Research findings and working tools answer different questions. A finding asks whether something is true; a tool asks whether something is usable, at the moment a person needs it, inside a workflow they already have. Prototyping is how a team learns the second answer cheaply.
Translate the finding into a moment of use
Write a single sentence in this form: "When [person] is doing [task] and needs to [decide something], the tool shows [output] so that they can [act]." If the sentence is hard to write, the gap is usually that the finding has not been mapped onto anyone's real working day.
Choose the smallest thing that can be judged
Each of these can be produced in days. The purpose is to provoke a specific reaction from a specific stakeholder, not to demonstrate technical capability.
- A clickable screen flow, when the question is about workflow fit.
- A single working calculation on real data, when the question is about credibility of output.
- A printed or emailed mock output, when the question is about whether anyone would act on it.
- A spreadsheet operated by a human behind the scenes, when the question is about demand.
Use real data early, in a safe form
Synthetic data hides exactly the problems prototypes exist to reveal: missing fields, inconsistent coding, unexpected distributions, and edge cases that dominate real practice. Where governance allows, use a small, de-identified real extract from the start, and treat data handling as part of the prototype rather than a later concern.
Test with the people who would actually use it
Five sessions with intended users will surface more than a large formal review. Watch what they do before commenting on what they say, note where they hesitate, and record every question they ask — questions mark the places the output is not self-explanatory.
Decide what the prototype proved
Close each prototyping round with a written decision: continue as designed, change the design, change the user group, or stop. A prototype that does not end in a decision quietly becomes a permanent demonstration, which is the most expensive possible outcome.
Name the moment of use, build the smallest artefact that can be judged, use real data safely, test with actual users, and end each round with a written decision.