Customer Feedback Categories: A Codebook With Examples

TL;DR: Customer feedback categories are the labels you sort feedback into so patterns become countable. Use three levels: a type (bug, feature request, pricing, question, objection, praise, complaint), a theme (the part of the business it is about) and a code (one short sentence of what the customer said). Count distinct people per code, not mentions, and route each category to one owner.

Customer feedback categories are a fixed set of labels used to classify every comment, review, survey answer and chat so you can count how often each issue comes up. The standard support-side list, from Intercom's feedback guide, is usability issue, new feature request, bug, user education issue, pricing or billing, generic positive, generic negative, junk and other. That list was built for software support teams. A business that sells on its website also needs sales categories: unanswered pre-purchase questions, objections and requests for things you do not sell. This page gives a full codebook with both, then the steps to analyze customer feedback with it.

What are the main customer feedback categories? (codebook)

The codebook below is the reference. Each row is a feedback type; the example is how it typically sounds in a chat or survey.

Three rules keep the codebook useful:

• One item, one type. If a message makes two points ("love it, but shipping was slow"), split it into two rows. Intercom's guide recommends capturing separate points in separate columns.
• Keep "other" small. If more than about one item in ten lands in "other", there is a category you have not named yet. Read them and add it.
• Fix the list for a quarter. Changing category names every week breaks trend lines.

How do type, theme and code differ?

Type, theme and code are three levels of the same label, from broad to specific. Intercom's analysis method uses exactly these three (Intercom: customer feedback strategy).

• Type says what kind of feedback it is, which tells you who should act (the codebook above).
• Theme says which part of the business it is about: for a shop, Delivery, Checkout, Product range, Returns, Pricing; for a service business, Booking, Quote, Onboarding, Results. Themes are specific to you. Intercom notes that with roughly 50 pieces of feedback or fewer you may not need themes at all.
• Code is one short sentence, true to what the customer said, that someone outside the project would understand: "Wants same-day delivery in Pune", not "delivery".

Codes should be split when they hide two problems. Intercom's example: "Email issues" turns out to be "Email composer bug" and "Email delivery bug", which need different fixes. When you split a code, go back and recode the earlier rows.

How do you analyze customer feedback, step by step?

Analyzing customer feedback means turning a pile of comments into a short, ranked list of things to fix or build. The method below combines Intercom's coding steps with thematic analysis, the six-phase method from Braun and Clarke (familiarise, generate initial codes, search for themes, review themes, define and name themes, report) published in Qualitative Research in Psychology 3(2), 77-101 (Braun and Clarke, 2006).

1. Collect into one sheet. One row per piece of feedback, with columns date, source (chat, survey, review, email), customer_id, page, raw_text, then type, theme and code.
2. Read a sample first. Scan the first 50 rows (about 10% of a 500-row sheet) before coding anything. If they are all about one issue, you will need fewer rows to see the pattern; if every row is different, you need more.
3. Code every row. Add type, theme and code columns. Use the customer's own words in the code where you can.
4. Refine. Merge codes that mean the same thing and split codes that hide two problems. Recode earlier rows.
5. Count. For 100 to 500 rows, Intercom suggests a helper column with a 1 per row and a sum per code; above that, use a pivot table. Count distinct customers per code as well as rows.
6. Rank and route. Build a top 10 list per type (top 10 bugs, top 10 requests, top 10 objections) and send each list to its owner from the codebook.
7. Repeat weekly or monthly with the same categories so you can see whether a fix moved the count.

For a team doing this together, affinity diagramming is the workshop version of steps 3 and 4: write each comment on a sticky note, cluster them, then prioritise the clusters. Nielsen Norman Group suggests 5 to 10 minutes for the note-writing round and warns not to discount a small cluster just because it has few notes (NN/g: affinity diagramming).

The sheet, the people count and the mention count, as exact values you can paste (Google Sheets; column C is customer_id, column H is code):

• date,source,customer_id,page,raw_text,type,theme,code
• 2026-10-28,chat,v_1042,/delivery,"Do you deliver same day to Pune?",Unmet request,Delivery,Same-day delivery to Pune
• People per code: =COUNTUNIQUE(FILTER(C2:C, H2:H="Same-day delivery to Pune"))
• Mentions per code: =COUNTIF(H2:H, "Same-day delivery to Pune")
• Pivot table: Rows = code, Values = customer_id summarised by COUNTUNIQUE, sorted descending

Why count distinct people instead of mentions?

Counting distinct people stops one loud customer from setting your priorities. One person who writes twelve times about gift wrapping produces twelve mentions and one data point. Nine different visitors asking once about same-day delivery is a pattern. Keep both numbers: mentions show intensity, people show demand.

Two biases distort raw feedback counts:

• Extremes speak up. Intercom points out that unsolicited feedback clusters at "amazing" and "appalling", a J-shaped curve like restaurant reviews, while the "it's fine" middle stays quiet.
• Recent and frequent feels important. Intercom calls it "fre-cently" bias: things heard often or lately seem bigger than they are. A fixed weekly count corrects it.

Which customer feedback categories come from sales chats?

Sales chats produce the categories most feedback guides leave out, because they come from people who have not bought yet. Surveys and reviews only reach customers. A website chat reaches every visitor who had a question, including the ones who left.

• Pre-purchase questions show what your pages fail to say. If 15 people per month ask whether a product fits a certain size, that fact belongs on the product page.
• Objections are the reasons people hesitate: price compared with a named competitor, trust ("is this a real company?"), fit ("will this work for a 3-person team?") and timing. Code the objection and the competitor named, if any.
• Unmet requests are things you do not offer: a delivery area, a payment method, a size, a service. These are the most valuable rows in the sheet, because each one is someone who wanted to give you money.

LuckPanda Chatbot collects this for you on your website: every chat is saved word for word with the page it started on, leads come with a summary of the need, timeline and size, and with request logging on, things visitors asked for that you do not offer are grouped on a Requests board, ranked by how many different people asked, with a weekly digest. See how a knowledge base chatbot answers from your content, the live chat metrics to track alongside it, and chatbot use cases for lead generation.

How do you get feedback from customers in the first place?

Getting feedback from customers works best when you combine one prompted source with one unprompted source. Prompted feedback answers your questions; unprompted feedback surfaces issues you did not know to ask about.

NPS is the share of promoters minus the share of detractors, from -100 to +100 (Bain: measuring NPS). It tells you how many people would recommend you, not why; the open comments under it, coded with the categories above, tell you why.