Skip to content
Robin Schreiner

UX Management · Product Leadership

← All notes

Think End to End · 3 min read

When does a signal become an insight?

Feedback, interviews, and analytics can point to the same problem. They can also tell different stories if we ignore the context.

This is part three of Think End to End. The first article sets up the journey; the second gives it a measure of success.

A dashboard shows a drop at the document-upload step. A support agent says people keep asking which document to use. An interview participant calls the instructions confusing. It is tempting to put all three on a slide and write: “Users cannot understand the upload instructions.”

Maybe. But we should slow down long enough to check what each source actually tells us.

The dashboard shows where a behaviour happened. The support agent hears from people who contacted support, not everyone who struggled. The participant tells us about one experience. Together, they make a stronger case for investigation. They do not yet prove one cause.

Keep the evidence attached to the claim

I find it helpful to separate three levels:

For the application journey, an insight might be: “First-time applicants interpret ‘proof of residence’ differently from the service team. Several bring a document that the form rejects, and the event data shows repeated uploads at this step.” That is much more useful than “the upload has friction.” It gives us a group, a mechanism, and a place in the journey. We would still check how widespread it is before treating it as the whole story.

The hypothesis follows: “Showing accepted examples before people start will reduce rejected uploads and increase complete submissions.” Now we have something to act on and measure.

Ask what would change your mind

Before calling a pattern an insight, I would ask:

  1. Does the observation repeat across people or sources, or could it be one unusual case?
  2. Do we understand who is affected and under which conditions?
  3. Is there another explanation? A technical error, an eligibility rule, or a tracking change can look like a usability problem.
  4. Can we trace the claim back to a quote, behaviour, support record, or event definition?
  5. Would learning that the claim is false change the action we plan to take?

These questions are not a demand for perfect certainty. They help us decide whether to make a small, reversible change, run a focused study, or invest in a larger redesign. The stronger the decision, the more important it is to understand the evidence behind it.

Do something, then listen again

Insights are useful when they change what the team does. We might clarify the document list, change the upload instructions, or test a different point in the journey. Then we watch the primary success metric, the relevant lever, and the guardrails. We also return to applicants and support colleagues to see whether the problem actually changed.

If the numbers improve but people still feel unsure what they submitted, we are not finished. If interviews sound positive but completed applications do not improve, we need to understand the gap.

The goal is a loop: signal → investigation → insight → hypothesis → activity → observed outcome. It is not a pipeline that turns every comment into a feature request. Sometimes the best action is to learn more.

Previous: One journey, one success metric, several levers. Next: Can AI find the journeys we missed?.

Keep reading

Can AI find the journeys we missed?