Introducing LakeFusion 6.0: 17 Capabilities for Smarter, Governed Master Data

Master Data Management is moving beyond creating clean records for analytics.

Enterprise teams now need trusted master data that supports applications and AI agents, governs products alongside other enterprise entities, maintains complex relationships without creating another layer of duplicated data, and prevents bad records from entering the master in the first place.

LakeFusion 6.0 is built around that shift.

It is LakeFusion's largest release to date, bringing 17 capabilities across Master Data Management, Product Information Management, LakeGraph, and Platform.

The release is built around three ideas:

Smarter. Deeper. Open.

Smarter matching and onboarding. Deeper governance across relationships, nested values, and product information. And more open, governed access that allows trusted master data to move from something teams simply consume to something applications and AI agents can safely act on.

What's New in LakeFusion 6.0?

LakeFusion 6.0 expands how enterprise teams match, govern, connect, enrich, and operationalize trusted data, extending the broader enterprise data management platform across MDM, PIM, LakeGraph, and platform capabilities.

Together, these additions support a broader goal: trusted master data your teams and your AI agents can act on.

Smarter Matching and Master Data Onboarding

Enterprise MDM has to balance matching accuracy with speed, explainability, and operational cost. LakeFusion 6.0 strengthens that foundation by introducing a learned probabilistic layer, relationship-driven hierarchy management, AI-assisted mapping, and real-time matching before new records are created.

Probabilistic Matching Adds a Learned Layer Before the LLM

Sending every ambiguous record pair to an LLM can increase latency, cost, and non-deterministic behavior.

LakeFusion 6.0 introduces a Random Forest probabilistic matching layer trained on an organization's own verified Ground Truth decisions, extending the intelligent matching and entity resolution capabilities within LakeFusion MDM.

The tiered matching pipeline becomes:

Deterministic Rules → Random Forest → LLM

Deterministic matching handles straightforward cases first, clearing approximately 80% of clean records.

The Random Forest then scores the remaining candidate pairs between 0 and 1 and routes them into three bands:

• AUTO_REJECT — the records should not be matched
• AUTO_MATCH — the records can be automatically merged
• POTENTIAL — the pair is uncertain and requires LLM evaluation

Only the POTENTIAL band reaches the LLM.

This can reduce LLM calls by approximately 50% while preserving AI-based reasoning for cases where it is actually needed.

SHAP feature contributions also provide visibility into why a pair received its score, making the probabilistic layer more reproducible and explainable.

For example, two patient records may have slightly different names and partially available dates of birth while strongly matching on address. Instead of sending the pair directly to an LLM, the Random Forest can evaluate those signals first and escalate only when the evidence remains uncertain.

Hierarchy Management Uses Relationships as the Source of Truth

A golden record establishes an entity's trusted identity. Enterprise teams also need to understand where that entity belongs within a broader organizational or business structure. LakeFusion 6.0 extends the relationship capabilities introduced with LakeFusion 5.0 by composing hierarchies directly from existing relationships.

Each hierarchy level can be composed from a relationship and an entity type, allowing teams to manage:

• Ragged hierarchies
• Mixed entity types
• Parent-child relationships
• Multiple hierarchy placements
• Full breadcrumb paths
• Relationship provenance

The hierarchy is computed live from the underlying relationship edge tables.

No additional hierarchy tables to materialize, no separate synchronization jobs, and no duplicate hierarchy structure to maintain.

The edges remain the single source of truth, helping ensure the hierarchy cannot drift away from the relationships used to create it.

If an underlying relationship changes in Databricks, the hierarchy reflects that change on refresh without requiring another hierarchy-specific synchronization process.

AI Column Mapping Reduces Manual Mapping Work

Bringing a new enterprise dataset into MDM often begins with manually mapping source columns to the correct entity attributes.

LakeFusion 6.0 introduces AI Column Mapping through a suggested mapping workflow.

The LLM evaluates:

• Column names
• Data types
• The top 10 sample rows
• Target attribute names
• Attribute labels
• Target attribute types

It then proposes the most appropriate source-to-target mappings for review.

Teams remain in control of the result. Teams can override any suggestion, incompatible data types are discarded, and attributes without a sufficiently confident match remain unmapped.

The capability works with master and reference entities, while manual mapping remains available.

This turns the onboarding of wide datasets from a series of repetitive dropdown selections into a faster review-and-approve workflow.

Search Before Create Prevents Duplicates Earlier

Traditional deduplication often begins after a new record has already been created.

Search Before Create moves entity resolution ahead of that write.

An application can send a partial incoming record through a real-time API. LakeFusion applies the same deterministic and probabilistic matching logic used in the broader matching process and returns ranked golden-record candidates with scores.

The response can route the application into three paths:

use_existing — a high-confidence match already exists

route_to_steward — the result is ambiguous and needs human review

create_new — no suitable match was found

Partial payloads are supported.

For example, a returning patient can be checked against existing golden records before creating a new patient identity. If a high-confidence match already exists, the new interaction can link to that trusted record instead of creating another duplicate.

Deeper Governance Across Enterprise Data

LakeFusion 6.0 also expands what can be governed inside the master data model. Governance now reaches dependent reference values, nested structs and arrays, more sophisticated access controls, and the product information that often sits alongside traditional customer, supplier, and organizational master data.

Dependent References Prevent Invalid Combinations

Independent reference lists can allow individually valid values to be combined in incorrect ways.

For example, selecting one payer while assigning a plan belonging to another payer creates an invalid relationship even though both values exist independently.

LakeFusion 6.0 introduces Dependent References, where available child values are constrained by the selected parent.

The relationship can be enforced during:

• Data entry
• Data loading
• Validation

If Aetna is selected as the payer, only the plans associated with Aetna can be presented as valid options.

Incorrect combinations can be flagged during load and routed to stewardship rather than being silently mastered.

The same approach can be used for structures such as:

Country → State

Specialty → Sub-specialty

This prevents bad reference combinations from entering the trusted master in the first place.

Governed References Now Extend into Structs and Arrays

Enterprise data frequently contains important values inside nested structures rather than simple top-level columns.

LakeFusion 6.0 extends governed references into structs, arrays, and other complex data types.

A nested attribute can reference a governed entity by link instead of copying the reference value into every record.

For example, a plan code inside a patient's coverage array can resolve directly to a governed plan reference entity.

Nested values can then be matched, validated, and standardized in the same way as top-level attributes.

If a governed value changes, it can be updated once at the reference level instead of modifying every individual occurrence across nested records.

This extends consistent reference governance into parts of enterprise data that previously remained much harder to control.

Role-Based Access Gets More Flexible

LakeFusion 6.0 also introduces an in-house Role-Based Access Control model that no longer depends on Databricks workspace-admin privileges.

LakeFusion becomes the source of truth for roles and permissions.

Built-in roles include:

• Admin
• Developer
• Data Steward

Organizations can also create custom roles.

Permissions can be assigned at multiple scopes, including:

• Global
• Entity
• Object group
• Object

Object groups can include areas such as Integration, Match Maven, Relationships, and Model Serving.

Permissions are enforced at both the application and Databricks layers. Application permissions are checked on API calls, while per-user Databricks grants are applied through objects managed by the LakeFusion service principal.

This provides more granular control without requiring customers to grant LakeFusion Databricks workspace-admin rights.

Product Information Management Becomes a Full LakeFusion Vertical

Product information creates a different set of governance challenges from traditional entity mastering. Enterprises may need to reconcile product feeds, maintain taxonomies, manage attributes and specifications, support localization, and publish consistent information across several channels.

LakeFusion 6.0 expands Product Information Management into a full LakeFusion vertical, bringing product mastering, taxonomy management, enrichment, localization, and publishing into the same broader governance foundation as other enterprise data.

Manage Product Data from Import to Publication

Product information can be imported from Delta tables and flat files, mapped through N-tier field mapping, and loaded into Lakebase.

Teams can then organize and author products through:

• Multi-level taxonomy trees
• Attribute libraries
• Specification libraries
• Tiered product entities

A working catalog supports richer product operations such as:

• Media assets
• SEO information
• Localization
• Channel-specific data

Products can then be published into a live catalog.

This creates a governed workflow from product onboarding and mastering through enrichment and downstream publication, while keeping product information close to the same Databricks environment used by the broader LakeFusion platform.

PIM Import Mapping Simplifies Catalog Onboarding

Every vendor or supplier catalog can arrive with a different schema.

One source may use prod_ref for a product identifier, another may use item_code, while a third may call the same field catalog_id.

The PIM Import Mapping Tool provides a visual interface that reads the incoming schema and suggests mappings against the PIM taxonomy.

Sources can include CSV files, Delta tables, and API feeds.

AI can propose mappings such as:

• prod_ref → SKU
• item_desc → Description
• cert_class → Regulatory.Classification

Teams review, adjust where necessary, and save the approved mappings.

Once a mapping has been saved, LakeFusion can automatically reapply it when future imports arrive from the same source.

This reduces repeated engineering work and creates a more reusable catalog-onboarding process.

Multiple Taxonomies Stay Connected Through Crosswalks

The same product may need to be classified across multiple systems.

An enterprise may maintain an internal taxonomy while also supporting GS1, regulatory, channel-specific, or partner classifications.

Managing each taxonomy independently can create drift.

LakeFusion 6.0 introduces Multiple Taxonomies + Crosswalk.

Teams can define a primary taxonomy as the source of truth and create crosswalk rules linking equivalent nodes across additional classification systems.

For example, an internal product category can be connected to the corresponding GS1 and regulatory classifications.

A classification change in the primary structure can then propagate through the mapped taxonomy relationships.

This reduces manual reconciliation and keeps multiple product views aligned.

PIM also integrates with LakeFusion RBAC, allowing product catalogs and workflows to follow the same governed access model as the wider platform.

Making Trusted Master Data Agent-Ready

AI agents introduce a new requirement for enterprise data: access alone is not enough. They need trusted identities, controlled access, provenance, and relationships that remain governed as they move into operational workflows. That foundation starts with trusted Master Data Management on Databricks.

Agents need to know whether the information they receive is trusted, where it came from, and what actions they are allowed to perform.

LakeFusion 6.0 adds capabilities that bring governed master data closer to agentic workflows without removing the controls that make the data reliable.

MCP Gives AI Agents Governed Access to Master Data

LakeFusion 6.0 introduces an MCP server that exposes governed LakeFusion capabilities to AI agents.

Agents can work with:

• Golden records
• Entity resolution
• Match and deduplication
• Graph data
• Hierarchies

Each response can also carry a trust envelope containing information such as the value, confidence, and provenance.

For example:

• Value: Metformin 500mg
• Confidence: 0.95
• Provenance: Epic

The goal is not simply to expose another API to an AI system.

It is to allow agents to interact with governed enterprise information while receiving the trust context required to understand what they are acting on.

This brings agent workflows closer to the same trusted master-data foundation already used by enterprise teams and applications.

Documentation Hub Makes the Platform Easier to Navigate

LakeFusion 6.0 also introduces a searchable, versioned Documentation Hub.

The hub brings together guidance covering:

• Product setup
• Matching
• Reference Data Management
• Structs
• REST and SDK APIs
• MCP tools and schemas
• Connectors
• Databricks
• Snowflake and additional sources

Instead of relying on disconnected documentation or internal tribal knowledge, teams can work from one searchable source aligned with the LakeFusion version they are using.

License Server Centralizes Entitlements

A new License Server provides centralized control over licensing and feature access across LakeFusion instances.

The LakeFusion control plane can issue, renew, and revoke licenses while managing feature entitlements and usage limits.

Each instance validates its entitlements before access is granted.

Licensing includes approximately 24-hour cached validation, a seven-day expiry grace period, a 72-hour offline grace period, and encrypted, tamper-proof entitlements that cannot be modified on the customer instance.

External Environments also forms part of the LakeFusion 6.0 Platform capability set, further extending the platform foundation supporting enterprise deployments.

LakeGraph Moves from Relationship Insight to Operational Action

Relationships provide critical context around enterprise entities, but that context loses value when graphs become stale or require manual intervention before anything can happen.

LakeFusion 6.0 makes LakeGraph increasingly operational, connecting scheduled graph processing, MDM stewardship, AI-agent access, alerts, and downstream actions.

Scheduled Pipelines Keep LakeGraph Current

Graph pipelines can now run automatically according to a defined schedule.

This creates an operational loop where LakeGraph stays current without manual refreshes.

Graph traversal can also generate potential entity-match suggestions and route them back into MDM stewardship.

This creates a closed loop between Graph Intelligence and Master Data Management.

Instead of relationships remaining a separate analytical layer, graph context can actively improve matching and stewardship.

MDM and Graph Work Through a Unified Agent Connection

The LakeFusion MCP experience also brings MDM and Graph capabilities together for AI agents.

A unified MCP connection can expose 65+ MDM and graph tools through one agent interface.

That matters because enterprise questions rarely depend on a single entity.

An agent may need the trusted supplier golden record and the ownership structure around it. Or it may need a customer identity together with its hierarchy and related organizations.

Providing both through one governed connection gives the agent access to trusted entities and the relationships that give those entities business context.

Alerts and Actions Turn Relationship Changes into Workflows

LakeFusion 6.0 also introduces Alerts and Actions around graph conditions.

The workflow follows a simple pattern:

Condition → Alert → Action

Teams can define conditions against graph properties or relationships. When the condition becomes true, LakeFusion can trigger a governed response.

Actions can include:

• Updating a golden record
• Triggering a pipeline
• Calling an external system through a webhook

For example, if a healthcare graph identifies a member with no active coverage relationship, LakeFusion can flag the condition, update the golden record status, and initiate a re-verification workflow.

The same model can extend to other enterprise relationships where changes in graph state should result in action rather than waiting for someone to manually query the graph.

Managing Data Pipelines Without Leaving LakeFusion

LakeFusion 6.0 also brings more of the operational MDM experience directly into the application.

The In-App Pipeline Editor provides a familiar Databricks Jobs-style canvas where teams can view and manage workflows such as:

Ingest → Dedup → Match → Validate → Survivorship → Publish

Teams can insert custom tasks, configure external steps, monitor runs, review logs, and manage retries without moving between LakeFusion and the Databricks Jobs UI.

This makes it easier to incorporate organization-specific validations and downstream processes into the broader MDM workflow while managing the pipeline from one place.

What LakeFusion 6.0 Adds Up To

LakeFusion 6.0 introduces a broad set of capabilities, but together they support three clear changes in how enterprise master data can be managed and used.

Smarter

Probabilistic Matching adds a Random Forest layer trained on verified decisions, reducing unnecessary LLM calls. Relationship-composed hierarchies eliminate additional materialized hierarchy structures, while AI Column Mapping accelerates dataset onboarding.

Deeper

Dependent References and governed references inside structs and arrays extend governance into more complex enterprise data. Product MDM and PIM add catalog authoring, enrichment, taxonomy management, crosswalks, localization, and channel publishing.

Open

MCP exposes governed master data and graph capabilities to AI agents with confidence and provenance. Search Before Create brings trusted matching into real-time application workflows, while RBAC, documentation, licensing, and operational LakeGraph capabilities make trusted enterprise data easier to access and act on.

Together, these changes support the core promise of LakeFusion 6.0: trusted master data your teams and your agents can act on.

From Trusted Master Data to Agent-Ready Enterprise Data

Creating a golden record is still essential, but trusted enterprise data increasingly needs to do more.

It needs to sit within accurate hierarchies, remain governed even inside complex data structures, support product information across channels, prevent duplicates before they are created, respond to changing relationships, and provide AI agents with enough confidence and provenance to act safely.

LakeFusion 6.0 brings those capabilities together across Master Data Management, Product Information Management, LakeGraph, and Platform.

For enterprises moving toward agentic AI, the goal is no longer simply giving AI access to more data. It is giving teams, applications, and agents access to trusted, governed enterprise data they can confidently act on.

LakeFusion has already applied governed master-data approaches to enterprise use cases such as Customer 360 on Databricks, showing how trusted identities can become a foundation for broader analytics and operational workflows.

Ready to See LakeFusion 6.0 in Action?

See how LakeFusion 6.0 brings smarter matching, deeper MDM and PIM governance, operational LakeGraph, and governed AI-agent access together on Databricks.

Request a LakeFusion Demo

Frequently Asked Questions

What is LakeFusion 6.0?

LakeFusion 6.0 is LakeFusion's largest release to date, bringing 17 capabilities across Master Data Management, Product Information Management, LakeGraph, and Platform.

What is new in LakeFusion 6.0 Probabilistic Matching?

LakeFusion 6.0 introduces a Random Forest layer trained on verified Ground Truth decisions between deterministic matching and LLM evaluation, allowing only genuinely uncertain record pairs to reach the LLM.

How does LakeFusion 6.0 support AI agents?

LakeFusion MCP gives AI agents governed access to golden records, entity resolution, hierarchies, and graph capabilities. Responses can include confidence and provenance information, providing a trust envelope around enterprise data.

What Product Information Management capabilities are included in LakeFusion 6.0?

LakeFusion 6.0 supports product importing, taxonomy-based authoring, enrichment, localization, channel publishing, AI-assisted import mapping, multiple taxonomies with crosswalks, and RBAC integration.

What changed in LakeGraph with LakeFusion 6.0?

LakeGraph adds scheduled pipelines, graph-driven match suggestions that feed into MDM stewardship, unified MCP access, and Alerts and Actions that can trigger governed operational workflows.