A unified design language for five B2B apps, and the first time AI entered the design process.
As a design intern embedded in Huawei's B2B department, I proposed design improvements grounded in real user behaviour across three projects covering user identity, structural optimisation, and cross-app consistency.
In B2B, identity is everything. Yet on most platforms, users are reduced to a name and a job title. How do you bring offline trust online without overcomplicating the product?
An app often serves multiple professional roles simultaneously, whose needs vary greatly. Some roles spend most of the day facing clients; others work across departments and regions. My product team and I analysed user types and typical interaction scenarios, and two findings emerged:
Most B2B platforms are straightforward and restrained, prioritising efficiency over expression. A good online experience should preserve, rather than strip away, the trust and emotional connection people have offline, bringing that sense of connection back without adding complexity.
First-hand research: through workshops and user interviews, I identified key identity-verification scenarios: first meeting, internal collaboration, client meetings. Secondary research: how "trust" is visually established across professional contexts, and the tension between genuine ability and social stereotype that shaped our subsequent design choices.
Presented to the product team across four occupations and two card types compatible with all five apps, helping users build a stronger first impression and faster trust online.
What I learned: in B2B design, "feeling" is not optional; it determines whether a tool is tolerated or trusted. Most B2B platforms underestimate this.
In the second phase, I moved to a large-scale enterprise-level web system serving users with multiple roles. This phase focused on structural design: simplifying information hierarchy and improving efficiency based on real behavioural data and user interviews.
Behavioural data: click-through rate, dwell time, bounce rate. User interviews: what's convenient, what's confusing, what do users expect.
An evidence-based classification that replaced intuition with a data signal and a behavioural hypothesis for every design suggestion. Each pattern maps straight to an optimisation strategy, not a guess.
| Pattern | Data signal | Behavioural hypothesis | Optimisation strategy |
|---|---|---|---|
| Dispersed entrances | The same function appears on multiple pages, and clicks on it are scattered. | The user's goal is unclear, or different pages share irrelevant entry points. | Reorganise the entry path to reduce redundancy. |
| High click-through rate + short dwell time | Users exit immediately after entering the function. | Accidental touch, or a misleading function name. | Verify through interviews and adjust labels or layout. |
| Low click-through rate + long dwell time | The feature is hard to find, but once found, is used extensively. | An important function is visually underestimated. | Enhance visual hierarchy and entry-point placement. |
| High bounce-rate pages | The task was interrupted midway. | A lack of clear task-completion feedback loses the user's confidence. | Strengthen completion feedback and guidance for the next step. |
What the framework looks like in use. Each read below starts from a data signal, not a preference, and lands on a different kind of fix.
Alongside the previous two projects, I maintained design consistency across all five B2B applications: analysing each app's business structure, identifying inconsistencies in shared components, and aligning the design team on a unified standard. This work is quiet, without dramatic before/after comparisons, but it's the foundation that makes everything else feel coherent.