How we cut survey analysis from 2 days to 20 minutes

RMillington

Sharing this because our old process was painful and a few people asked.

Before: every month one analyst exported NPS verbatims to a spreadsheet, hand-tagged them into themes, and built a deck. Two full days, and the themes drifted depending on who tagged them.

What we changed:

1. Set up Loop's auto-theming on the open-text field, then spent an afternoon correcting the first batch so the model learned our categories (billing, onboarding, performance, support speed).
2. Built one dashboard filtered to detractors only, grouped by theme, with a trend line per theme.
3. Saved it as a shared view and pointed leadership straight at it.

Now the monthly "analysis" is reading the dashboard and writing three bullets. About 20 minutes. The one thing I'd do differently: spend longer on step 1. The first month of auto-themes was rough until we corrected enough examples. Happy to share the theme taxonomy if useful.

RMillington

This is great, thank you. Would love the theme taxonomy. We're still hand-tagging and the drift problem you describe is exactly ours, two analysts produce two different sets of themes from the same verbatims.