Persona-driven user simulation

How it works

Scripted tests follow the path the designers intended; real users don't. A persona-driven agent is given a profile (a first-time user in a hurry, an expert who uses keyboard shortcuts, someone who distrusts forms and abandons them easily) and a goal (book a return, change the billing address, export last year's data). It then drives the product through the UI, choosing actions in character, and keeps a log of its reasoning. The useful outputs are the points where it hesitated, backtracked, misread a label or reached a dead end, and the differences in path between personas. Several runs per persona are needed, because each run is different.

When to use it

Early usability checks before real users are available; comparing how a redesign changes the path through a task; checking that common tasks are completable at all after a change; and generating realistic behaviour sequences for other kinds of testing.

Limitations

A model playing a confused user is not a confused user — its difficulties come from different places, and its successes don't prove a human would succeed. Personas tend towards stereotypes, and simulating users with disabilities this way is no substitute for testing with them. Runs are costly, slow and vary from one to the next, so conclusions need many runs. Findings are suggestive at best and need confirming with real people.

What you still own

Writing personas grounded in real user research, and treating every finding as a question to check with real users.

Cite this