The 5 Signals You Have the Right Number of Feedback Themes

Most mature feedback programs settle at roughly 30 to 50 themes, and most teams that ask "how many themes should we have?" are either well under 15 or well over 150. Both failures look fine from a dashboard. Fifteen themes produce a clean chart where every bar is too broad to act on. Three hundred produce a long tail where the same problem is counted in four places and no single bar looks urgent. The count is a symptom. The useful question is what the distribution of feedback across those themes is telling you.

You have the right number of feedback themes when five signals hold: the catch-all stays small and boring, no single theme carries an outsized share of volume, almost every theme earns feedback every quarter, each cross-cutting problem exists exactly once, and every theme has one owner. Each signal is measurable from data you already have, and each one points to a specific fix when it fails: merge, split, or add structure.

What a theme structure actually needs to get right

Theme count is an output of four properties. Score your structure on these before you count anything.

1. Granularity that matches a decision. A theme is the right size when one team can act on it in one planning cycle. Bigger, and it hides several problems. Smaller, and it fragments one problem into numbers too small to win a roadmap argument.
2. Structure learned from the feedback. A theme list written in advance encodes what the team expected customers to say. An adaptive taxonomy derives the theme layer from what customers actually said, so the count tracks reality rather than a planning meeting.
3. Size measured in consequence, not mentions. Twelve mentions from enterprise renewals can outweigh 300 from free trials. A customer context graph attaches account, segment, and revenue to each theme, which is what lets you keep a low-volume, high-ARR theme separate instead of merging it away.
4. Stability over time. If the structure changes substantially every month, trend lines break and nobody trusts the counts. The theme layer should grow with your product surface, not churn with your feedback volume.

The right number is whatever count satisfies all four. For most B2B products, that lands in the 30 to 50 range at the theme level.

The 5 signals you have the right number of feedback themes

1. The catch-all stays small and boring

Pull the "Other" or "General" bucket and read 100 records. If most of them belong in an existing theme, your theme names are too narrow. If many of them describe the same new problem, you are missing a theme. The healthy state is a small bucket full of feedback that is genuinely unclassifiable: one-word replies, off-topic comments, spam. See what an overloaded catch-all bucket means for how to read the contents.

Fails when: you have too few themes. Fix: add the missing theme, or rename an existing one so it describes the problem rather than one symptom of it.

2. No single theme carries an outsized share of volume

Rank themes by volume. If one theme holds a large share of all feedback, it almost certainly contains several distinct problems, usually two to five. "Performance issues" at 18% of volume is not a finding. It is a category masquerading as a theme.

Fails when: you have too few themes in one area. Fix: split by root cause, not by customer vocabulary, following splitting a theme that has grown too broad.

3. Almost every theme earns feedback every quarter

Count the themes that received fewer than a handful of records in the last 90 days. A few is normal: a resolved bug, a seasonal issue. A long tail of near-empty themes means the structure is more granular than the feedback. Those themes never win a prioritization review because their numbers are too small, even when the problem they describe is real.

Fails when: you have too many themes. Fix: merge near-empty siblings under the problem they share, and keep the detail as sub-themes if you need it for engineering.

4. Each cross-cutting problem exists exactly once

Search for a problem that touches several product areas, such as "export times out" or "permissions are confusing." If it appears as a separate theme under three different parents, the count is inflated and the volume is split three ways. Enterpret research on AI-generated taxonomies, built by an off-the-shelf coding agent on two proprietary feedback corpora, found zero nodes shared across branches, against 48 and 73 in the production taxonomies. Concepts that cut across the tree get minted once per branch, so their true size never shows up in one place.

Fails when: you have too many themes, in a way that hides problems. Fix: consolidate into one theme and let product area be an attribute of the record. Finding duplicate themes in your categories covers the search.

5. Every theme has one owner

Map each theme to the team that would fix it. If a theme maps to three teams, it is too broad. If one team owns 40 themes that all lead to the same backlog item, the structure is more granular than the decisions it feeds. Theme count should roughly track the number of distinct problems your teams can pick up, which is why routing customer feedback to the right teams and theme design are the same exercise.

Fails when: either direction. Fix: split themes with shared ownership; merge themes that always resolve to the same ticket.

Why theme count is the wrong number to optimize

Teams that set a target ("we will have 40 themes") end up forcing the feedback to fit the number. The count should fall out of the five signals, not drive them.

The bottleneck is maintenance, not design. A structure that passes all five signals in January drifts by June, because every release creates new surfaces and new failure modes. Hand-maintained taxonomies drift toward too few themes, since adding structure takes effort, and feedback about new things lands in the catch-all where signal 1 eventually catches it. Generated structures drift toward too many, because each run invents its own labels, which is the pattern explained in why zero-shot categories change between runs.

A structure that is learned from the feedback and persisted between runs holds the count steady while still adding themes when genuinely new problems appear. That is the property worth paying for, and it matters more than the number itself.

How to run the check

Run all five signals quarterly and after every significant launch. Signals 1 and 2 tell you where you have too few themes. Signals 3 and 4 tell you where you have too many. Signal 5 tells you whether the structure maps to the organization that has to act on it. If the signals disagree, trust signal 5: a theme count that matches how your teams make decisions beats one that matches a benchmark.

FAQ

How many themes should a customer feedback taxonomy have?

Most B2B teams land between 30 and 50 themes at the theme level, under six to ten top-level categories. The exact number matters less than the distribution: a small catch-all, no dominant theme, few empty themes, and one owner per theme.

How do I know if my feedback themes are too broad?

A theme is too broad when it holds a large share of total volume, when its records split evenly between praise and complaints, or when it maps to more than one team. Reading 50 records from it usually reveals two to five distinct problems that should be separate themes.

How do I know if I have too many feedback themes?

You have too many when a long tail of themes receives almost no feedback each quarter, when the same problem appears under several parents, or when many themes resolve to the same backlog item. Merging near-empty siblings is usually the fastest fix.

Should the number of themes grow as feedback volume grows?

No. Theme count should grow with your product surface, such as new features, integrations, and segments, not with volume. More feedback about the same problems should make existing themes larger, not create new ones.

How does Enterpret decide how many themes to create?

Enterpret's adaptive taxonomy learns the theme layer from your feedback and keeps it stable between periods, adding a theme only when genuinely new problems appear rather than re-deriving the structure each run. Its customer context graph ties every theme to accounts and revenue, so small but high-value themes stay visible instead of being merged away.

If your theme count keeps drifting, see how Enterpret's adaptive taxonomy keeps the structure current without a rebuild.

‍