How Unity Accelerated Internal Revenue-Generating Processes from Days to Minutes
Run Unity Ads' Huron platform, covering 23B+ monthly ad impressions and 2.5B+ players across three time zones.
From Slack request to compliance verdict, down from a multi-day manual process, the first agent built on the new AI context layer.
Surface at least one issue, such as schema gaps, missing PII tags, or weak descriptions, before shipping.
Gaming / Technology
Apache Flink, Apache Paimon, Google BigQuery, Great Expectations, Cursor, Slack
From documentation to infrastructure: metadata that AI can call
The shift underneath everything Unity built is a change in what metadata is. Traditionally it is documentation, accurate only if a human keeps it current and useful only if a human reads it. Unity turned it into infrastructure. The Collate MCP Server exposes Unity's governed metadata (schema conventions, glossary definitions, lineage) as a live, queryable interface that any AI agent or workflow can call directly, rather than a person hunting through docs. That single shift, from something engineers read to something systems call, is what makes every capability below possible.
Every Unity Ads data product needs a review to ensure accurate billing and avoid PII violations. That review used to be manual, one engineer at a time across Copenhagen, Montreal, and San Francisco, losing days to round trips, with around 80% of reviews surfacing at least one issue. Now a request in Slack triggers Dax to check schema conventions, PII tagging, glossary coverage, and description quality, all pulled live through the Collate MCP Server, then post a verdict back to the thread in about seven minutes. The eight-person team keeps pace with 100% of data products routing through the platform, with no added reviewers as volume grows.
When a pipeline incident hits, ads stop serving and revenue stops with them, so resolution is time-sensitive. The first 20 to 25 minutes typically went to manual archaeology: tracing what fed the affected table, who consumed it downstream, and where the pipeline was defined. Unity is extending the same callable-metadata model here. When an incident channel opens, Dax is being built to surface upstream producers, downstream consumers, and pipeline definitions automatically, replacing the multi-person scramble with an instant, governed response.
Because the governed metadata is callable by any AI tool or workflow, each new use case builds on the same foundation instead of rebuilding it. Unity can layer on automated documentation, SQL generation grounded in governed definitions, and data-quality test creation, all working from the same live context that already powers compliance and incident response. Once metadata is infrastructure rather than documentation, the cost of the next AI use case drops.
Why Unity moved from self-hosted OpenMetadata to managed Collate
Huron's governance began on open-source OpenMetadata. Unity kept the open standard it had standardized on and moved to managed Collate to power its AI context layer while shedding the operational load an eight-person team could no longer justify carrying. What the move delivered:
The Collate MCP Server turns governed metadata into a live, AI-ready interface, the foundation every use case above depends on, delivered as a managed service rather than something the team has to build and run.
Self-hosted OpenMetadata means provisioning, scaling, and tuning search clusters, databases, and ingestion pipelines as usage grows. Managed Collate lets Unity redirect senior engineers to governance and platform work instead of upkeep.
Self-hosted deployments own their upgrade cadence, where a fix in one version can introduce an issue in the next. Managed Collate ships only tested, stable releases, insulating Unity from version churn.
Where typical open-source data-quality tooling surfaces only a pass/fail percentage, Collate's alerts return the failing rows themselves with a sample and a debug query, so engineers can see exactly what broke and why.
Why it matters for data leaders
The strategic takeaways beneath the compliance win:
Governed metadata exposed as a callable context layer means AI agents and tools work from schema, glossary, and lineage that are correct by design, not from a copy that drifts out of date.
The first agent built on the foundation cut compliance review from days to about seven minutes, on 100% of Unity Ads data products.
The same eight-person team keeps pace with rising volume, while managed Collate carries the operational load that self-hosting would demand.
