Market shifts, listening to the Gartner Magic Quadrant, and looking at the road ahead
Every wave of enterprise technology eventually runs into the same wall.
Business intelligence needed reliable data. Robotic process automation needed reliable data. Now artificial intelligence, with all its promise, needs reliable data too. Yet most organisations are still struggling with the same fundamental problem.
That is the real story behind the renewed interest in Master Data Management (MDM). AI has not created a new data problem. It has made an old one impossible to ignore.
For years, organisations have invested in data platforms, integration technologies, analytics capabilities and increasingly sophisticated architectures. Yet the underlying information remains fragmented, inconsistent and difficult to trust.
AI changes the equation because it consumes information at a scale and speed that makes these weaknesses considerably more visible. It also creates an opportunity to address problems that organisations have struggled to resolve for decades.
I believe this is where the next generation of MDM will distinguish itself.
Trust is the bottleneck
Across the industry, organisations repeatedly identify a lack of trust in their data as a major obstacle to digital transformation. Whether the objective is reporting, automation, operational decision-making or AI, the underlying challenge is remarkably consistent.
But trusted data is not a single characteristic. It is the result of several interconnected dimensions:
- Accuracy: Does the information reflect reality?
- Completeness: Are the necessary attributes present?
- Consistency: Does the information mean the same thing across systems?
- Timeliness: Is the information sufficiently current for its intended purpose?
- Integrity: Do the relationships between records and business entities make sense?
- Lineage: Can we understand where the information originated and how it has changed?
- Accountability: Do we know who is responsible for the information and the decisions made about it?
The problem is that organisations frequently approach AI with a new use case rather than addressing these underlying dependencies.
They connect a model to existing enterprise data, demonstrate an impressive proof of concept, and then discover that the information cannot support reliable decisions.
AI has intensified this problem because it consumes data at greater scale and with considerably less human review than many previous technologies.
The uncomfortable truth is that an organisation cannot automate its way out of a lack of understanding, ownership and accountability.
A loop that feeds itself
One of the more interesting developments is the growing recognition that AI can become part of the solution rather than simply another consumer of enterprise data.
This creates a recursive relationship:
- Organisations need trusted data to make AI useful.
- AI can help identify, classify, analyse and improve the underlying data.
- Improved data makes subsequent AI applications more reliable.
- Better results generate further opportunities to improve the information foundation.
This is an important shift in perspective.
For decades, data management has been constrained by the amount of human effort required to understand and improve enterprise information. Data profiling, source analysis, attribute mapping, duplicate identification and validation design are all labour-intensive activities.
Large language models introduce a new level of scalability. They can perform much of the investigative work that previously required substantial specialist effort.
However, there is an important distinction: AI can accelerate the work, but it cannot independently establish business agreement or assume organisational accountability.
That distinction should remain central to how we design the next generation of MDM capabilities.
My take on the Gartner Magic Quadrant
The return of Gartner’s Magic Quadrant for Master Data Management, alongside renewed attention to the Data Quality market, is an interesting development.
The market is clearly moving. AI capabilities are becoming increasingly important in vendor strategies, and traditional MDM platforms are evolving towards more intelligent, automated and connected environments.
But I am less interested in where individual vendors land in a quadrant than in what this shift tells us about the future of the discipline.
For me, AI will accelerate the MDM journey across several layers, from the semantic foundations of enterprise data to the operational work of identifying and correcting poor-quality information.
I see five particularly important areas.
1. Ontologies and semantic foundations
Building an ontology or understanding how business entities relate to one another has traditionally been specialist work.
AI can help draft these relationships, analyse existing data structures, identify implicit connections and translate technical representations into language that business stakeholders can understand.
This is particularly interesting because an ontology should not exist purely as a technical artefact. It should represent how the business understands its world.
The business must still validate the meaning and relationships, but AI can substantially reduce the effort required to get started.
2. Source onboarding and attribute mapping
Anyone who has onboarded a new data source into an MDM platform knows how much effort is involved.
Understanding source structures, identifying corresponding attributes, analysing value distributions and mapping information into a canonical model can consume considerable time.
AI can help guide this process, suggest mappings, identify semantic similarities and explain inconsistencies.
Instead of starting with a blank mapping specification, practitioners can start with an informed proposal and concentrate on the exceptions.
3. Identifying data problems
AI agents can continuously investigate enterprise information for duplicates, missing attributes, inconsistent relationships and potential quality issues.
This is particularly valuable when dealing with large and fragmented data landscapes.
A traditional stewardship team simply cannot investigate every record, every relationship and every possible inconsistency manually.
AI changes the economics of that work.
4. Rules and validation
Historically, translating business requirements into data quality rules has often required technical specialists.
A business expert might understand perfectly well that a supplier must have a valid registration number, or that a particular product attribute is mandatory under certain circumstances.
Turning that understanding into executable validation logic is another matter.
AI can bridge this gap by translating natural-language requirements into proposed rules, validation conditions and test scenarios.
The subject matter expert can concentrate on whether the rule reflects the business requirement rather than how to implement it.
5. Operational remediation
Eventually, all data management initiatives encounter the same reality: broken information that has accumulated over many years.
Incorrect addresses, duplicate customers, inconsistent supplier hierarchies, incomplete product descriptions and obsolete classifications rarely originate from a single incident.
They are the accumulated consequences of operational processes, system limitations and insufficient accountability.
AI agents can investigate these problems, suggest remediation strategies and prepare changes in bulk.
People remain responsible for approving the proposed changes, while deterministic processes execute the actual updates.
This is where I see one of the most immediate opportunities.
Much of the MDM backlog is not particularly intellectually challenging. It is simply enormous. AI provides additional capacity to investigate and prepare the work that organisations have never had sufficient people to complete.
From data specialists to business participation
One of the most important consequences of AI in data management is the potential change in who can participate.
Historically, many data management activities have been restricted by technical complexity.
Business experts understand the meaning of information, but they are often dependent on developers and data specialists to translate that understanding into models, mappings and validation rules.
AI can reduce that dependency.
Consider a seemingly simple example: a status column containing the values A and I.
A language model might reasonably interpret these as Active and Inactive. In many contexts, that would be correct.
But what if A actually means Approved, or I means In Review? What if the interpretation changes depending on the source system or business process?
Only someone with the relevant business knowledge can establish the correct meaning.
The opportunity is to make that knowledge accessible to AI agents and incorporate it into the data management process.
This is not about replacing data stewards. It is about giving them better tools and allowing them to contribute at a scale that was previously impossible.
From answers to attribution
One of the most compelling developments is the way AI can interact with enterprise data.
Imagine asking an AI assistant about a customer, a supplier or a commercial opportunity.
Connected directly to Salesforce, Dynamics and HubSpot, a language model might identify several matching records across these systems.
It may even produce a convincing summary.
But which record represents the actual customer? Which information is current? Are there conflicting addresses? Who is responsible for the relationship? Can the information be trusted?
The language model can retrieve information, but retrieval alone does not establish authority.
Now consider the same question being answered through an MDM layer, exposed through the Model Context Protocol (MCP) to tools such as ChatGPT, Claude or Microsoft Copilot.
The response can include much more than a collection of matching records:
- The canonical record and its authoritative attributes.
- Relevant data quality scores and warnings.
- Source system references and data lineage.
- Relationships to other business entities.
- The history of merges and other changes.
- The identity of the people responsible for approving important decisions.
Integration with Microsoft Fabric IQ can further contribute a business-oriented semantic layer, helping AI understand how enterprise entities and concepts relate to one another.
This is where the distinction becomes important.
I do not simply want AI to provide an answer. I want to understand why that answer should be trusted.
An answer without attribution is just another piece of generated information. An answer with its history, ownership and quality context attached becomes something that can support an accountable business decision.
Agents as first responders
The next evolution of MDM will increasingly be agent-oriented.
Data landscapes are constantly changing. New sources appear, business processes evolve, regulatory requirements change and new use cases emerge.
Traditional approaches rely heavily on people periodically identifying and investigating these changes.
That model does not scale.
Instead, I see a future where agents act as first responders in the data management landscape. They continuously investigate, identify potential issues, prepare recommendations and bring decisions to the people who have the authority to make them.
Several principles are important here.
Agents are teammates, not autonomous data owners. They should be given defined objectives, tasks, performance indicators, budgets and access to relevant business documentation. Data dictionaries, glossaries, process descriptions and operational guidelines should provide the context needed to perform meaningful work.
Read-only access and suggestions first. Agents can identify hundreds of potential rules or thousands of possible duplicates. They can investigate and prepare remediation plans, but they should not independently change authoritative enterprise records.
Human decisions, at scale. The objective is not to force people to approve every individual record manually. AI should assemble the evidence, group related decisions and present them in a way that allows experts to approve or reject changes in batches.
A deterministic execution backbone. Language models are probabilistic. Their interpretations can be useful, but they are not guaranteed to be correct. The actual execution of approved changes should therefore remain within predictable, controlled processes.
An agent can propose a merge. A human can approve it. A deterministic pipeline performs the merge and records the result.
This separation allows AI to introduce flexibility and intelligence without sacrificing control.
Scale beyond traditional tooling. The real opportunity is not simply having an AI assistant generate a script for a developer. It is having agents capable of investigating millions of records, analysing patterns, preparing recommendations and operating within an established enterprise data management framework.
That is a fundamentally different proposition.
Meeting people where they work
Another important development is federation.
For too long, data governance has operated on the assumption that if organisations build a dedicated platform, business users will eventually adopt it.
In practice, this rarely happens as intended.
Business experts already have established ways of working. They use Microsoft Teams, Slack, email, Jira, ServiceNow and other operational applications.
They should not have to learn an entirely new MDM interface simply to approve a duplicate or validate an attribute.
Instead, decisions should be delivered to the places where people already work.
Imagine receiving a request to determine whether two company records represent the same legal entity.
Rather than navigating an MDM platform and searching for supporting evidence, you receive a decision package containing:
- Both candidate records.
- The AI’s assessment and confidence.
- The differences between the records.
- Their respective source systems.
- Existing ownership information.
- Relevant third-party enrichment.
- The potential consequences of merging them.
You review the evidence and approve or reject the proposal.
The complexity remains within the MDM environment. The decision is brought to the person who understands the business context.
This is what I mean by federation: distributing decisions and responsibilities without fragmenting the underlying governance and control.
The same thinking should apply to agent development.
Instead of requiring users to understand complex agent configuration, a copilot should guide them through defining objectives, selecting capabilities, analysing available data and preparing an execution plan.
The technology should make participation easier, not introduce another layer of complexity.
What this means for the MDM market
I believe the next generation of MDM platforms will be judged by a different set of expectations.
AI capabilities should no longer be considered an optional feature or an isolated assistant. They should become an integral part of how the platform investigates, understands and manages enterprise information.
For me, a modern MDM platform should demonstrate several characteristics.
AI as a core capability. The platform should support intelligent data discovery, semantic analysis, mapping, quality investigation, rule generation and agent-based stewardship.
Workflow and federation. Business decisions must reach the people who have the authority and knowledge to make them. The platform should integrate with existing operational tools rather than expecting everyone to work inside a central MDM application.
Cloud scale and SaaS delivery. The ability to process large data volumes, support distributed operations and evolve continuously is increasingly important. Modern platforms must be designed for cloud-scale operations.
Lower barriers to entry. MDM has traditionally been difficult to evaluate and expensive to implement. Organisations should be able to start with a focused business problem, demonstrate value and expand progressively.
Integrated data quality. I have long argued that MDM and data quality belong together. You cannot establish a credible canonical record without understanding its integrity, accuracy and completeness. Equally, data quality rules become significantly more meaningful when they operate against a properly understood and unified representation of the business entity.
Treating MDM and data quality as entirely separate disciplines creates unnecessary gaps in accountability and execution.
They are two sides of the same problem.
How AI entered data management
Looking at the development of AI in data management, I see three broad stages.
The first was the use of large language models to assist with analysing data and applying rules at scale. This began before ChatGPT became widely available and demonstrated that language models could contribute to data profiling, classification and quality analysis.
The second stage was the introduction of copilot-style experiences. These made it possible for non-technical subject matter experts to interact with data management capabilities through natural language, translating business requirements into more complex validation and operational tasks.
The third stage was the emergence of AI agents with defined capabilities, tools and skills.
By October 2024, the concept of agents equipped with specialised skills was becoming increasingly visible. Since then, the approach has developed rapidly across the market.
What connects these stages is a common objective: making business knowledge an active part of the data supply chain rather than something locked away in the heads of a few specialists.
This is arguably more important than the technology itself.
The architectural challenge we should not ignore
There is, however, a risk in the current enthusiasm around AI-driven data management.
Organisations may end up automating their existing complexity rather than addressing it.
An agent can identify duplicates, but it cannot independently decide which business process should be responsible for preventing them.
It can suggest a validation rule, but it cannot establish who has the authority to define the underlying business requirement.
It can recommend a canonical model, but it cannot resolve organisational disagreements about what a customer, supplier or product actually represents.
These are architectural and organisational questions.
AI can make the investigation considerably faster, but the underlying governance still needs to be established.
This is why I believe process-driven MDM becomes even more relevant in the age of AI.
Master data should not be treated simply as an IT asset that needs to be cleansed and maintained. It is the information created, changed and consumed by business processes.
If the process consistently creates incorrect information, no amount of downstream AI will permanently solve the problem.
The real opportunity is to combine AI-assisted investigation with clear business ownership, process accountability and controlled execution.
That is how organisations move beyond repeatedly correcting the same problems.
What I would tell data leaders
For organisations considering their next MDM investment, I would focus on five priorities.
- Treat data trust as a prerequisite for AI. Do not treat data quality and governance as activities that can be postponed until after an AI implementation. Address the information foundation alongside the use case.
- Combine MDM and data quality. They should operate as connected capabilities. Unified master records need quality controls, and quality controls need a meaningful understanding of the entities they evaluate.
- Use AI to scale stewardship, but maintain guardrails. Let agents perform the investigation, analyse patterns and prepare recommendations. Keep human approval for consequential decisions and use deterministic processes for execution.
- Bring decisions to where people already work. Federate the workflow. Make it easy for business experts to contribute without requiring them to become MDM specialists or learn another platform.
- Demand attribution. An AI-generated answer should be accompanied by its lineage, quality context, source information and approval history. If a vendor cannot explain where an answer came from, the conversation about trust has not even started.
The road ahead
I believe we are approaching an important transition in Master Data Management.
For many years, MDM has been associated with relatively complex implementations, centralised stewardship teams, elaborate governance structures and long-running data cleansing initiatives.
AI creates an opportunity to change the economics of this discipline.
It can reduce the effort required to understand enterprise information, make business participation easier, accelerate data onboarding and provide the capacity to address years of accumulated data debt.
But it does not eliminate the need for architecture, ownership or accountability.
In fact, as AI becomes more deeply embedded in business operations, these foundations become more important.
The organisations that succeed will not necessarily be those with the most sophisticated AI agents or the largest MDM platforms.
They will be the organisations that understand their business information, know who is accountable for it and have designed processes that consistently produce information they can trust.
AI can give us more hands. It can give us more speed. It can even help us understand information that has remained poorly documented for decades.
But it cannot replace the business decisions that make data meaningful.
That, to me, is the real future of Master Data Management.
Leave a Reply