Unified Data Lineage
for Salesforce Data Cloud

The product exposed disconnected pieces — never the full picture. Debugging required stitching together data models, transformations, and activations across multiple tabs. Without a clear picture of the entire data pipeline, users couldn't trust what the platform was telling them — and that distrust was a direct threat to Data Cloud adoption.

I led Data Lineage from 0 to 1 as its sole designer, defining not just the experience but the product foundation it would grow from — its scope, MVP, core use cases, and long-term direction. I owned the experience end-to-end: information architecture, graph interaction model, progressive disclosure patterns, side panel, visual language, and the design system tying it together, turning invisible complexity into an explorable, trustworthy system.

From ingestion to transformation to activation, users can now trace their data's full journey, debug issues in context, and understand the impact of any change, without switching tabs or relying on tribal knowledge.

Being the sole designer also meant every major UX tradeoff was mine to investigate and advocate for — including challenging engineering-driven scope reductions and making the case for what research showed users actually needed. That advocacy is reflected in the results below — evidence that betting on what the research showed, not what was fastest to ship, was the right call — turning Lineage into something the business could trust as a foundation for Data Cloud adoption.

Lineage was powerful in the graph — but most users never needed to leave their workflow to get there. I designed an agentic experience that exposes lineage context directly inside Agentforce, so users can understand upstream and downstream impact without switching surfaces.

← →

Step 1 — The Question A user preparing to modify the Customer Transactions Stream asks Agentforce what downstream objects will be impacted. The agent queries the Lineage API in the background.

Step 2 — Lineage in a List The agent surfaces 36 downstream objects — dashboards, activation channels, agents, and reports — as a scannable list. No graph required. Users can act immediately: go to full lineage or contact owners.

Step 3 — Full Context on Demand Expanding the list reveals created-by, business unit, and type for every downstream object. Lineage context — without ever opening the graph.

← Prev Next →

1 / 3

As Agentforce matured, a new class of data entered the picture: unstructured data — knowledge articles, help files, connectors, source citations. These weren't structured records in a data model. They were the raw material that agents used to generate responses. And when an agent gave a bad answer, there was no way to trace why.

I extended the lineage model to include this new class of data, enabling agentic troubleshooting through full pipeline visibility — so users can trace exactly what their agent knew, where it came from, and where things went wrong.

← →

Step 1 — The Entry Point An agent flags a poor response on a support case. The "View in Lineage" button takes the user directly to the pipeline behind that response.

Step 2 — The Agent Pipeline One click from a poor response takes you directly to the execution chain that produced it. Every step — retriever, prompt, action, agent — laid out and inspectable.

Step 3 — Unstructured Objects as First-Class Nodes Knowledge articles, connectors, and file chunks were previously invisible to lineage. Now they're native nodes — and the pipeline behind every agent answer is finally complete.

Step 4 — Tracing the Answer to Its Source Clicking the Source Citation node reveals exactly what the agent cited and where it came from. If the wrong source was used, the owner can pinpoint it here and fix it — preventing the same bad response from happening again.

← Prev Next →

1 / 4