Turning a confusing first run into a guided first success — so new users feel Copilot's value before they're asked to trust it.
The first run leaned on generic prompts and silent failures. Most new users never sent a meaningful prompt — and most never came back.
Organic vs Inorganic acquisition across iOS & Android paid segments
Source: Historical app usage data from M365 Copilot Mobile telemetry
Only ~26% of new iOS users — and 19.5% on Android — ever became chat-active. The funnel was leaking at the very first step.
"If the first interaction feels useful, users come back. If it doesn't, they won't."
So the whole first run was rebuilt around a single moment: the user's first prompt succeeding — and a Week-1 target of every new user submitting one.
Work-graph prompts for M365 subscribers; quick-win prompts for consumers. Never generic suggestions.
Eligibility and setup run in the background. Users reach chat only when Copilot is ready.
Dismissable at every step. Trust earned through quality, never forced engagement.
Early flows and iterations — working through structure, density and hierarchy before landing on the final direction.
Perceived latency turns a technical wait into a moment of personalization. The hero prompt arrives pre-filled and contextual — built to succeed. The response confirms Copilot understands the user's world.
Two design patterns were on the table — Treatment A and Treatment B. To find out what actually worked before committing, we shipped both into a small A/B test with 5% of users, then finalized the pattern based on what moved the metric.
Set direction and pressure-tested the spec with the PM.
Grounded every decision in telemetry and prior studies.
Concept to ship — flows, prompt design, iOS & Android.
Engineered first prompts for first-time success.
Presented at milestones; defended rationale to senior leaders.
Read live metrics weekly to validate and iterate.