AI use-case discovery

AI use-case discovery is the work of turning a vague goal ("use AI in operations") into a list of concrete, bounded tasks: answer tier-one support tickets about order status; extract fields from supplier invoices; draft the weekly sales summary from CRM data. Each candidate is written down with the input it needs, the output it produces, who checks it, and what happens when it is wrong.

Good discovery starts with the people doing the work, not with a model catalogue. Interviews and screen-sharing sessions reveal where hours go, which steps are repetitive and rule-bound, and which require judgement. Candidates are then scored on three axes: value (hours saved or revenue affected), feasibility (data available, system integrable, decision well defined), and risk (cost of a wrong answer, regulatory exposure). The ranked list feeds the AI roadmap.

It is not brainstorming. A brainstorm produces fifty ideas and no evidence; discovery produces eight to twelve candidates with a rough size, a named owner and a reason each one is or is not first. It is also not a vendor demo, which starts from what a product does rather than from what your staff do.

• Founder / CEO: it replaces "AI strategy" as a slogan with three or four projects you can actually fund and measure.
• Product manager: it gives you use cases described as user jobs with success criteria, which is what you need to write a spec.
• Operations head: it identifies where your team's hours really go, which is usually not where leadership assumes.
• CFO: ranking by return, not excitement, means the first project is the one most likely to pay back.

Companies rarely lack AI ideas; they lack a way to choose among them. Without discovery, the first project tends to be the loudest executive's favourite or the vendor's demo, and it is often the hardest one technically. Discovery exists to make the choice on evidence: which task is frequent, rule-bound, data-rich and low-risk enough to be a first win. The trade-off is that the top-ranked use case is often unglamorous, such as invoice extraction rather than a customer-facing assistant. Teams that accept that usually reach production; teams that chase the exciting one often stall.

• A SaaS support team discovering that 40 per cent of tickets are password and billing questions answerable from existing docs and account data.
• A lending business finding that bank-statement analysis, not chat, is where underwriters lose most hours.
• A D2C retailer identifying that returns-eligibility decisions follow a written policy and could be automated with policy gates.
• An education provider mapping admissions-enquiry handling as a high-volume, low-risk candidate for a student support agent.
• A logistics company finding that delivery-exception communication is repetitive and data-rich while route planning is not the bottleneck.

Technique / practiceA technique combining interviews, process observation and scoring. Eazyware runs it in the first week of Sprint Zero and as part of AI product strategy engagements; the scoring rubric is simple enough for a product manager to reuse internally.

Eazyware service that covers it: AI Discovery Sprint. Starting prices are on the pricing page.