Data Governance Without Handbrakes: How AI Accelerates Time-to-Value in Data Mesh

A Data Mesh promises decentralized and flexible data use. In reality, many initiatives fail because data governance is too complex and time-consuming. Policies, quality standards, and access permissions often have to be checked manually—a process prone to error and delays.

Our Solution

INNOQ developed AI-powered features for Data Mesh Manager to automate data governance. These tools check whether data products and contracts comply with internal policies, flag issues, and suggest improvements. Business teams can define data quality rules in plain language, which are then automatically converted into executable SQL queries.

The Outcome

The AI-powered solution simplifies and accelerates federated data governance. Companies can organize their data architecture in a decentralized way—without sacrificing control over quality, access, or compliance. This unlocks data-driven innovation and speeds up time-to-value.

Data Mesh Manager

Data Mesh Manager removes key obstacles to becoming a data-driven organization. As a central platform for managing data products and contracts, it allows companies to decentralize their data landscape without losing grip on governance and compliance. It establishes an enterprise-wide data marketplace where business units can publish data products to other teams—while meeting all regulatory requirements.

The tool was born out of INNOQ’s employee innovation program, which supports self-initiated ideas with time, budget, and mentorship.

Two smiling people, Jochen Christ and Simon Harrer, in INNOQ t-shirts (apricot and white) against a beige background — Jochen Christ and Simon Harrer, founders of Data Mesh Manager

The solution helps companies build decentralized data architectures where data is treated as a product—with clear ownership, defined quality criteria, and transparent terms of use. Democratizing data access in this way fuels innovation and creates competitive advantages.

By integrating AI through Large Language Models (LLMs), the tool automates complex governance tasks. This case study shows how these AI features make governance easier and faster.

The Challenge: Governance Complexity That’s Hard to Automate

Implementing a Data Mesh means defining and enforcing numerous governance rules. Data products must be described, classified, and documented correctly. Data contracts need to meet quality criteria, and access to sensitive data must be controlled. Doing all of this manually is tedious, error-prone, and simply doesn’t scale.

The Approach: AI Features That Automate the Pain Points

To address this, we extended Data Mesh Manager with several AI-powered features based on Large Language Models (LLMs) and the Spring AI Framework. These features work with models like OpenAI’s GPT-4o—but model selection is fully configurable. For sensitive use cases, organizations can use self-hosted or on-prem LLMs.

Data Governance AI

This feature automates compliance checks for data products and contracts based on your organization’s policies.

Screenshot of a data governance user interface with a list of policies such as 'Ownership' and 'Data Classification.' Some policies are marked as 'Accepted,' others as 'Draft.' Notes like 'Issues with 4 data contracts' indicate problems. Buttons like 'Add Policy' are visible. — The Policies overview shows how Data Governance AI automatically checks for compliance. For certain rules—like "Data Classification," "Snowflake Naming Conventions," and "PII"—violations were flagged in up to four contracts. The AI gives early signals of compliance risk.

The AI receives policies in Markdown and metadata in YAML. Using Spring AI and Function Calling, it supplements this with additional info (e.g., the schema) before analyzing everything with the LLM. The LLM checks compliance and returns structured JSON containing any violations it finds.

This video shows how Data Governance AI checks all defined policies against existing data contracts. Deviations are highlighted and quantified, enabling a quick overview of potential compliance issues.

It also provides actionable suggestions to fix them:

Screenshot of the 'Data Governance AI' user interface with a PII check. An issue is highlighted: The field 'contact.email' is a potential PII candidate but has no 'PII flag'. Recommendation: Add the 'PII flag' to the 'contact.email' field and set it to true.

Users can flag false positives and save their feedback for future checks.

Another use case: evaluating access requests. Users can request access to data; the data product owner must approve or reject. The AI reviews the request for purpose clarity and potential policy conflicts. Its recommendation is shown directly in the decision UI.

Screenshot of Data Mesh Manager user interface with an access request for the data product 'Orders'. A yellow warning message indicates possible policy violations: 'Unclear purpose' and 'Processing of PII'. Options to 'Approve' or 'Reject' are available. — While reviewing a request, the AI identifies risks like unclear purpose or insufficient justification for accessing PII. The recommendation appears directly in the decision dialog for product owners.

Data Quality AI

This feature helps data engineers and product owners define quality checks. Users describe requirements in plain English (e.g., “The value must always be positive” or “Less than 5% nulls allowed”). The AI turns that into executable SQL and saves both query and threshold in the YAML-based data contract. Using the Data Contract CLI, the checks can be run continuously.

This natural-language approach is more efficient—and more powerful—than building a UI for every rule.

No one enjoys doing data governance—or manually checking if every rule was followed. These processes often take too long. With LLMs reviewing metadata in Data Mesh Manager, we can automate most of it. If something’s wrong, the product owner gets immediate feedback.

Jochen ChristCo-Founder Data Mesh Manager

User interface of a data quality tool with an SQL-based check for the 'ORDER_TOTAL' field, verifying that at most 5% of values are null. — Data Quality AI translates requirements like "May contain at most 5% null values" into SQL checks. These are stored in the data contract and can be run for continuous validation.

Technical Implementation Details

The AI features are built in Java using Spring AI, which wraps various LLMs. It uses Function Calling and structured JSON to allow models to perform tool use and return structured, machine-readable outputs.

Data Mesh Manager supports custom model configurations:

Dialog box in a data management tool titled 'Quality attributes for field ORDER_TOTAL'. Shows options for defining attributes with text or SQL, an SQL example, and a description explaining the verification of null values using a 5% threshold. — In the AI settings of Data Mesh Manager, organizations can configure anything from managed models (e.g., Azure OpenAI) to custom deployments. This gives full control over infrastructure, data security, and model choice.

Flexible model options: Azure OpenAI (EU), Ollama, or fully on-premises
GDPR-ready: No training data sent when using managed models
Spring AI integration: Easy to plug into Java-based systems
Function Calling: LLMs can execute structured checks
Structured JSON output: Auditable, machine-readable, automatable
Bring-your-own-LLM: Supports in-house models

Case Study

Data Governance Without Handbrakes: How AI Accelerates Time-to-Value in Data Mesh

TL;DR

The Challenge