This article was co-written by Erika Patrikainen, Data Scientist, and Heidi Hämäläinen, Data Consultant at Zure.
Microsoft has announced 12 November 2026 the retirement of Purview Data Governance, with support ending on 12 November 2027. OneLake Catalog becomes the center of Microsoft’s data governance.
Whenever a governance platform is retired or significantly changed, the first question organisations ask is usually:
“What should replace it?”
It is a reasonable question.
We just do not think it is the most important one.
After spending the past few years working on Data and AI Governance, including Microsoft Purview Data Map and Unified Catalog, it is clear that organisations should first look at what they have already built. Over time, governance work creates business definitions, ownership information, lineage, data products, classifications, rules and relationships between business concepts and technical assets. Some of this is technical metadata, some of it is business related metadata, and some is governance content. All these contribute to the context that is needed to interpret data correctly. The platform gives us a place to manage it, but the investment is also in creating, connecting and maintaining that content. Before deciding what technology comes next, we should understand which parts are worth preserving and remember that now we can serve AI agents as well.
The value of this shared business and governance context becomes obvious in data governance projects. We may start by discussing data sources, scanning, classifications, lineage, governance domains and metadata structures. The more interesting challenges appear when people discover that they do not actually agree on what the data means. What is a customer? What exactly counts as revenue? Who owns a concept used across several business units? We have seen organisations realise that resolving these questions takes time, and that it must come before configuring the platform. We see that as a prerequisite rather than an optional activity: assumptions that previously lived in meetings, documents and people’s heads become explicit definitions, ownership and relationships that others can use – and they evolve, so they need to be maintained.
AI gives us another reason to preserve and improve this content. An AI agent can find a table called Customer, but finding the table is the easy part. The agent still needs context: which definition of Customer is approved, which source should be trusted, what exceptions apply and who owns the concept. It may also need to understand how Customer relates to contracts, products, invoices and revenue. A database schema alone does not answer those questions. Business definitions, relationships, ownership, lineage and rules can provide much of the missing context. Governance content and metadata can therefore become part of the context AI systems use to interpret organisational data.
Where to define the context then? Since most of the business terms are not specific to a certain data asset or data platform, they should be created and governed outside the data platforms, e.g. on a metadata platform or even as simple text files to start with. The context does need to be bound to the data for the AI, and for that Fabric IQ has the Ontology feature. Business context should be served there: for example, the meaning of "Customer", the relationship to other entities, such as "Customer purchases Products", what type of Products there are, and where the data is located.
Increasingly, this context may not only be consumed inside a single governance platform or even within one organisation. Industry developments such as Microsoft’s Fabric IQ Sharing suggest that business context itself is becoming a reusable and portable asset. Definitions, relationships, ontologies, instructions and governance knowledge can be shared across teams, platforms, partners and AI systems. This makes the quality and maintainability of governance content even more important than the platform used to store it.
This is why we would not start a Purview transition by comparing feature lists between Microsoft technologies, Databricks, Atlan, Collibra, Informatica or another catalog. We would start with an inventory of what already exists. Which business definitions are actively used? Which assets have meaningful ownership and lineage? Which data products, rules and classifications matter outside the current platform? Which relationships between business concepts and technical assets have already been created? At the same time, we would identify what is missing or disconnected: definitions with no link to data assets, ownership that exists only on paper, business rules hidden in project documentation, or important context that still lives in people’s heads. This gives us a much clearer picture of what should actually move forward.
The next question is what this content needs to support in the future. Analysts need trusted definitions and enough context to understand the data they use. Data product teams need ownership, rules, quality expectations and relationships they can maintain consistently. AI agents increasingly need governed context when retrieving and interpreting organisational data. If we were designing this foundation today, knowing what we now know about analytics, data products and AI, we doubt many organisations would simply recreate yesterday’s catalog. We would instead focus on how business context, governance content and technical metadata connect, how they are maintained and how they can be made available to the people and systems that need them. The technology should follow those requirements, not define them.
So before selecting the next governance technology, start with these four questions:
What business content and metadata have we already created?
- Refer to Purview content to identify which definitions, ownership information, lineage, rules, data products and relationships are described.
- If ownership and business term definitions are not defined yet, we recommend common workshops for the business and data experts to build a solid data governance foundation.
- Reusable, high-quality data products can only be created on top of a working data governance process. Figure out with us what these could be in your business.
Which parts are valuable independently of Purview?
- Determine what should survive even if the platform changes. Technical metadata like refresh dates may not be important to keep, but metadata of ownership is definitely valuable as are descriptions from business perspective.
What important context is still missing or disconnected?
- Look for knowledge that still resides in project documentation, Teams conversations or people’s heads.
- Business area specific concept modelling is an efficient way to clarify important entities and their relationships; and concept models work as input for ontologies consumed by AI.
- Connecting the unstructured sources to your AI tools can give the reported figures an enriched context, which you can efficiently utilize in your business processes. The question is, how to do this securely.
What needs to be available to analytics, data products and AI agents in the future?
- Use those needs to define what the next architecture must support.
- Do you already know how AI will change your business? Our customers do have an idea.
A platform transition should not be an exercise in recreating the current catalog. It is a good opportunity to decide what is worth preserving, what should be improved and what is not worth rebuilding at all. Once we have that picture, we can evaluate technologies against actual requirements instead of starting with feature lists and trying to make the organisation fit the next platform.
Begin by deciding which organisational knowledge is to be stored and developed further. Then kick off the work with the OneLake catalog in Microsoft Fabric and Fabric Ontology capabilities.
