Turn customer feedback into a theme backlog

Convert free-text reviews into a small, stable set of product themes, then count themes without losing the original feedback. This produces an input to product prioritization rather than a sentiment-only dashboard.

Before you begin#

You need a project where you can create transforms and datasets, plus a configured AI provider and a chat model you can use. These are implementation recipes: no live model result or quality level is claimed. The sample contains only synthetic records.

Follow Run a use-case recipe for the shared input, destination, and app steps.

Sample input#

Download the CSV, or copy it into Transforms → Library → Seed → Paste CSV.

feedback_id,comment
F-1,Checkout takes too long and I gave up
F-2,The delivery arrived three days later than promised
F-3,Checkout was quick this time
F-4,I like the blue color but wish the bag had more pockets

Build it in WhoDB#

1. Create Feedback theme proposals, add Seed, and paste the sample.
2. Add AI in Each row mode. Keep input columns and send only comment. Configure the controlled theme and polarity fields below.
3. Paste the instructions and preview all four rows. Preserve one primary theme per feedback record for this first report.
4. Write the row-level result to feedback_theme_review. Review proposed tags before including them in an approved theme report.
5. For the report, add a separate transform with Dataset input, then Aggregate. Group by theme and polarity; add a row-count measure named feedback_count.
6. Display the report alongside a table of the source feedback, so a product owner can inspect what a count represents.

AI output schema#

Instructions to paste#

Assign one primary feedback theme: checkout, delivery, product_design, or other.
Prefer the concrete issue the customer emphasizes; do not create extra themes.
Assign polarity positive, negative, mixed, or unclear from the comment.
If both praise and criticism are present, use mixed.
Return a short quoted evidence phrase. Do not turn a feature request into an existing fact.

Download these instructions.

Check the result#

The targets are checkout/negative, delivery/negative, checkout/positive, and product_design/mixed. The proposed aggregate should contain four groups with count 1 each; the total count must remain 4.

Test a comment that mentions several themes. With one primary theme per record, the report counts records, not every topic mention. Changing to multiple tags changes that denominator.

Find the controls#

These screenshots show the current AI configuration controls used in this recipe. They come from the manual’s synthetic demonstration workspace; they do not show this use case already executed.

Put it to work#

Compare review rates across channels and time periods only after checking collection volume and duplicates. Keep feedback_id so repeated exports do not count the same review twice. Record the theme-list version when changing definitions; otherwise a taxonomy change can look like a trend.

Related: Campaign quality for duplicate and attribution checks.

For output mapping and rerun behavior, follow Write pipeline outputs. For a team-facing screen, follow Create an app and Connect state and data.