Databricks’ Ad-Genie standardizes marketing metrics—but Meta conversions still need a local contract

Databricks has released Ad-Genie, a deployable accelerator that turns Lakeflow Connect advertising data into normalized silver and gold tables, three Unity Catalog metric views, an AI/BI dashboard and a curated Genie agent. The launch post says the bundle targets Google Ads, Meta Ads and TikTok Ads, with governed metric views supplying the same KPI definitions to both the dashboard and conversational interface. Databricks launch Implementation tutorial

What the bundle standardizes

The tutorial shows the useful part plainly: source adapters rename platform-specific identifiers into a common schema, convert Google’s cost_micros into a standard spend field, normalize status values and union the results into campaign, ad-set and ad fact tables. Metric views then define measures including impressions, clicks, spend, return on ad spend, click-through rate and cost per acquisition across shared dimensions. Implementation tutorial

That architecture gives the natural-language layer a better starting point than asking a model to reconcile three advertising schemas at query time. Databricks’ metric-view documentation says measures are defined centrally and can be queried from SQL editors, notebooks, dashboards, Genie Agents and alerts; the query engine generates the computation for the dimensions selected at runtime. Metric-view documentation

Deployment is also concrete rather than conceptual. The published guide packages Ad-Genie as a Declarative Automation Bundle, requires Unity Catalog, serverless compute, a SQL warehouse and partner-powered AI features, and offers profiles ranging from pipeline-only to the full dashboard-and-Genie stack. Implementation tutorial

The conversion trap

The accelerator’s own changelog exposes the most important production caveat. In version 0.2.0, Meta Ads conversions default to 0.0 because Meta stores actions and conversion values in nested arrays and each advertiser must decide which action types count as a conversion. The changelog says configurable action-type mapping is planned for a future release. Ad-Genie changelog

That means a technically successful deployment can still produce a business-wrong answer to questions such as “Which campaign had the best conversion rate?” or “What was our return on ad spend?” if the local conversion contract has not been implemented. Before exposing Genie to marketers, teams should define the accepted Meta action types, currency and attribution-window rules, then compare the resulting metric views with each platform’s native reports for a fixed set of campaigns and dates.

The repository includes a test-harness directory and documents pipeline setup, but the launch tutorial does not publish an answer-quality benchmark for the Genie room. GitHub’s repository metadata also reports no detected license even though the tutorial describes the accelerator as open source. Teams should therefore treat the bundle as an inspectable starting point, not a drop-in proof of semantic correctness or a completed licensing review. Test-harness directory Repository metadata Implementation tutorial

The practical rollout gate is simple: do not test only whether Genie returns SQL. Test whether the shared metric contract returns the same answer the marketing team already accepts—and make Meta conversion mapping the first case.