Thousands of open-ended replies, turned into a barrier map.
The #1 reason carts almost didn't convert
Barriers, friction, and LTV drivers
Dec 2023 – Mar 2025, coded
The problem.
Echo Water was running a post-purchase survey and getting exactly what open-ended questions produce: thousands of free-text replies. The signal, what almost stopped a purchase, where the friction was, what drove the highest-value buyers, was buried in prose nobody had time to read at scale.
The approach.
I derived themes from the replies using keyword frequency and topic modeling, then segmented them by AOV, product type, gender, age, and landing page, and, in a second phase, by lifetime value and acquisition source, down to which podcasts drove high-LTV buyers. Each segment's replies were summarized with a reasoning LLM so every theme stayed grounded in what customers actually wrote. Two phases, six objectives.
The outcome.
The survey became a ranked, segmented read on conversion barriers and LTV drivers instead of a wall of text. The clearest signal: over 70% of respondents named high prices and unclear or ineffective discounts as what almost stopped them, with unexpected shipping fees and surprise charges close behind. Concrete CRO and pricing direction, pulled straight from the customers' own words.
If your post-purchase survey is a pile of free-text nobody reads, I'll turn it into a ranked barrier map, segmented by who's answering. One slot open for Q3 2026.