01 / Overview
The wait felt too long.
People were waiting 60–90 seconds for a translator, often during stressful situations. We reviewed over 10,000 chat logs and found a mix of simple requests and cases that needed more care.
90s → <3sInitial wait
25% → 8%Abandonment
02 / Two paths
Match the help to the request.
Quick support for simpler requests.
Spark could handle short translations and clarifications. Sensitive or unclear requests still needed a person who could understand the context.
03 / Human handoff
Keep the conversation going.
Users could ask for a translator at any time. When someone joined, the conversation context came with them so the user did not have to start again.
04 / Interaction detail
Show what is happening.
I designed the processing, confidence, review and completion states. I used Lottie for product micro-interactions and Jitter for the wider motion work.
The shorter wait gave users help sooner and left translators more time for cases that needed their judgement.