Skip to main content

Guide: build your topic plan from the customer journey

The 7-step method to get a taxonomy that reflects the stages of your customer journey.

F
Written by François Forest

Goal: by the end of this guide, your verbatims are classified by the stages of your customer journey. For each stage, you know what your customers are talking about and where the pain points are.

Who it's for, and before you start

Who it's for

The person setting up Text Analysis, with the Admin role (required to create and publish a taxonomy).

Prerequisites

Your feedback sources are imported into Feedier.

To prepare

A description of your customer journey, even a rough one.

Time

One to two hours, including review.

The details of each screen are in the article Text Analysis: taxonomies, segments and topics.

The steps

1. List the stages of your journey

Write down the 5 to 8 main stages your customers go through, in order. Each stage will become a segment.

Stay in the customer's point of view: "Booking" rather than "Sales department".

2. List the subjects to track for each stage

For each stage, write down the subjects you want to be able to act on. Each subject will become a topic. Aim for 5 to 10 topics per segment, with neutral names: "Delivery time" rather than "Delivery too slow".

3. Generate the taxonomy with AI

In Text Analysis → Taxonomies, click + then Create with AI.

Field

What to choose

Analysis type

Customer journey

Existing topics to include

Plain text, pasting your list of stages and subjects

Number of segments

Your number of stages

Sources

The sources related to this journey

Visibility

The teams that track this journey

Click Generate draft. See Create a taxonomy.

4. Review the draft in the Taxonomy Lab

The draft is tested on a sample of 1,000 verbatims. Check three things:

  • Each segment does match a stage of your journey.

  • No topic captures a disproportionate share of verbatims. A topic that's too broad is worth splitting.

  • The untagged rate stays low. The Untagged column suggests missing topics.

5. Clarify the rules of ambiguous topics

Open topics that look alike or that use your internal vocabulary, and add MUST and MUST NOT rules. If the examples from the last test are right, don't add anything. See Write a topic's rules.

6. Validate, then publish

Move to the Validation step to check the result on 10,000 verbatims, then publish.

Publishing analyzes your existing verbatims and uses analysis budget. Publish a finished version rather than several successive versions.

7. Keep the taxonomy up to date

Track the Untagged rate at the top of the Text Analysis page. If it rises, new subjects have appeared. Group your adjustments and go back through the Taxonomy Lab once a quarter, rather than with every idea.

Example

An airline structures its taxonomy into seven segments that follow the passenger's trip:

Segment (stage)

Example topics

Booking and pre-flight

Ticket prices, website navigation, seat selection

Airport and boarding

Check-in counters, baggage drop, boarding organization

Cabin and comfort

Seat comfort, legroom, cleanliness

Food and entertainment

Meal quality, crew attentiveness, onboard Wi-Fi

Arrival and baggage

Punctuality, delayed baggage, lost baggage

Customer service and disruptions

Cancellation handling, refunds, phone wait times

Loyalty program

Account management, award tickets, status benefits

At a glance, the team sees that disruptions and customer service concentrate the most verbatims, and which topics weigh the most there.

Common mistakes

Mistake

Why it's a problem

What to do

Putting sentiment in the name

"Unpleasant welcome" excludes positive reviews about the welcome.

Name the topic "Welcome".

Creating too many topics

Very narrow topics overlap and become unreadable.

Start broad, split later.

Mixing journey and internal organization

Segments no longer match a clear stage.

Create two taxonomies: one by journey, one by business area.

Publishing too often

Each publication uses budget.

Test in the draft, publish when it's stable.

Going further


Last reviewed: November 11, 2026, version 4.23.5.

Did this answer your question?