Discoverability at Scale: Rebuilding BP’s Metadata Experience
Client: BP
Role: Product Designer
Team: 1 Product Designer, 1 Service Design Lead
The challenge: making data not just technically available, but humanly accessible.
BP’s enterprise data platform had one critical mission: to make datasets discoverable and usable across thousands of employees. But for both engineers and business users, it had become a bottleneck. Metadata was scattered, search results were opaque, and access flows felt broken.
Strategic groundwork: How We Reimagined the Grid
Across interviews and heuristic reviews, the friction points became tangible:
Analysts spent hours chasing datasets across silos, often finding duplicates or incomplete copies.
Business users couldn’t navigate technical jargon, so they leaned on data engineers for even the simplest queries.
Access requests were opaque: they submitted a form, then waited sometimes weeks with no visibility.
Metadata itself lacked trust signals. Users didn’t know if what they found was current, certified, or reliable.
The outcome: frustration, low adoption, and wasted hours in a company where speed and accuracy matter.
Design Principles
Simplify
- Speak in business language, not technical shorthand.
Contextualize
- Surface quality, freshness, certification at first glance.
Empower
- Clarify ownership, make access requests transparent.
Integrate
- Embed into existing enterprise workflows and tools.
Anticipate
- Use AI to suggest relevant data and streamline bulk actions.
Designing for AI: from raw fields to trusted context
One of the biggest challenges in enterprise data was the constant noise around quality, duplicate datasets, missing owners, and incomplete fields created endless queries back to data teams. By introducing AI-driven metadata enrichment, we aimed to minimize data quality issues caused by manual work, reduce support overhead, and give users immediate confidence in what they were seeing. Features like overlap detection, owner suggestions, quality signals, and autofill were added to reduce errors and make trust and accountability clear at the point of use.
Reduce duplication and confusion
AI detects overlapping datasets and surfaces merge/deprecate actions directly in context. This prevents clutter and makes the catalogue easier to navigate.
Make ownership transparent
Suggested owners are displayed with confidence levels, helping users quickly identify accountability without relying on guesswork.
Minimize repetitive metadata work
Inline AI autofill suggestions streamline fixing missing or incomplete fields, reducing manual edits and keeping metadata consistent.
Surfaces trust indicators
AI suggestions include quality signals such as High Confidence, making it clear how reliable the information or action is.
Final Reflection: Designing for Explainability
Designing for explainability helped bridge the gap between raw data and user trust. By turning complex metadata into clear summaries, surfacing trust signals, and reducing repetitive manual tasks, the design made it easier for both technical and business users to work with data confidently. The main lesson was that automation alone isn’t enough, people need transparency and context to trust what they see.