Case Studies

Production planners had no way to view or edit routing data across multiple routings at once — every field-level change had to be made one routing, one field, at a time. There was no prior tooling for this in the product.

Designed a comparison view that surfaces routing data for multiple routings side by side, then layered a mass-edit interaction on top of it — letting planners select fields across routings and apply a single change everywhere at once, rather than repeating the same edit by hand.

Shipped as a production capability spanning 150+ editable fields across multiple routings simultaneously — the first mass-maintenance workflow the product has had. Delivered roughly 6 months after joining the manufacturing domain, following an earlier AI-powered routing search capability shipped in the first 2 months.

Master data — Material Master, Routing, BOM, and Production Version — is the foundation every downstream manufacturing application depends on, but it's split across four separate applications with no unified view. Gaps and inconsistencies for a given material or product go unnoticed until they cause downstream failures.

Designing a single view that brings all four sources together, surfaces gaps and inconsistencies through trend and AI-based analysis, and suggests concrete fixes rather than just flagging problems. Ran validation with 20 customers across 5 iterations — their feedback directly shaped what "material readiness" should mean, which factors it depends on, and what information customers need to see: the gap itself, its downstream impact, and the hurdles to resolving it.

Customers responded with strong, specific signal that the concept solves a real problem. Findings from this research are directly shaping a new application, currently in development.

Agent design is new enough that most designers aren't sure what actually needs to be defined for an agent — the important decisions are about boundaries and behavior, not how the agent looks on screen, and there was no framework guiding designers through those discussions.

Built a playbook that walks designers through the right questions to ask and the right details to define — an agent's boundaries, behavior, and decision logic — then generates artifacts tailored to whoever needs them next: pitch material for senior decision-makers, specs for development teams and architects, scenarios for testers, and guidance for language experts and user-assistance teams.

Validated with design and development team members; currently used by 5 designers and 1 product manager ahead of a planned internal rollout.

Newly built agents get evaluated for technical correctness, but not for whether interacting with them is actually a good experience — no framework assessed user-facing risk, bias, or appropriate behavior before an agent reached production.

Building an evaluation agent that reads a target agent's domain and context, applies established UX guidelines for both screens and agent interactions, identifies short- and long-term risks and user biases that could come into play, and defines the appropriate agent behavior across scenarios. It then generates test scenarios from that analysis, runs them against the target agent, and records the results.

In testing as an internal "agent gate" — the goal is a holistic report on experiential quality and concrete improvement recommendations, so agents are evaluated for being experientially usable, not just technically capable, before reaching production.

There was no clear way to tell whether a routing — or a BOM, Material Master, or Production Version — was actually complete enough to send to production. The fields that need filling aren't standardized; they depend on material type, plant, customer manufacturing process, best practices, and dependencies, so they can't simply be marked mandatory. In practice, people filled in what was required and added the rest from experience, or only after a downstream escalation flagged something was missing.

Defined the vision and capabilities, at the empowered-product-team level, for a proactive/reactive agent that analyzes trends across these factors — material type, plant, customer process, and their dependencies — and determines whether a material is production-ready, rather than waiting for a person to notice a gap. Scoped to start with routing data first, since that's directly in the team's control, with the approach designed to expand across the other three master-data objects and be handed to their respective teams to customize.

Not yet shipped — currently in active design and definition at the empowered-product-team level, ahead of implementation.