One of the most valuable things a business can do before starting an AI project is to evaluate it rigorously — before any money changes hands. This sounds obvious. In practice, it almost never happens, because the people proposing AI projects are usually enthusiastic about them, and enthusiasm is not a great filter.
Start with the problem, not the solution
The first question is not "what AI should we use?" It is "what is the specific outcome we are trying to achieve?"
"We want to use AI to improve customer service" is not a specific outcome. "We want to reduce average response time from 4 hours to under 30 minutes" is. One of these can be evaluated. The other cannot.
If you cannot state the desired outcome in measurable terms, you are not ready to evaluate an AI project. You are still in the problem definition phase.
The four questions that matter
1. Is there data to support it? AI learns from data. What data does this project require? Does it exist? Is it clean, consistent, and accessible? A project that requires data you do not have is a data infrastructure project first.
2. Is AI the right tool for this specific problem? Can you write down every rule needed to solve this problem? If yes, automation will be faster, cheaper, and more reliable. AI is the right choice when the task involves ambiguity, pattern recognition, or unstructured data at scale.
3. What does success look like in 90 days? Any AI project should produce a measurable signal within 90 days — not necessarily the final result, but evidence that the approach is working.
4. What happens if it does not work? What is the fallback? How much will you have spent? Understanding the downside before you start is not pessimism — it is due diligence.
The build vs buy decision
For most SMEs, the first instinct when considering AI is to buy an existing tool. This is often the right instinct — existing tools are faster to deploy and carry lower upfront costs.
But off-the-shelf AI tools are built for the general case, not your specific case. They process your data on someone else's infrastructure. The vendor controls the roadmap, the pricing, and the continuity.
Custom AI development costs more upfront but gives you a solution that fits your processes exactly, keeps your data under your control, and creates a competitive asset rather than a shared one.
The right answer depends on how differentiated the capability needs to be. If AI is supporting a generic process — buy. If AI is at the core of what makes your product or service different — build.
A final note on ROI
The return on investment of an AI project is almost always harder to measure than the proposal suggests and easier to achieve than the sceptics predict. The honest approach is to define your ROI metrics before you start, measure them seriously, and be willing to adjust course based on what you find.
The businesses that get the most from AI are not the ones that spend the most. They are the ones that ask the sharpest questions before they spend anything.