 and 2: measure the current process. This is your baseline.
• Weeks 3 to 6: build the narrowest version that does one job, with a person approving the output.
• Weeks 7 to 10: run it on real work and measure the same numbers.
• Decide: scale it, change it or stop it. Stopping a project that did not move the number is a good outcome.

A note on forecasts

Gartner has forecast that over 40% of agentic AI projects will be canceled by the end of 2027 because of escalating costs, unclear business value or inadequate risk controls. That is a forecast from an analyst firm, not a measurement, and it is worth reading as a prompt to define value and controls up front, which is what the filter above does.

Sources

• McKinsey, "The State of AI: Global Survey 2025"
• Gartner, "Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027" (June 25, 2025)