ProblemStrong features, almost no discovery.

The new navigation shipped two genuinely useful capabilities — global Search and Favorites (pinning). Both convert well when used. The catch: almost nobody found them. When the problem is "nobody finds it," the fix is surfacing, not redesign.

InsightAdoption is a behavior, not a demographic.

Pinning turned out to be a power-user behavior: the more distinct pages someone visits, the more likely they are to pin. Casual users (1–2 pages) almost never adopt; exploratory users adopt many times more. That inflection is where onboarding should intervene.

• Low — 1–2 pages: ~0.5% adoption
• Moderate — 3–10 pages: 8.5–14.2% adoption
• High — 10+ pages: 17.3–35.4% adoption

Design decision: trigger the active nudge once a user crosses ~3 distinct pages — the exact point where adoption leaves the floor.

PrincipleEvidence-based, Jobs-to-be-Done onboarding.

Onboarding should fire on observed behavior, not assumptions, and frame everything around what the user is trying to get done — activities, not RBAC roles. Roles vary across organizations and add ambiguity; intent doesn't. It also stays deliberately lightweight: no Copilot checks, access validation, upsell, or background processes in the flow.

Design decisionsPromote the value — don't rebuild it.

• Promote, don't rework. The features already convert; lead with "here's how to pin and search," not new UI.
• Value before effort. Silently pre-seed a conservative set of default Favorites mapped to the user's job, so the payoff exists with zero effort — inverting the discovery problem.
• Ambient discoverability. A hover-visible pin star on the real page, not a modal tour.
• One just-in-time hint. A single teaching bubble, bound to the actual page at the 2nd–3rd visit ("Pin Devices?"), that disappears after use.
• Surface Search passively for everyone — it already converts ~1 in 3.

Micro-interactionsTeaching inside the real product.

Instead of a modal tour, discovery lives in context — a hover-visible pin star, a single just-in-time hint, and lightweight prompts that surface only when behavior warrants it.

Core flowIntent in, personalized navigation out.

• 1 · Intent. The user states their job in free text ("What are you trying to get done?") or picks from a short list of activities.
• 2 · Mapping. That activity maps to a relevant set of navigation pages — backend logic, invisible to the user.
• 3 · Pinning. Those pages become the user's starting Favorites, ready on first load.

Six triggersBehavior-driven, never interruptive.

Every teaching moment is bound to an observed behavior — and each can be tuned or switched off. Six triggers span the journey from first visit to power use:

1. 1 · Navigation exploration — a one-line teaching bubble on the pin star that disappears after first use.
2. 2 · Successful search — an animated pin pulse on results, with a toast after repeated searches.
3. 3 · High-frequency page — a recommendation to pin pages the user visits often.
4. 4 · Workflow misalignment — a nudge to customize navigation around what they actually do.
5. 5 · Pin churn — an offer to help organize favorites when pins change frequently.
6. 6 · Empty favorites — a subtle personalization hint when nothing is pinned yet.

Constraints & guardrailsKeep it honest and reversible.

• Ten-pin cap. Favorites max out at 10; overflow affects visibility — and ~11.8K users already hit that ceiling, so seeding has to be careful.
• Removable by design. Seeded favorites must be visible and easily removable, validated by a default-favorites retention metric.
• Accessibility & privacy respected throughout.
• Performance wasn't the lever. Median navigation load is sub-second; onboarding ships on top of it.

Outcome & learningSurface value at the moment of need.

The lesson that stuck: when a feature works but nobody finds it, the answer isn't a redesign — it's surfacing at the moment of need. Pre-seed value, keep discovery ambient, and teach once, in context. The MVP stays tight: default favorites, passive search discovery, ambient pinning discovery, and a single contextual teaching moment.