Skip to main content
Context in Practice · On Demand

Inside Mastercard's Context Agents Rollout

Zhenni Hu, Manager of AI and Data Governance at Mastercard, on the scope, evidence, and certification decisions behind enriching 30,000 assets in two weeks. Recorded September 10, 2026.

What you'll see

01

The gap between using AI and trusting it

Zhenni polls the room. Nearly everyone uses AI at work, almost nobody would let it make a decision they are held accountable for. Documentation now feeds AI as well as people, and an agent that cannot find an answer produces one anyway, so the quality of your metadata sets a ceiling on the quality of your AI.

02

Three decisions, and the thinking behind them

Scope held back deliberately, a bounded set of critical assets instead of one click across everything. Evidence reasoned from real usage, lineage, and existing enrichment rather than column names. And certification, a subject matter expert signs off before anything goes live.

03

The results leadership asks for

30,000 assets enriched across roughly 700 tables and views in two weeks, an estimated 6,200 steward hours saved. Roughly half the sampled drafts needed no changes at all. A 12-month effort you celebrate once, a two-week run repeats next quarter without a task force.

04

The stewardship role got upgraded

Less drafting, more judging. A steward's value lives in knowing when something is subtly off and which of three near-identical tables is authoritative, and a certified description gives every downstream consumer, human or agent, a stronger signal than unverified model output.