SURVEY ANALYSIS · UX RESEARCH
Analyzed the month over month satisfaction trends of a large media company, rebuilding how a premium subscription measures customer value.
ROLE
UX Research Intern · Summer 2026
TIMELINE
5 weeks
METHODS
Surveys · Thematic Analysis · Quantitative Trend Analysis
01 / PROBLEM
A large media company I interned for in the summer of 2026 has an all access premium subscription plan, which runs an ongoing satisfaction survey to understand why people subscribed and how sentiment moved over time. Responses were sorted into established drivers like content quality, relevance, ease of access, device reliability, support, value for price.
One reporting cycle surfaced a contradiction. Customers still rated the content quality and relevance highly. But their sense that the premium tier was worth paying for was slipping.
If the content was good, why were people questioning the price?
The existing framework of the survey couldn't answer that. While covered a high level overview of where problems appeared, it failed to help us identify how those problems surfaced and worked together.
02 / APPROACH
Reading through the open-ended responses, I kept hitting comments that refused to sit in one bucket. Someone complaining about ads was also describing a broken expectation. Someone describing a login failure was also describing eroded trust. Tagging each comment into a single predefined category was actually destroying important patterns.
So instead of forcing the data into the existing scheme, I rebuilt the coding segments around what the responses were actually saying. I identified an opportunity to alter the survey so it would lead to more impactful findings.
This meant running two layers of analysis: the standard driver-level reporting stakeholders expected, plus a second-layer synthesis that explained the relationships between drivers.
03 / PROCESS
01 - Familiarize with the metrics
Reviewed satisfaction trends and driver performance from survey data (blurred for privacy) to locate where sentiment was actually shifting.
02 - Read past the buckets
Went through open-ended responses looking for repeated tensions that cut across categories, rather than tagging into predefined ones.
03 - Rebuild the coding scheme
Restructured the segments when overlapping themes broke the existing categories.
04 - Name the cross-cutting themes
Identified three that explained the relationship between frustration and perceived value: expectations, habits, and trust.
05 - Connect to the business question
Mapped the new themes back to declining value satisfaction to show customers weren’t judging value of the premium subscription on content alone.
06 - Build the narrative
Translated the analysis into a story stakeholders across growth, strategy, engineering, product, and design could argue with, act on, and use.
04 / ARTIFACTS
See my finished report below!
05 / IMPACT
Customers were actually judging value through an interconnected system of expectations, habits, and trust instead of content quality alone.
That explains something the driver-level view couldn't: a product can have strong content perception and still lose its value through other avenues.
I presented across three stakeholder forums to 60+ cross-functional partners the new framework I had developed. It changed the question in the room from "which driver scored low" to "how are several drivers interacting to erode value."
Opportunities identified
Clarify the premium promise of the all access plan before and after purchase; support existing routines across formats and devices; reduce operational friction around reliability and support; treat trust as a product problem, not only an editorial one.
Recommendations from the work drove a full redesign of the satisfaction survey itself, so future cycles could capture these relationships directly and more distinctly, filling in unclear gaps in the data and speeding up analysis.
06 / LEARNINGS
1) Question the categories you're handed. The most useful thing I did on this project was notice that the framework did not encompass all patterns before I dug into what the data said. Everything else followed from that.
2) Overlap is a signal, not a mess. My instinct at first was to pick the "best fit" bucket for each ambiguous comment. The comments that deviated from the existing classifications were the ones carrying the actual insight.
3) Descriptive reporting doesn't change decisions. Explanatory research does. Nobody acts on "value for price is down four points." Teams act on "here's the mechanism producing that number, and here's where each of you sits in it."
What I'd do next: validate the three themes with findings from other research departments, and test different survey questions to see which formats would tell a story and get the most detail out of the data.