Voice of Consumer as a standing program
Customer signal was scattered across seven systems and five functions. Everyone had dashboards. No one had a shared account of what was actually going wrong, or which problems were worth solving first.
Led qualitative and behavioral insight synthesis across four priority brands, translating fragmented data into clear opportunity areas. Partnered across Brand Experience, NA eCommerce, Customer Service, Digital Experience Activation, and Supply Chain so insights informed both experience design and operational priorities.
- Built a friction measurement framework in Fullstory to quantify struggle across PDP, PLP, cart, and checkout
- Connected behavioral signals to experience breakdowns
- Integrated post-purchase insight through Bizrate and custom Looker dashboards
- Fullstory, GA4, Bizrate
- Talkwalker, Sphere
- Customer service contacts
- Google Search Console
Turning seven data sources into one monthly account of what to fix.
- Promotional confusion was the highest-frequency pain point across brands, driven by unclear stacking rules, SKU exclusions, and sample eligibility
- Vague error messaging and missing real-time feedback compounded it
- Users attempted retroactive promo application, driving appeasement requests, reships, and service contacts
- Separated cross-brand patterns from isolated brand issues
- Prioritized opportunities by impact and frequency rather than by who raised them
- Aligned product, CX, and operations on one source of truth
- Translated raw signal into executive decisions with named owners
Behavioral signal to executive decision, every month
Fullstory quantifies frustration through rage clicks, dead clicks, and drop-off, identifying high-friction journeys.
Bizrate structures qualitative feedback into themes and maps sentiment by journey stage.
Behavior and voice connected to root causes, then built into a narrative by brand, category, and journey.
Insight translated into actions with cross-functional alignment on sequence and ownership.
Most of the signal was the
instrument, not the customer.
March put frustration events at 45% across North America brands, four points above February, and the program had to explain it. I ran the deep dive across seven brands. Three defects in how frustration was captured accounted for most of the score, and none of them described anything a customer experienced.
Real friction, and where each finding went
Definitions governed. Findings
assigned to named owners.
Running monthly across four brands meant the measurement itself had to be stable enough to argue from, and the output had to land somewhere with a name on it.
One published sentence. Eleven records behind it.
The program worked. What it did not do was keep anything. Every month the raw consumer language was synthesized into a narrative, the narrative shipped, and the substrate underneath it had nowhere to go.
Accurate, actionable, and gone. It routes a ticket, then expires with the site build it describes.
“The site kept saying ‘coupon could not be applied’ when it really meant that you were sold out of the drykit.”
“I used MOTHERSDAY code, but it doesn't show 16-piece gift in the order.”
Plus nine more across Estée Lauder, MAC, Origins, and Aveda. Not confused about the rule, distrustful of what the interface said.
Two lanes off one pipeline.
The readout does not change.
Frustration events and dead click URLs describe a build that will not exist in a year. Consumer language does not decay on that schedule. Splitting them by shelf life lets the operational loop stay fast without discarding the material that holds.
Operational stream
- Metrics, frustration events, defects
- Monthly cadence, named owners, live tracker
- Expires with the build it describes
Consumer reference library
- Verbatims, contact reasons, category framing
- Continuous ingest, rolling 18 to 24 month window
- Provenance on every record, queryable after publish
What a persona should have been
The site kept saying “coupon could not be applied” when it really meant that you were sold out of the drykit. When I entered oilykit, the coupon was accepted.
Tensions: promotion opacity, error legibility · recurred Mar–May
She decoded the real cause herself. She remembers the system as dishonest, not as out of stock.
Grounded panels, and the discipline that makes them defensible
A library dense enough to ground synthetic respondents is worth more than an off-the-shelf panel, because the grounding is what determines whether the output holds. The discipline matters more than the capability.