Session Ratings
A lightweight feedback loop that helps customers measure and improve the experience of every video call.
Whereby offers two video calling products: a standalone tool for individuals and small teams, and an embeddable product that lets companies build video calls directly into their own services. In telehealth, education, and financial services, the embeddable product powers sessions that feel native to the customer's platform.
Session Ratings gives those customers a lightweight way to capture user sentiment at the end of every call — turning subjective experience into something measurable and improvable.
I led design across discovery, the rating experience, and the analytics layer customers use to act on it.
From One Ask to a Pattern
The opportunity started with a single customer. They needed to identify which participants had hit technical issues, and during which calls — a clean, solvable request. Before touching pixels, I wanted to understand the landscape. The deeper I looked, the more the same need surfaced, phrased differently each time: telehealth providers monitoring session quality, education platforms watching host performance, customers in a proof-of-concept phase wanting a quick way to gather feedback during a trial. Diagnosing failures, evaluating hosts, building confidence during onboarding — the feature kept finding new use cases the more we explored.
That reframed the work. This wasn't a one-off feature for one customer; it was a feedback loop that could carry across customers without becoming bespoke each time. The design decision was upstream of any pixels: solve once for the pattern, not many times for the request.
Stars, Not Smileys
The mechanic centres on a five-star rating. Stars over thumbs, NPS, or emoji — because the primary context is telehealth, and emoji felt too casual for a moment that often follows a clinical conversation. Stars carry the right register: professional, familiar, low-effort. I wanted the act of rating itself to feel good, so the stars use gradient and motion — fun, but quiet about it. Delight without disrespect for the context.
The flow is asymmetric on purpose. A satisfied user passes through quickly. A dissatisfied user — one to three stars — sees a short list of common issues: poor audio, poor video, audio/video sync, disconnection, poor internet, blocked devices, device compatibility, an issue with a feature, or "something else." The list came from product instinct and support patterns, then I proof-checked it with customers. "Something else" opens a free-text field as an escape hatch for anything the list misses. Two ratings ran in the prototype: one for call quality, one for the host.
A Floor, With Headroom
Most of our customers had no formal feedback mechanic before this, and that shaped the dashboard. A clean average score — per room, and overall — is a meaningful step up when the baseline is nothing, so that's where the dashboard leads. From there, customers can drill down: overview to room, room to individual session, session to the participant's rating and tags. Same data, at the depth they actually need it.
I held back on more sophisticated mechanics — weighted scoring, custom metrics, customer-defined tag sets. Build the floor people can stand on first. The complexity is genuinely useful, but only once a customer has lived with the simple version long enough to know what to ask for. That headroom is the next phase of work, not this one.
Scoping for Clarity
Two ratings made it into the prototype: call quality and host quality. Only one made it into the release. Call quality landed — the mechanic worked and the tag list resonated across almost every account, with small additions here and there but the same shape holding. Host quality didn't. Each customer wanted it slightly differently — different tags, different flow logic depending on who was in the room, different views on when to even ask the question (every session, only the first, after some threshold of familiarity). There was no shared answer to design toward.
That's not a "needs more iteration" problem — it's a "no shared answer exists yet" problem. We made the deliberate choice to de-scope host feedback. The core experience shipped without it: focused, reliable, immediately usable. Host feedback stays on the roadmap, backed by real demand and the research already done. When we return to it, we'll return with clarity.
Beta & Honest Read
We ran a closed beta with explicit go/no-go criteria — engagement, conversion, and whether the feedback distribution matched the way we already see customer sentiment. All three came through cleanly, and the numbers gave us confidence to release.
- 68%
- 30%
- 4.6
- Normalised
The more interesting signal came afterwards. Enterprise customers liked the feature but tended to build their own — feedback ties directly to their KPIs, and they have the engineering capacity to do it. The customers actually using Session Ratings are the ones who didn't have either, which reframed the audience: this is for teams that need a reliable starting point, not for organisations already running their own analytics stack. Frequency is the next thing to watch — we ask after every session today, but rating fatigue is a real risk in telehealth, where the moment right after a session isn't always a great moment to ask anything. Some things are still being figured out, and that's the honest read after a closed beta.
