Skip to content
Robin Schreiner

UX Management · Product Leadership

← All notes

Think End to End · 3 min read

Can AI find the journeys we missed?

Usage paths can reveal behaviour that does not fit our journey map. AI can help us spot patterns and propose explanations to test with people.

This is part four of Think End to End. The first three articles start with a journey we know about. This one starts with behaviour we did not expect.

We usually draw journeys from research, conversations with a team, and the paths we expect people to take. Then we instrument those paths and watch the numbers. That is useful, but it has a blind spot: what if the journey we chose is incomplete?

People may leave an application, search for a document, return on another device, contact support, or ask someone else to do part of the work. The journey in the dashboard may look like a simple abandonment. To the person, it was a pause in a longer task.

Let the data question the map

AI can help explore large sets of event paths and suggest patterns worth investigating. For example: Which sequences repeat before a successful submission? Are there unexpected loops? Do people who use search before starting complete the form differently? Are there groups whose paths look nothing like the “happy path” we designed?

Product analytics tools now offer AI-assisted exploration of behavioural data; Mixpanel describes agents for investigating changes and behavioural signals. That makes discovery faster. It does not make the explanation automatic.

An unusual path is a candidate journey, not a new persona. A cluster of events is a pattern, not a motive. The model can propose “people are comparing options before committing.” We still need to check whether those events belong to the same people, whether the instrumentation is sound, and whether the behaviour means what we think it means.

Raw data is never the whole experience

Even detailed event logs only contain what the product records. They do not show the document lying on someone’s desk, a conversation with a colleague, a failed phone call, or the reason a person chose to stop. Missing events, changing event names, shared devices, and consent settings can also distort a path. Before asking AI to explain behaviour, I would have a data partner check what the events actually represent and use only data we are allowed to analyse.

Then I would take a few candidate paths back to people: “We noticed that some applicants return several times before submitting. What is happening between those visits?” That question may reveal a missing step in the journey, or show that our original interpretation was wrong. NN/g recommends grounding journey maps in direct research and supplementing them with quantitative data.

From a surprise to an activity

Suppose we find that many successful applicants visit the document checklist, leave, and return days later. Interviews reveal that they need time to obtain a document from another person. Our current journey map assumes they have everything ready when they begin.

Now we have a useful hypothesis: “If we make the requirements clear before the form and offer a reliable way to resume, more applicants will complete it without starting over.” We can test that with people, release a change, and check completion and rework. The discovered path becomes part of the journey we track.

This second path runs in the other direction from the first three articles: behaviour → candidate journey → research → insight → hypothesis → activity. Both paths meet in the same place: a better decision about what to do next.

Previous: When does a signal become an insight?. Related: Can AI do user research without users?.

Keep reading

UX is rather easy.