Comcast's commerce engineering group faced a familiar BI bottleneck: unclear requirements, slow report cycles, dashboard bloat, and conflicting metrics across tools eroded confidence and delayed decisions.
The team reset its approach by consolidating analytics on AI-powered business intelligence on the Databricks Platform, pairing a governed semantic layer with conversational analytics and automated delivery. For a broader look at how organizations are rethinking reporting and self-service, the complete guide to business intelligence and analytics in the AI era covers the key patterns driving this shift.
The result was a 90-day execution plan that replaced weeks-long cycles with days, created a single source of truth for KPIs, and packaged self-service insights in an experience leaders actually use.
Why traditional BI broke, and what needed to change
Three recurring patterns kept value out of reach. First, the "one more chart" spiral turned focused dashboards into clutter, burying golden KPIs beneath incremental slices that different stakeholders requested over time. Users could not find what already existed, and requests repeated.
Second, there was an art-versus-engineering gap. Data engineers can deliver against specs and tests, but data storytelling requires a different craft. Forcing engineering teams into narrative work led to long feedback loops and little closure.
Third, tool sprawl and redundant pipelines meant the same metric often appeared with different values across multiple BI platforms. Leaders lost trust, and engineers spent cycles reconciling numbers and maintaining parallel data flows.
The 90-day execution playbook on the Databricks Platform
The team rejected incremental fixes and executed a staged plan anchored in Unity Catalog, Genie, and a shared semantic layer. In week one, they validated AI/BI Dashboards and Genie against a manually loaded dataset to prove the experience.
By day 30, one domain had been replumbed into a medallion architecture in Unity Catalog, with a certified metrics view as the backbone for dashboards and Genie. Conversational querying was productionized for stakeholders.
Days 31 to 60 focused on automating dashboard development using Genie Code. By day 90, the team packaged dashboards, Genie, and a structured feedback loop into a branded app, turning feedback into requirements that flow straight into engineering.
A packaged, governed self-service experience
The internal e-commerce insights app replaced scattered links and Slack requests with one destination. A personalized home lists accessible dashboards, and a marketplace lets users discover new ones and request access through a governed workflow. A standard template keeps each dashboard tight: a readme, a small set of golden KPIs, and a few supporting trends to avoid cognitive overload.
When users need more, they ask Genie. Questions like "slice by region," "compare periods," or "change to a line chart" happen conversationally without SQL or analyst queues. Structured feedback in the app creates a Jira ticket with full context. An agent built with the Databricks CLI and open-source agent skills performs impact analysis, proposes a spec, and generates draft changes to the metrics view, dashboard, and Genie space. Engineers review and publish. Enhancements that previously took three or four days can now ship in minutes.
Impact: faster cycles, stronger trust, fewer tools
Development timelines shrank from weeks to days, and engineers avoided BI narrative work that sat outside their core skill set. Unity Catalog centralized test-certified KPIs so leaders stopped debating whose number was right. Consolidation also set the stage to retire redundant BI tools, reducing both license and operational costs.
Turning backlog into leverage with one governed semantic layer
Consolidating data, semantics, dashboards, and conversational AI on a single governed foundation changed the team's contract with the business. Leaders get quick, trusted answers. Engineers ship faster against a stable metrics layer. The blueprint is less about any one feature and more about removing friction end to end.
