Why categories decide everything

How you categorize customer feedback decides what you can see. Categories that are too broad hide problems; too narrow, and nothing has enough volume to compare. Categories named after departments rather than customer experiences leave gaps where customers' real concerns fall between teams.

Use three levels

  • Theme: a broad area of the experience customers recognise. Staff, Billing, Product, Store environment, Delivery, App.
  • Sub-theme: a distinct part of that area, usually with a clear owner. Under Billing: Waiting time, Errors, Payment options.
  • Specific issue: a recurring problem or praise described in customers' terms. Under Waiting time: Only one counter open at weekends.

Themes are for leadership reporting. Sub-themes are what owners track. Specific issues are what gets fixed. Themes and sub-themes should be fixed in advance; specific issues are best discovered from the feedback itself.

Building it

How to build a review taxonomy from your own reviews.

How to build a review taxonomy

What makes a good category

  • Customer language. "Waiting at billing", not "POS throughput".
  • One idea. "Staff and cleanliness" is two categories.
  • Clear boundaries. A one-line definition says what is in and what is out.
  • Examples. Two or three real comments that belong, and one that looks similar but does not.
  • An owner, at sub-theme level.
  • Enough volume to matter, but not so much that it hides several different problems.

An example category card

Sub-theme: Billing > Waiting time. Definition: The time customers spend queuing or waiting to pay, from joining the queue to completing payment. Includes: queue length, number of counters open, slow processing at the counter. Excludes: errors in the bill (Billing > Errors), waiting for staff on the shop floor (Staff > Availability). Examples: "Only one counter open on a Saturday." "Waited 25 minutes to pay." Owner: Head of Store Operations.
Category card

Tag mentions, not reviews

A comment can belong to several categories with different sentiment. Tag each mention separately: "Lovely staff but the billing queue was endless" is Staff (positive) and Billing > Waiting time (negative). Tagging whole reviews to one category forces you to drop one of them.

Allow unclear

Some comments carry no topic: "ok", "nice". Mark them unclear rather than forcing them into a category. Count them so you know how much of your feedback they are.

Free review snapshot

Want this done for your own reviews?

No commitment. We use your details only to send your snapshot and follow up. Privacy policy

Test the categories

  1. Take 200 recent comments and have two people tag them independently.
  2. Where they disagree, the definition is unclear. Fix it.
  3. Look at what landed in "Other". If it is large, you are missing a category.
  4. Look at the biggest categories. If they mix several different problems, split them.

Changing categories over time

Categories need to evolve, but every change breaks a trend line. Add specific issues freely as new problems appear. Change sub-themes a few times a year at most, and record the date. Change themes rarely. When you change a category, reclassify historical feedback if you can, so before-and-after comparisons stay meaningful.

A starter set for retail and restaurants

Every business should build its own categories, but a starting point helps. These themes cover most store and restaurant feedback; adjust the sub-themes to your business.

  • Staff: friendliness, knowledge, availability, behaviour.
  • Waiting: queues, billing, service speed, appointments.
  • Product: quality, range, availability, freshness or fit.
  • Price and value: pricing, offers, value for money.
  • Billing and payment: errors, payment options, invoices.
  • Returns and exchanges: policy, process, refunds.
  • Store environment: cleanliness, layout, comfort, facilities, parking.
  • Delivery and online: order accuracy, timing, packaging, app.

Then let specific issues emerge from your own feedback under each sub-theme.

Common mistakes

  • Department names as categories. Customers do not complain about "Operations".
  • A huge "Service" bucket that hides five different problems.
  • A large "Other" category that nobody reviews.
  • Categories that overlap, so the same comment lands in different places depending on who tags it.
  • Sentiment as a category. "Complaints" is not a topic; it is a sentiment about a topic.

Categorising at scale

By hand, one person can tag a few hundred comments a day with care. Beyond that, use software, but keep the same discipline. Give the model your category definitions and examples, not just names; require it to point to the words that justify each tag; send uncertain cases for review instead of guessing; and spot-check a sample every month against a person's tagging. Automation should apply your categories consistently, not invent new ones silently.

Doing this with Falcon

In Falcon, each brand has its own taxonomy of themes and sub-themes, which can be generated from a sample of the brand's reviews and then reviewed. Every review is classified against it at the mention level with sentiment, specific issues are discovered by clustering what customers wrote within each sub-theme, and each theme can carry an owner.