Research Project Guide: A practical method to surface non-obvious AI use cases

This guide describes a compact, evidence-first method you can run in 1–3 weeks to discover AI opportunities that matter. It focuses on observable problems, measurable outcomes, and quick validation. The method is deliberately lightweight so teams can move from discovery to prototype without getting bogged down in lengthy analysis.

What this project delivers

• A prioritized list of 6–12 candidate AI opportunities with evidence (data signals, process pain points, stakeholder quotes).
• For 1–3 high-potential candidates: a clear hypothesis, success metrics, quick experiment plan, and a go/no-go recommendation.
• A short synthesis report that explains why each candidate matters, what data and integrations are required, and what risks to watch.

Who runs this project

A small triad is ideal: a project owner (product/ops/pm), a researcher/analyst (data-savvy but not necessarily a modeler), and a domain expert (frontline lead or senior practitioner). If you have a librarian or knowledge manager, include them—this work benefits from good question framing and evidence capture.

Core phases (practical checklist)

1. Define the scope and outcome — Pick a domain slice (e.g., returns processing, triage calls, lab sample intake). Clarify the outcome you care about (reduce handling time, increase first-contact resolution, reduce defects). Keep scope small and measurable.
2. Quick data & document scan — Find the obvious data sources: logs, CRM notes, spreadsheets, SOPs, sample documents, and recent incident reports. Note access constraints and obvious gaps. (If you need deeper evaluation of data readiness, run the existing Data & Knowledge Readiness Audit.)
3. Map the process — Use the Process Mapping Worksheet to identify handoffs, decision points, manual steps, and where knowledge is hidden in people or documents.
4. Stakeholder interviews — Run 4–6 short interviews using the provided interview form. Focus on what people actually do, their workarounds, and recent failures—evidence, not hypothetical asks.
5. Synthesis and opportunity ideation — Combine quotes, process friction, and data patterns to create opportunity statements. Use the Opportunity Archetypes reference to spark creative matches without inventing solutions first.
6. Prioritize — Score candidates by evidence strength, expected impact, implementation complexity, and alignment with organizational priorities. Use the AI Opportunity Audit if you need a formal scoring framework.
7. Design quick experiments — For top candidates, turn them into small experiments with clear hypotheses, one or two success metrics, a short timeline, and defined roles.
8. Run and learn — Execute the experiment, capture results, and decide whether to scale, iterate, or stop.

Practical interview and synthesis tips

• Ask for recent examples: "Tell me about a specific case from the last month when X happened." Concrete stories beat hypotheticals.
• Listen for workarounds—these are gold. Workarounds reveal what people value and where tools fail.
• Corroborate claims with simple data checks (counts, timestamps, error logs). If you find a repeating pattern in the data, it’s a stronger candidate.
• Avoid solution bias. Phrase opportunities as problems and desired outcomes, not technology solutions.

Deliverables and templates

Produce a one-page synthesis for each candidate: problem statement, evidence, expected benefit, required data, quick experiment idea, and implementation constraints. Use the Stakeholder Interview and Experiment Planner tools to standardize outputs.

Common mistakes to avoid

• Chasing flashy models without evidence of need or data readiness.
• Ignoring the human workflows that will use or validate the AI output.
• Trying to solve long, complex processes in a single experiment—start small and expand.

Next steps

Start by running the stakeholder interview form and the process mapping worksheet. Once you have 4–6 interviews and a mapped process, use the synthesis steps above to create 1–3 experiment plans and run the Experiment Planner.