AI, yes – but who's liable for the nonsense?

Imagine this: your CFO presents the board with a data-driven forecast generated by an AI model. Two weeks later, it turns out the numbers were based on outdated, unvalidated data. The question that then hangs in the room isn't just uncomfortable – it's existential: who is actually liable for decisions like this?

This question is currently occupying many companies that are integrating AI and analytics into their decision-making processes. Because while enthusiasm for generative AI and data-driven innovation is high, responsibility for data quality, access control, and traceability often remains unresolved. This is exactly where Databricks Unity Catalog comes in – as a central governance framework that not only minimizes risk but also enables innovation.

In this article, we'll walk you through a structured master path to data governance with Databricks Unity Catalog. You'll learn why Unity Catalog is particularly relevant for SAP analytics teams, which four phases you should go through, and how to achieve quick wins – without falling into typical governance traps.

---

Why Unity Catalog matters for SAP analytics teams

Many companies in the SAP ecosystem face a paradoxical situation: on one hand, business units demand faster access to data and self-service analytics. On the other, regulatory requirements, compliance mandates, and the introduction of AI applications are putting increasing pressure on central IT and data owners.

The consequence? Governance is often perceived as a brake – not an enabler. Projects stall because permissions are unclear. Data silos persist because no one takes responsibility for cross-cutting metadata. And AI initiatives fail because data lineage can't be traced.

The challenge of hybrid data landscapes

Organizations close to SAP know this problem particularly well: their data landscape has grown over decades. SAP BW, S/4HANA, SAP Datasphere, external cloud platforms, data lakes, and legacy systems all exist side by side. Each component has its own security concepts, metadata structures, and access logic.

If you now introduce a modern analytics platform like Databricks to unify SAP and non-SAP data, you face the question: how do you establish governance across all these systems – without slowing down innovation?

Unity Catalog as a central control instance

Databricks Unity Catalog offers a pragmatic approach here: it acts as a central governance layer for your entire data platform. Much like SAP systems have a central control instance for permissions, object management, and metadata, Unity Catalog takes on this role for your hybrid analytics environment.

Specifically, this means:

• Central metastore: All metadata – whether from SAP, cloud DWH, or data lake – is managed in one place.
• Unified access control: You no longer define permissions per system, but centrally via catalogs, schemas, and tables.
• Data lineage: You can trace at any time where data comes from, how it was transformed, and who accessed it.
• Auditability: Compliance requirements can be met through automated logging and traceable data flows.

For SAP analytics teams, this means: you can organize management reporting, decision support, and data provisioning across system boundaries – while still retaining control.

---

The 4 phases: setup → federation → ABAC → AI governance

Governance isn't a one-time project – it's a maturation process. We recommend structuring this process into four sequential phases. Each phase has clear goals, typical pitfalls, and concrete quick wins.

---

Phase 1: Setup – Laying the foundation

Goal:

Build a functional, central governance structure with Unity Catalog that serves as the foundation for all subsequent steps.

Typical pitfalls:

• Perfectionism over pragmatism: Many teams want to define all governance rules from the outset. This leads to paralysis.
• Lack of role clarity: It's unclear who is responsible for metadata, permissions, and catalog structures.
• Technical hurdles: Integration with existing SAP and cloud systems is underestimated.

Quick wins:

• Start with a pilot catalog for a clearly scoped use case (e.g., financial reporting).
• Define three to five central roles (e.g., data owner, data steward, analytics user) and assign them.
• Use predefined schemas and naming conventions to create structure without reinventing everything.
• Set up lineage and audit logs from the start – even if you initially only use them passively.

Business benefits:

• Faster access to reliable data for business units
• Reduced risk through clear accountability
• Foundation for scalable analytics initiatives

Governance benefits:

• Central metadata management instead of system chaos
• Traceable permission structures
• Basis for compliance and auditability

---

Phase 2: Federation – Onboarding data sources and domains

Goal:

Federate various data sources (SAP, non-SAP, cloud, on-premise) and business domains (finance, sales, supply chain) into Unity Catalog without losing central governance.

Typical pitfalls:

• Fear of losing control: Central IT worries that decentralized teams will bypass governance rules.
• Inconsistent metadata: Every data source brings its own structures and terminology.
• Lack of alignment: Business units and IT don't speak the same language.

Quick wins:

• Establish data products as the central unit: every data product has a clear owner, documentation, and defined access rights.
• Use Delta Sharing to integrate SAP data and external sources without costly ETL processes.
• Define central governance rules that apply across all domains (e.g., naming conventions, data classification), while leaving room for domain-specific adaptations.
• Foster cross-functional teams made up of IT, business units, and data stewards who jointly develop data products.

Business benefits:

• Business units get faster access to relevant data
• Data reusability increases – less redundancy, more efficiency
• Innovation speed increases because teams can work independently

Governance benefits:

• Federation doesn't mean anarchy – central rules remain in place
• Transparency across all data sources and their origin
• Clear accountability per data product

Why federation is not a loss of control:

Many SAP teams associate federation with uncontrolled decentralization. But Unity Catalog enables controlled autonomy: business units can independently develop and use data products – but always within the central governance framework. Think of SAP authorization roles: there, too, central guidelines exist alongside business flexibility in their design.

---

Phase 3: ABAC – Introducing attribute-based access control

Goal:

Shift access control from static roles to dynamic, attribute-based rules (ABAC = Attribute-Based Access Control) to make governance more precise and scalable.

What is ABAC?

Traditional access models work with fixed roles: "User A is a controller and may access financial data." ABAC goes a step further and defines access based on attributes – such as cost center, region, project code, or confidentiality level.

Example:

A sales manager in Switzerland may only access sales data from Switzerland – not from Germany. With ABAC, you define this rule once centrally, instead of manually maintaining permissions for every user.

Typical pitfalls:

• Underestimated complexity: ABAC requires clean data modeling and clear attribute definitions.
• Poor data quality: If attributes (e.g., cost center) are inconsistent in the source data, ABAC won't work.
• Lack of alignment: Business units and IT must jointly define which attributes are business-relevant.

Quick wins:

• Start with one use case that has clear attributes (e.g., regional sales data).
• Use Unity Catalog to implement row-level security and column-level security based on attributes.
• Automate the assignment of permissions via attributes from your identity provider (e.g., Active Directory, Okta).
• Document your ABAC rules transparently so everyone involved understands why certain access is granted or denied.

Business benefits:

• Higher data security combined with more flexible access
• Scalability: new users automatically receive the right permissions
• Less manual effort for permission management

Governance benefits:

• More precise control over sensitive data
• Traceability: every access is documented on an attribute basis
• Compliance requirements (e.g., GDPR, SOX) are easier to meet

Prerequisites for ABAC:

• Clean master data and consistent attributes in source systems
• Clear definition of which attributes are business-relevant
• Technical integration between Unity Catalog and identity management

---

Phase 4: AI governance – Building trust in AI applications

Goal:

Extend governance so that AI and GenAI applications are built on a reliable, traceable, and legally sound data foundation.

Why AI governance?

With the introduction of AI models – especially generative AI – governance requirements change fundamentally. It's no longer enough to know who may access which data. You also need to ensure:

• Data provenance: Which data was used to train the model?
• Usage boundaries: Is this data even allowed to be used for AI applications (e.g., GDPR, licensing questions)?
• Traceability: How did the model arrive at this result?
• Accountability: Who is liable if the model produces incorrect recommendations?

Typical pitfalls:

• Unclear data usage rights: Many companies don't know whether their data may be used for AI training.
• Missing lineage for AI models: It's unclear which data flowed into which model.
• Lack of transparency: Business units don't understand how AI decisions are made.

Quick wins:

• Use Unity Catalog to link AI models and their data sources – creating lineage for AI as well.
• Define data classifications (e.g., "AI-ready," "AI-restricted") to clearly mark which data may be used for AI applications.
• Implement model governance: document who trained a model, which data was used, and which approvals are in place.
• Create transparency for business units: explain in plain language how the model works and what data it uses.

Business benefits:

• Trust in AI applications increases – internally and externally
• Faster scaling of AI use cases because governance questions are resolved
• Reduced liability risk for AI-supported decisions

Governance benefits:

• Traceability across the entire AI lifecycle
• Compliance with AI regulation (e.g., EU AI Act)
• Clear accountability for AI models and their data foundations

Why Unity Catalog helps here:

Unity Catalog provides governance not just for data, but also for AI assets. You can centrally manage, version, and assign permissions to models, feature stores, and training data. This creates the foundation for trustworthy, scalable AI applications.

---

Quick-win checklist: How to get started today

You don't need to tackle all four phases at once. Here's a practical checklist you can start with immediately:

Weeks 1–2: Assessment

• Identify one concrete use case (e.g., financial reporting, sales analysis).
• Clarify which data sources are relevant (SAP, non-SAP, cloud, on-premise).
• Define three central roles (data owner, data steward, analytics user).

Weeks 3–4: Set up a pilot catalog

• Set up Unity Catalog and create a pilot catalog for your use case.
• Define naming conventions and a simple structure (catalog → schema → table).
• Enable lineage and audit logs.

Weeks 5–8: Federate the first data products

• Onboard two to three data sources (e.g., SAP BW, cloud DWH).
• Create a first data product with a clear owner and documentation.
• Test Delta Sharing for integrating external data.

Weeks 9–12: Launch an ABAC pilot

• Choose a scenario with clear attributes (e.g., regional access rights).
• Implement row-level security based on attributes.
• Document the ABAC rules transparently.

From week 13: Prepare AI governance

• Classify data by AI usability ("AI-ready," "AI-restricted").
• Link a first AI model to its data sources in Unity Catalog.
• Establish model governance for new AI projects.

---

Want to assess your governance maturity in a structured way?

Download our Decision Brief and find out where your company stands today – and which steps to take next.

Or start directly with a Clarity Audit:

In a structured workshop, we jointly analyze your data landscape, identify quick wins, and develop a roadmap for your governance transformation.

What has been your experience with data governance in hybrid data landscapes? What challenges are you currently facing? We look forward to your feedback and the exchange of ideas.

Want to assess your governance maturity in a structured way?

Download our Decision Brief and find out where your company stands today – and which steps to take next.

Or start directly with a Clarity Audit:

In a structured workshop, we jointly analyze your data landscape, identify quick wins, and develop a roadmap for your governance transformation.