Why Your AI Vendor Can't Answer the Seven Questions That Matter

A vendor walks into a conference room with a polished deck and a compelling demo. Twenty minutes later, someone in the room says "this looks promising." Forty minutes later, there's a verbal agreement to run a pilot. Nobody asked the seven questions that would have told them whether the pilot was a good idea.

This happens more often than anyone admits. Not because the people in the room are careless, but because they have no structured framework for evaluating AI proposals. They know how to evaluate budgets, timelines, and vendor credentials. They do not know how to evaluate whether a process is suitable for AI in the first place.

The vendor won't help. Vendors sell solutions. They are not in the business of helping you decide whether you need one. Their deck will show you what the tool does. It will not help you determine whether your process, your data, your regulatory environment, and your team's readiness make this a sensible decision.

That evaluation is your job. And most managers don't have a methodology for doing it.

The questions nobody asks

When I work with managers across enterprise organisations, I find the same pattern. The conversation about AI jumps from "this is interesting" to "let's try it" without passing through the stage that matters: structured evaluation. The questions that get skipped are not technical. They are governance questions. And they fall into seven categories.

Volume. How much of this work actually exists? AI is expensive to implement and maintain. If the process handles twelve cases a month, the economics rarely justify automation. If it handles twelve thousand, the conversation changes. But volume alone doesn't decide anything. It's the first filter, not the verdict.

Error tolerance. What happens when the AI gets it wrong? In some processes, errors are caught downstream and corrected with minimal cost. In others, a single mistake triggers regulatory exposure, customer harm, or reputational damage. The vendor will tell you the accuracy rate. They will not tell you what happens when the remaining percentage goes wrong in your specific context.

Regulatory and compliance rules. Is this process governed by rules that require human judgment, audit trails, or explainability? Certain industries and functions have constraints that make full automation either illegal or professionally irresponsible. The vendor's compliance section will describe their platform's certifications. It will not map those certifications against your organisation's actual obligations.

Data readiness. Is the data this process runs on clean, structured, accessible, and sufficient? Most AI failures are data failures. The process might be perfect for automation, but if the underlying data lives in thirty spreadsheets across four departments with no consistent formatting, you have a data project before you have an AI project. Vendors demo on clean data. Your data is not clean.

Integration complexity. How does this tool connect to your existing systems? A standalone AI tool that requires manual data export and import is not automation. It's a new step in the same manual process. The integration question determines whether the tool genuinely saves time or merely redistributes effort. Vendors describe integrations as "seamless." Ask your IT team whether they agree.

Change readiness. Will the people who currently do this work accept the change? Will their managers support it? AI implementation is an organisational change project, not a technology deployment. If the team sees the tool as a threat to their roles, adoption will fail regardless of how well the technology works. The vendor's implementation timeline does not include the six months of change management that most organisations need.

Time to value. How long before this investment produces measurable results? Not the vendor's projected ROI. Your actual timeline, given your data readiness, your integration complexity, your change management needs, and your IT team's capacity. Most organisations underestimate this by a factor of two or three.

Why vendors can't answer these for you

This is not a criticism of AI vendors. They build products. Good ones build very good products. But their incentives and yours are structurally misaligned at the evaluation stage.

The vendor wants the pilot to happen. You need to know whether the pilot should happen. Those are different objectives, and they require different questions.

A vendor can tell you what their tool does. They cannot tell you whether your process is a good candidate for it. They can describe their accuracy benchmarks. They cannot tell you whether your error tolerance accommodates the gap. They can list their platform integrations. They cannot assess whether your systems are ready to receive them.

The evaluation is inherently internal. It requires knowledge of your processes, your data, your regulatory context, your team dynamics, and your organisational readiness. No external vendor has that knowledge. And no external vendor is incentivised to tell you that the answer might be "not yet."

What a structured evaluation looks like

The seven questions above are not a checklist to run through quickly. Each one opens a line of enquiry that requires investigation, discussion, and documentation. A structured evaluation of an AI proposal takes time. It should produce a documented verdict that can be shared, challenged, and defended.

The output is not a yes or a no. It is one of three outcomes: go, no-go, or pilot with conditions. Each outcome carries specifics. A "go" specifies what success looks like and how it will be measured. A "no-go" specifies what would need to change for the decision to be revisited. A "pilot with conditions" specifies the scope, the timeline, the success criteria, and the exit conditions if the pilot does not meet them.

This is what governance looks like in practice. Not a policy document that sits in a shared drive. A repeatable methodology that managers can apply every time an AI proposal lands on their desk.

The cost of not asking

Organisations that skip structured evaluation don't save time. They spend it later, in less productive ways. Failed pilots consume budget and erode internal confidence in AI. Poorly scoped implementations create new problems instead of solving existing ones. Vendor relationships sour when expectations weren't aligned from the start.

Worse, the precedent compounds. Once an organisation approves one AI initiative without structured evaluation, every subsequent proposal follows the same path. Enthusiasm replaces analysis. Speed replaces rigour. And when something goes wrong, nobody can point to the evaluation that should have caught it, because the evaluation never happened.

The seven questions are not difficult. They do not require technical expertise. They require a framework and the discipline to use it consistently. That is a capability most organisations need to build deliberately. It does not emerge on its own.