Blog · AI for small and midsize businesses
How to define a business problem before buying an AI tool
A one-page problem brief gives colleagues and suppliers the same question to answer. It explains who is affected, what evidence shows the bottleneck and what improvement would count, while leaving room to choose a simpler solution if AI is unnecessary.

In this article
“We need an AI assistant” describes a proposed solution. It does not tell a supplier which problem to solve or help a manager decide whether the investment worked. Before buying a tool, write a brief that another colleague can understand without seeing a demonstration.
The brief should fit one page and help the team agree on the problem. It is not an exhaustive specification. OpenAI's use-case guide describes purpose, scope and audience as design inputs. The template below turns that principle into a practical buying document.
Six sections for a useful problem brief
- Observed problem. Describe the current bottleneck with a concrete task and consequence. “Support searches three policy folders before replying” is more useful than “information is fragmented.”
- Affected users. Name who performs the task, who receives the result and who can approve a change. Include the downstream person who fixes mistakes.
- Baseline evidence. Record case volume, total handling time, corrections and delays. State the observation period and whether figures are measured or estimated.
- Desired outcome. Specify a useful change and the quality that must be preserved. Leave numerical targets provisional until the baseline is credible.
- Boundaries. List included request types, permitted information and exclusions. Identify decisions or actions that remain with people.
- First acceptance check. Identify sample inputs, expected results, reviewer and the conditions that would justify further testing.
Add the document owner, version and unresolved questions in a small footer. A brief without an owner becomes an attractive document that nobody updates when the underlying process changes.
A hypothetical brief, worked through
Consider an invented equipment-service company. Its initial request is “buy a chatbot.” A review of one ordinary week instead finds that coordinators repeatedly search approved service policies to answer routine coverage questions. The proposed scope becomes a draft-answer assistant for coordinators, using the current policy set.
The brief states that the service supervisor owns the problem; coordinators review every draft; and coverage exceptions, commercial concessions and technical safety questions remain outside scope. It identifies twenty-five historical inquiries as an initial comparison set, including missing policy information. This is a practical starting sample, not a universal evaluation standard.
The desired result is less end-to-end preparation time without unsupported policy claims. The baseline must include source searching and checking as well as typing. The team records open questions about document versions and which roles may consult each policy. It does not conceal those questions inside a broad requirement such as “secure and accurate.”
Separate facts, assumptions and choices
Label an observed delay as a fact only when there is evidence. Label an anticipated benefit as an assumption. Label a proposed workflow as a design choice. This distinction prevents a supplier's proposal from repeating an untested assumption as a guaranteed requirement.
Ask a colleague outside the project to read the page. Can they explain whose work changes, what inputs exist and how a result will be judged? If they can only repeat a product name, the problem definition needs another pass.
Use the brief to compare responses
Send the same brief to potential suppliers. Ask them to identify missing evidence, propose the smallest test and explain whether an existing feature, process correction or conventional automation could address the bottleneck. A useful response makes assumptions visible and names what it cannot yet estimate.
Keep consequential decisions and review responsibilities explicit, consistent with the lifecycle approach in the NIST generative AI profile. Once the team agrees on the page, develop the detailed project scope. If the page is still contested, a focused AI problem-definition session can help resolve the business question first.