Scout24 turns institutional knowledge into trusted AI context with Collate

Each subsidiary posts data to a shared lake with data contracts that describe team, table, fields, types, and PII. But the assets themselves still live in different technologies and formats. Without a common context layer, AI had no reliable way to reason across the organization.

Producers and owners had been expected to document their sources at Scout24, though producing that knowledge by hand and keeping it current was time-consuming for the team. This documentation had also been designed for human users and was not structured for agent consumption. When fed this information, agents returned seemingly trustworthy responses that were incorrect.

Scout24’s previous self-managed catalog required engineers to write descriptions manually, which drove low adoption beyond a small engineering audience. This left the broader data community disengaged, with description coverage thin and falling behind as the estate grew.

Data quality was uncoordinated across the organization, with analysts spending significant time chasing freshness issues, tracking down owners, and resolving discrepancies with no central model and no safe way to create and monitor checks at scale.

Scout24 had made AI-led product innovation a priority and had MCPs running on Starburst so users could query without SQL. But without curated, trustworthy context, those answers couldn’t be relied on. There was no mechanism to make the estate contextual, transparent, or governed for AI agents to consume.

Rather than ask people to document more, Scout24 published every source it deemed legitimate (team communication tools, docs and knowledge bases, code repositories, database systems, file storage, and even external sources) as an MCP on a central, approved MCP registry, including Collate’s own read/write MCP.

An LLM reads from those sources and, through the Collate MCP, generates and enriches documentation inside Collate for assets that had none: table and column descriptions, PII classification, lineage, certification, and usage. Every AI-generated entry carries a visible “AI generated” tag; when a domain or entity owner reviews it, they remove the tag and it becomes original, human-verified documentation.

A single Collate instance spans Scout24’s 10+ subsidiaries, connecting Starburst/Trino, Delta Lake, AWS Glue, PostgreSQL/Aurora RDS, DynamoDB, and MicroStrategy under one discovery layer. RBAC and tag-based policies control visibility by team: analysts see only consumption-ready assets while ETL-internal tables stay restricted, without breaking lineage.

Scout24’s internal data agents are instructed to look for context in Collate before generating SQL. When an agent answers, the end user sees a confidence score calculated from how much trusted context the agent found versus how much it had to improvise. The less context in the trusted source, the lower the score, so users can evaluate the answer for themselves.

Scout24 set out to let anyone in the business get the insights they needed to drive decisions. Now they can: people ask in plain language, and the agents answer from trusted context in Collate rather than guessing. Users get answers directly across the entire company, without queuing for them.

The AI virtuous cycle is live: AI enriches context, humans verify it, and verified context sharpens the next answer. Accuracy climbed once Collate was placed first in the agents’ answer hierarchy, directing them to check the context before building any query. The benefits: more accurate answers, higher confidence, transparency, and governance from a single source of truth.

Assets that never had documentation now get it automatically, drafted by AI and verified by owners, and scattered tribal knowledge across the company is centralized in one place. Coverage grows as the data grows instead of falling behind it, and every entry is either human-verified or clearly flagged as AI-generated.

Within weeks of go-live, Scout24 announced to executive leadership the deprecation of its self-managed legacy catalog. The team also consolidated other data tools into Collate as the new unified context layer for their human and AI data teams.

Governance and context were always needed, but now they serve a new outcome. AI and humans have different needs, and not every source of institutional knowledge is equally trustworthy or well-structured for AI. Collate gives the team one place to govern, tag, and expose only trusted context, whether the consumer is an analyst or an agent.