The right answer, surfaced mid-sentence

Stage 1 — Problem Identified

The answer gap is measured in seconds

Call shadowing made the core constraint concrete: whatever we built had to deliver value inside a ~3-second window, hands-free, without pulling the rep's attention from the client.

Stage 2 — Problem Scoping

One surface, ruthlessly scoped

We scoped v1 to the live-call panel plus a minimal knowledge-base manager — no CRM sync, no analytics suite. Every feature was tested against 'does this help in the 3-second window?'

• →Latency budget agreed with engineering before high-fi design
• →Knowledge-base quality flagged as the real success dependency

Stage 3 — Solution Shaping

Designing for peripheral vision

I prototyped card anatomies, motion timing and information density until suggestions could be parsed without reading — bold answer line, muted source line, calm entrance animation that never startles mid-sentence.

Stage 4 — Impact Testing

Simulated calls, real pressure

We ran mock client calls where testers had to use surfaced answers live. Cards that animated too eagerly got ignored or resented; the calm version got read. Confidence thresholds were tuned so the panel stayed silent rather than guessing.

• →Measured glance duration and answer usage per call
• →Iterated timing, density and suppression rules

Stage 5 — Solution Deployed

Shipped with the trust layer intact

The live product pairs every suggestion with its source and freshness, keeps the panel dockable beside any call window, and gives teams a simple flow to keep the knowledge base current — because stale answers kill the product.