Ten Questions Before You Spend Anything on AI

61 percent of SMEs use at least one AI-enabled application. Only 21 percent report significant or transformational impact from it. Those two numbers describe the same group of businesses, which means the problem is not access to AI tools.

What the gap actually tells you

The SME adoption data shows that only 3.6 percent of firms have deployed AI across the enterprise. Japanese deployment data from the same research puts transformational impact at 2 percent for isolated tool use and 23 percent for enterprise-wide deployment. The difference between those two figures is not which tools a business chose. It is whether the business had the data infrastructure, governance, and staff capability to absorb AI beyond a single task.

Tool-first experimentation is a reasonable way to learn whether a specific tool fits a specific workflow. It is a poor way to build the conditions that produce 23 percent transformational impact rather than 2 percent. A founder who deploys a generative AI writing tool, saves time on marketing copy, and concludes the business is ready for AI has learned something real and something incomplete at the same time.

The critics of readiness diagnostics make a fair point: self-assessments carry systematic overconfidence, and scores reflect self-perception rather than actual capability. A founder who rates their data quality 4 out of 5 because their CRM is organized has answered the question as written, not as intended. That design risk is real. It does not disappear with a better questionnaire. What the research also shows, across the AIRI framework, the SAS AI Readiness Index, and a multidimensional SME readiness model validated in industry settings, is that even imperfect diagnostics surface blind spots that tool-first pilots never reach, specifically around data governance and skills gaps that founders do not know they have until a deployment fails.

The diagnostic

Score each question 1 to 5. One means not in place, five means fully operational.

Score 1-5 on each of the following:

• Your business data for a target AI use case sits in one accessible place, not split across disconnected tools with conflicting records.
• You know which data fields an AI system would need, and you have checked whether those fields are complete and consistent.
• Someone in your business owns the question of what AI is allowed to do with customer or employee data.
• You have a written position, even a one-paragraph one, on what AI decisions require human review before acting.
• At least one person on your team has used an AI tool to complete a real work task in the last 90 days, not a demo.
• Your team knows how to identify when an AI output is wrong, not just that it might be wrong.
• You have named one specific business problem you want AI to address, with a metric that would tell you whether it worked.
• Your AI use cases connect to a revenue, cost, or customer outcome you track today.
• You have reviewed what data you would share with an external AI vendor and whether your contracts permit it.
• You have set a budget ceiling for AI tools and a timeline for deciding whether to continue or stop.

What your score means

10 to 20: Your business is not ready to deploy AI beyond low-stakes, single-task tools. The priority is data consolidation and naming one use case with a measurable outcome. Deploying further before fixing those two things produces the 2 percent transformational impact figure, not the 23 percent one.

21 to 35: You have partial readiness. One or two dimensions are solid, others are not. The German manufacturing SME research on maturity models shows a consistent pattern: firms at this stage benefit most from fixing the weakest dimension first, not from advancing the strongest one further. Identify your lowest-scoring question and treat it as the constraint.

36 to 50: You have the structural conditions for broader deployment. The remaining work is connecting pilots to business metrics and building governance for the use cases you scale. The 82 percent of micro-firm non-adopters who cite "not applicable to business" as their reason for avoiding AI are not describing your situation. You have legitimate use cases. The question now is sequencing them against the metrics you already track.

A score in the top range does not guarantee good outcomes. The validation samples behind the frameworks this diagnostic draws on, including a 39-business sample for one peer-reviewed SME adoption model, are not large enough to make that promise. What the score gives you is a more structured basis for deciding where to spend time and money than running pilots without any baseline at all.