Loading work...
Apple was paying a third-party vendor for every experiment its internal marketing teams wanted to run, and the dependency was getting in the way of how the company actually wanted to test. The studio led UX on Truth Serum, an Apple-native A/B and multivariate testing platform built on Apple Directory for permissions and surfaced through Apple Connect for admins, with a calendar dashboard, gated experiment setup, IP-granular audience builder, and multi-variant WYSIWYG editor with live preview.
Picked up mid-stream. Audited prior work, reviewed competitor tools (Adobe Test and Target, Optimizely), and grounded the product in the marketing team's existing testing patterns. Started at the end: reviewed prior results, then worked the system backward until the design served both business and user.
The top three questions on arrival were: which tests are running, how long are they running for, and where are they running. The calendar dashboard was chosen so this context is visible before anything else. Anything that did not serve those three questions on entry was pushed deeper.
Testers inside Apple run hundreds of experiments a week. Experiments frequently pause mid-setup when a higher-priority test comes in, so drafts were built as a first-class surface where tests can be started, stopped, and edited without penalty, without losing state, and without cluttering the live calendar view.
An approved offer going out globally is expensive. Experiment setup was designed as a gated five-step flow (metadata, audience, offer, variants, review) that forces users through specific approvals before launch. Asset requests and permissions route through the same approval path, so the gate is the product, not a layer on top of it.
The marketing team kept bumping into missed work because no one knew who owned a task. Every permission in Truth Serum routes through Apple Directory. Managers control access, track ownership, and approve via the same identity their teams already log in with. No parallel auth system to maintain, no second source of truth for who can do what.
Apple gets into the weeds when testing what works in each market. The audience builder was designed down to IP-block granularity through a drag-and-drop attribute palette so non-technical marketers could compose precise segments (country + device + logged-in state + cart value + IP) without writing rules or asking engineering.
Traffic allocation is a core A/B primitive but does not need complexity. A single slider that visually controls the percentage of traffic to each variant was faster to learn than a numeric field, impossible to enter invalid values, and matched the mental model marketers already had from prior tools.
An Experiment contains one or more Variants. A Variant can contain multiple pages but cannot mix test modes. The variant editor enforces a single mode (A/B or MVT) per experiment so users cannot accidentally ship inconsistent tests or mis-attribute results to the wrong control.
Variants change pixels on a live product. The WYSIWYG editor ships with a browser simulator so users can preview each variant across Safari, Chrome, Firefox, and Edge before launching.
Design ran as a dedicated lead engagement due to project secrecy. The mitigation was access: embedded directly with the marketing team that would use Truth Serum, which made requirements, validation, and iteration a same-day loop instead of a weeks-long artifact cycle.
Adobe Test and Target and Optimizely were the two closest competitor tools. Both served as grounding for what marketers expected in an experimentation product, then departed where Apple's constraints (Directory permissions, global storefront, IP-level audiences) demanded Directory-native gates and IP-block targeting.
An internal, Directory-permissioned A/B and multivariate testing platform designed for Apple's marketing team.
A complete design system and product designed for Apple's internal marketing team, built around existing Apple Directory permissions and approval flow.