What customers complain about most: grouping reviews into themes

A theme appearing 60 times and a theme appearing 4 times lead to two completely different decisions about what to fix first. So after reading reviews, the job is to count, not to conclude.

The 40-minute process

Step 1 — get the data. Google Maps reviews — 15 credits, 200 reviews, runs in the background — for your venue and one or two nearby competitors.

Step 2 — read the first 40 reviews and name your own themes. Do not use a pre-made taxonomy. Theme names should be the customer's words: "long wait", "too salty", "parking", "staff not attentive", "pricier than expected". Usually five to eight themes emerge.

Step 3 — read the rest and count. Each review goes into one main theme, or into "other". If "other" exceeds a third, your theme names are wrong — go back to step 2.

Step 4 — split by star rating. The same theme in a one-star review and in a four-star review have different severity.

Step 5 — record the date and keep the table to compare next time.

The result table

Theme	1–2 stars	3 stars	4–5 stars	Total
long wait
not like the photo
parking
price against quality
staff

The 1–2 star column decides the fix order. A theme appearing often there is what stops customers coming back.

Choosing what to fix first

Rank on two criteria, not one:

Frequency — how many people mention it.

Cost to fix — replacing photos is nearly free, changing the kitchen process is not.

Start with high frequency and low cost. For many venues that is photos and descriptions: the "not like the photo" theme is usually in the top three and fixable in an afternoon.

Comparing with competitors

A competitor's table gives you two things. A theme they are criticised for and you are not is your strength — say it in your content. A theme both of you are criticised for is an area-wide problem, and whoever fixes it first wins.

How often to repeat

Quarterly, and after any menu or process change. Compare to the old table: a theme that shrank means the fix worked.

One limit to state plainly

Review writers skew to the extremes: very happy or very unhappy. Most ordinary customers write nothing. Your table cannot measure overall satisfaction — it finds work to do, and that is enough.

The tool returns only a display name and the review text, with nobody's contact details.

Frequently asked questions

Is there a tool that groups themes automatically? CIA returns the reviews; naming themes is better done by you, because the right names depend on the kind of business.

Are old reviews still worth counting? Separate reviews from the last 12 months. A problem from two years ago may already be fixed.

Should I count reviews with no text? No. They only affect the average rating, not the theme table.