Customer insights: if it changes no decision, it is a fact

A customer insight is a finding about why customers behave the way they do that the team did not already know and that changes a decision. Plenty of what gets labelled customer insights is really data or findings: true, often useful, and not yet able to tell anyone what to do differently. Below is one worked example of the difference, then where insights hide, how to write one so a team acts on it, and how to test it before you bet a quarter on it.

1. Data says what happened, a finding says what repeats, an insight says why and what to change

Nielsen Norman Group splits research output into three levels. In Sara Paul's article for NN/g, data is the raw, unanalyzed material (transcripts, notes, metrics, survey output), findings are the patterns across it, and insights are focused explanations of opportunities that draw on other research and business context. Findings tell you what happened. Only insights tell you why, which is what a recommendation needs.

Her example is a Net Promoter Score (a survey question asking how likely people are to recommend the product) for a redesigned medical appointment booking site. The new score looks higher than the old one, but the margins of error overlap, so the change may be noise. The insight comes from adding usability testing: people struggled with medical jargon and with picking the right specialist, which points to plainer wording and ties to the company's goal of fewer support calls (NN/g).

Here is a made-up example that we will use for the rest of the post. Priya runs product at a scheduling tool for small dental clinics, with a 14-day free trial.

Notice that the finding invites the obvious response, which is a louder "invite your team" prompt. The insight says that prompt would annoy managers who are deliberately waiting. Same numbers, different decision.

Short version: if a sentence could be true and the roadmap would stay the same, it is a finding.

2. If the team already knew it, it is not an insight

The Wikipedia entry on customer insight lists four qualities: an insight is non-obvious (it usually takes several sources of evidence converging), actionable, able to change customer behavior, and good for both the customer and the company. The first two do most of the work in practice.

Run every candidate through two questions. Would a smart new hire have guessed this in their first week? Would anyone on the team argue for a different plan after hearing it? If the answers are yes and no, keep digging.

Here are three statements that show up in readouts and fail the test:

• "Customers want it to be easier to use." Everyone has always wanted that, so it changes nothing.
• "Price is a top reason for churn." This is a finding. It skips why the price stopped feeling worth it.
• "Users love the new dashboard." This is a reaction. It does not explain what they do with it or what they stopped doing.

Each can become an insight once someone asks why. "Price is a top reason for churn" might turn into "clinics cancel after their first slow month because the invoice arrives the week cash is tightest." That one suggests moving the billing date, which nobody had considered.

3. Look where customers explain themselves without being asked

Behavior data tells you what people did. It rarely tells you why. The why usually lives in places where customers are talking in their own words, often while annoyed, and nobody is steering them with a survey question.

If you already collect a lot of this, sort what you have before asking for more. The voice of customer post covers how to set that up as a routine, so this post will stay on the insight itself.

There is one more reason to favor these sources. A 2022 piece in Harvard Business Review by Ian Kahn, Paul Leinwand and Mahadeva Matt Mani argues that much of the customer data companies buy is also available to their competitors, so it rarely produces a distinctive understanding. Your own support queue is data nobody else has. (Your competitor's support queue is someone else's problem, thankfully.)

4. Write the insight as one sentence a team could argue with

Write each insight as a single sentence with four parts: who, what they do, why they do it, and what that means for a decision. Then add one line of evidence underneath.

Here is the template, followed by Priya's version:

• Who does what, because why, so we should change this decision.
• Office managers at single-dentist clinics start trials alone because they want proof before the monthly owner meeting, so we should let single-user trials run until after that meeting and give managers a one-page summary to bring to it.
• Evidence: the trial numbers in section 1, plus 6 of the 8 managers Priya called saying they planned to show the owner "at the next meeting."

Three rules keep the sentence honest. Name a specific group, because "users" covers people who want opposite things. Put the cause in the customer's world (a meeting, a budget cycle, a boss), since causes inside your product are usually findings in disguise. Point at a decision someone owns, because NN/g's advice is that an insight should narrow the options toward a direction to test, rather than prescribe a single design (NN/g).

Keep the evidence line short and linkable. If the team wants the full argument, a customer research decision memo gives each claim a record, a counterargument and a next test. The analysis that produces the sentence in the first place is its own skill, covered in how to analyze customer interviews.

5. Test it with a different kind of evidence before anyone builds on it

An insight is a hypothesis with good manners. Before it changes a roadmap, check it with evidence of a different kind from the evidence that produced it.

NN/g calls this triangulation: combining methods so one method's blind spots are covered by another's strengths. Kathryn Whitenton's article gives pairings such as interviews that suggest an unexpected purchase motivation followed by a survey that measures how common it is, and analytics showing high error rates followed by a look through support records. The post on combining interviews with product analytics goes further into that pairing.

The strongest test is a prediction you have not checked yet. If Priya's insight is right, a few other things should also be true:

1. Single-user trials that did convert should cluster just after the start of a month, when owner meetings happen.
2. Managers who were offered an extension should invite the dentist later in the trial, close to that meeting.
3. Clinics where the dentist signed up personally should not show the same gap.

Pick the prediction that is cheapest to check, and decide in advance what result would make you drop the insight. If the converted trials are spread evenly across the month, the meeting story is wrong, however good it sounded on the calls.

Then run the smallest real change. Priya could offer a 30-day trial to half of new single-user clinics for one quarter and compare conversion against the other half. That costs two weeks of free trial per clinic, which is cheaper than a redesign built on a guess.

6. Count an insight as done only when the decision changes

An insight that nobody acts on is a nicely formatted fact. NN/g's Brian Utesch and Tammi Fitzwater call the loss between research and delivery "breakage." They list the usual causes: teams cherry-pick the recommendations that match existing plans, rework them into easier fixes that miss the cause, or let them sit with no owner after the readout. Their advice is to track adoption as an outcome in its own right.

The practical version is small. Give every insight an owner and a decision date, and attach it to a ticket or a planned test. A month later, check whether the decision changed. If the answer is "we agreed it was interesting," it went into the drawer with the others.

Who this does not fit

If you have fewer than about ten customers, you do not need an insight process. Talk to all of them, and treat what you learn as hypotheses to check in customer discovery rather than insights to file.

It also fits poorly when the decision is already made and the research is there to decorate it. A team that has committed to a launch will hear any insight as an objection. Say so, and save the research for the next decision.

Your next step

Take the last research readout your team shared and rewrite its top three points using the who, what, why and decision sentence. Any point that will not fit the template was a finding, and now you know what question to ask next.

Sometimes that question needs a person you cannot easily reach, such as the practice owner who never joined the trial. Instant Expert lets you invite specific professionals, including consumer insights experts and user research experts, to a short call or a written or voice answer, and you pay only if someone books or answers.