You have Fabric and Purview. What is still missing before an agent can answer a business question?
Fabric holds the data. Purview scans it and labels the sensitive parts, automatically, but only the technical half and mostly the Microsoft-native half. The business half, what a term means, which rule constrains it, which field carries it, is left for somebody to curate by hand. Which is why, in most estates, it never quite gets done.
This page sets out what Fabric and Purview each cover, which half of the problem is automated and which is not, how the two compare with a governed context layer, and when the Microsoft stack on its own is proportionate.
Key takeaways
- Fabric stores and serves. Purview scans and labels. Both do their job, and neither is the piece that is missing.
- Purview's technical half is automated and its business half is hand-curated. That is why scan coverage climbs while glossary coverage stalls.
- Metagem automates the curated half: it extracts terms, rules and definitions and links each one to the fields that carry it, into an ontology it populates rather than one you fill.
- It reads structured and unstructured sources, inside and outside the Microsoft estate, and the result is held in open formats you keep.
What each one covers
Microsoft Fabric
Fabric is the platform: storage in OneLake, compute, pipelines, warehousing and the serving layer that BI sits on. It is where your structured data lands and where the queries run. As a place to consolidate an analytics estate it does the job it says it does, and nothing here proposes replacing any part of it.
Microsoft Purview
Purview catalogues the technical estate. It scans sources, records schemas, tracks lineage and applies sensitivity and compliance labels, and the scanning is genuinely automated. That credit is worth stating plainly, because the argument on this page is often made badly. Nobody is hand-tagging every asset in Purview. The scan runs, the inventory fills, the labels apply.
The part that is manual is the business layer sitting on top: the glossary, the definitions, the decision about which term a given asset represents, the link from a piece of meaning to a piece of data. That work is curated by people, and people who have a day job. Watch the two coverage numbers in any mature Purview deployment and they tell the story: scan coverage climbs steadily, glossary coverage flattens early and stays there.
The gap it leaves
Take active customer.
A Purview entry tells you the table exists, who owns the pipeline that fills it, when it last refreshed, and that it holds personal data. All accurate, all automatic, all useful.
It does not tell you that active excludes dormant accounts under a rule that lives in a finance policy nobody has scanned, that the rule has three exceptions for acquired entities, that two of your systems disagree about which flag carries it, or who to ask when they do. The scan reached the asset. The meaning was never in the asset to begin with.
This is the shape of the whole gap. The automated half of Purview covers the half of the problem that does not answer business questions, and the half that does is left to hand-curation with no automation behind it. An agent asking what active customer means gets an entry that describes a table.
- Scanning finds assets. It does not find meaning. A scan can tell you a column exists, its type, and where it flows. It cannot tell you which business rule governs it.
- The business glossary is the manual part. Terms, definitions and the links from meaning to asset are curated by people, which is why that number is the one that plateaus.
- Most governing rules were never in the warehouse. They are in a policy PDF, a finance manual, a slide made for an audit. Structured scanning was not built to read any of it.
- Fabric IQ starts where you already have a model. It gives you the tools to define and bind an ontology, which assumes a governed foundation and a team with time to hand-build it.
Side by side
| Fabric | Purview | Metagem | |
|---|---|---|---|
| What it holds | Data, compute, OneLake | Technical metadata, lineage, sensitivity labels | Terms, rules, processes, roles and the fields that carry them |
| Source coverage | OneLake and what you land in it | Structured assets, Microsoft-centric | Structured and unstructured, inside and outside Microsoft |
| How the technical half populates | Not applicable | Automated scan | Automated extraction |
| How the business half populates | Not applicable | Hand-curated | Extracted, then approved by domain owners |
| Business rules and exceptions | No | No | Yes, as first-class objects |
| Term-to-field linking | Not applicable | Manual | Automatic, with provenance |
| Sensitivity and compliance labelling | No | Yes | No |
| Stores and serves data | Yes | No | No |
| Inventories the whole technical estate | No | Yes | No |
The two rows worth reading together are how each half populates. Purview automates the technical half and hand-curates the business half. That is not a criticism of the product, it is a description of what it was scoped to do. It does mean that the coverage number people quote and the coverage number that decides whether an agent can answer a question are two different numbers.
When you do not need this
If your estate is Microsoft end to end, the data is already modelled in OneLake, the rules that govern your metrics live in the model rather than in documents, and you have a team with the time and the appetite to hand-build an ontology, Fabric and Purview with Fabric IQ on top is proportionate and it keeps you on one vendor. A separate context layer earns its place when the meaning is spread wider than one vendor can see, or when hand-curation is the step that has already failed once.
How they fit together
What Metagem supplies
The business half, automated. Terms, definitions, rules and exceptions extracted from documents and systems, each linked to the technical fields behind it, each traceable to its source and its owner.
What Metagem does not do
It does not store or serve data, scan and inventory the whole technical estate, or apply sensitivity and compliance labels. Those are Fabric's and Purview's jobs and it has no ambition to take them.
What you keep
Fabric as the platform, Purview as the technical catalog and the labelling authority, and the governance processes you have already built around both.
The division is one of scope rather than of quality. The Microsoft stack governs Microsoft-native data very well. A context layer reads structured and unstructured sources across the whole estate, including the documents that hold the rules and the systems Microsoft never sees, and it holds the result in open formats. The context stays yours, which is the difference between governing your meaning and renting it.
Purview automates the scan and leaves the meaning to be typed in. Metagem automates the meaning, extracted from documents and systems the Microsoft stack has never read, linked to the fields that carry it, and held in a form you own.
What's different hereThe mechanism behind this is described on active metadata and business ontology and context graph.
Frequently asked questions
Take one term and follow it down to the field.
We will show you the definition, the rule and its exceptions, the owner, and the columns it resolves to, alongside what your catalog holds for the same term today.
Talk to our team