Agentic AI Won’t Save You If Your Data Is Broken: The MDM Wake-Up Call

Agentic artificial intelligence has quickly become the new executive obsession. The promise is seductive: digital agents that can reason, plan, act, orchestrate workflows, and take initiative across customer service, operations, finance, supply chains, and compliance. McKinsey describes this shift as the move from reactive tools to “autonomous, goal-driven execution”, while the World Economic Forum warns that many organisations are rushing ahead while overlooking the quality, security, and governance of the data those systems rely on.

That is the real issue. Agentic artificial intelligence is not arriving in pristine operating environments. It is being layered onto fragmented enterprise resource planning systems, inconsistent customer records, duplicate suppliers, incomplete product hierarchies, weak metadata, contradictory reporting logic, and uncontrolled spreadsheets. In other words, it is being asked to make decisions on top of broken enterprise memory. NIST’s Artificial Intelligence Risk Management Framework is explicit that data used to build artificial intelligence systems may not be a true or appropriate representation of the context in which those systems are used, and that data quality issues can undermine trustworthiness.

This is why master data management is back at the centre of the conversation. Not as an old technology programme. Not as a back-office clean-up exercise. And certainly not as a purely information technology concern. It is becoming the discipline that determines whether artificial intelligence remains a pilot, becomes a liability, or scales into something commercially useful. DAMA positions data management as a globally recognised discipline grounded in best practice, while Microsoft’s Responsible Artificial Intelligence Standard states plainly that all artificial intelligence systems are subject to appropriate data governance and management practices.

1. The problem is not a lack of artificial intelligence ambition

Most leadership teams do not have an imagination problem. They have a readiness problem. IBM’s 2025 global study of chief data officers found that 81 per cent prioritise investments that accelerate artificial intelligence capabilities, yet only 26 per cent are confident their data can support new artificial intelligence-enabled revenue streams. In the Middle East and Africa, the pattern is similar: 77 per cent prioritise artificial intelligence investment, but only 25 per cent believe their data can support new revenue streams, and only 27 per cent are confident they can use unstructured data in a way that delivers business value.

That gap matters more in an agentic environment than it did in earlier analytics programmes. Traditional dashboards expose bad data slowly. Agents can operationalise it quickly. A flawed record is no longer just a reporting nuisance; it can become a trigger for action, escalation, recommendation, exception handling, or automated communication. When bad data is connected to autonomy, error moves from observation to execution.

2. Agentic systems amplify whatever foundations already exist

Artificial intelligence agents do not magically create truth. They inherit the logic, permissions, data access, and process weaknesses of the environment they are placed into. McKinsey notes that the real promise of agentic artificial intelligence lies in automating complex business workflows, but it also warns that agents introduce systemic risks such as uncontrolled autonomy, fragmented system access, lack of observability, and an expanding attack surface. In a telling line, the firm notes that what starts as intelligent automation can “quickly become operational chaos” if it is not built on a foundation of control, scalability, and trust.

That warning should land hard in any executive committee. An intelligent agent handling procurement, onboarding, forecasting, customer resolution, or financial reconciliation is only as good as the identifiers, hierarchies, statuses, and definitions it can trust. If the enterprise has five versions of a customer, three versions of a supplier, mismatched product codes, and inconsistent business rules across systems, the agent does not solve the confusion. It industrialises it.

3. Master data management is the control layer, not the clean-up crew

Too many organisations still frame master data management as a technology project whose purpose is to tidy records and reduce duplication. That view is now dangerously outdated. Modern master data management exists to create authoritative, governed, traceable records for the entities that matter most to the business: customers, suppliers, products, assets, sites, employees, contracts, and reference data. Informatica describes master data management as the integration of information about important business entities into a complete and consistent master record, while SAS argues that reliable access, integration, cleansing, governance, and preparation are essential to artificial intelligence and analytics.

Seen this way, master data management is not peripheral to agentic artificial intelligence. It is the operating control layer that allows agents to act on trusted entities rather than ambiguous approximations. It creates the identity discipline that prevents duplicate actions. It establishes lineage so decisions can be audited. It enforces stewardship so ownership is visible. It reduces semantic confusion so the same term does not mean three different things in three systems. Without that discipline, every agent becomes a confidence trick: impressive in demonstration, unreliable in production.

4. The wake-up call is bigger than structured data

One of the most important shifts in the current conversation is that the quality challenge is no longer limited to traditional master data domains. Harvard Business Review recently argued that proprietary unstructured content such as emails, contracts, forms, meeting recordings, and shared files can make generative artificial intelligence more distinctive, less likely to hallucinate, and more economically valuable. But the same article included a blunt warning from a chief data officer: “You’re unlikely to get much return on your investment by simply installing CoPilot.”

That point matters because many organisations assume structured master data and unstructured enterprise content sit in separate conversations. They do not anymore. Agents increasingly draw on both. A customer-facing agent may need the right customer record, the right price list, the latest contract language, the correct service history, and the appropriate policy interpretation at the same time. The more these worlds converge, the more master data management, metadata discipline, taxonomy design, and content governance have to work together. The companies that still manage them as separate islands will struggle to scale trustworthy artificial intelligence.

5. Good agents require more than access; they require alignment

McKinsey makes a crucial point that is often missed in the market noise around artificial intelligence agents. High-impact agents cannot simply be switched on from a software catalogue and expected to create strategic advantage. To generate real value, they need to be “deeply aligned with the company’s logic, data flows, and value creation levers”. That is another way of saying the hard work lies in process design, data architecture, governance, and control, not in the interface layer alone.

This is where many organisations are currently at risk of self-deception. They are treating the visible intelligence as the solution while ignoring the invisible disciplines that make intelligence usable. A smart assistant can draft a response, compare scenarios, or orchestrate a workflow. But it still needs the right legal entity, the right product mapping, the right organisational structure, the right approval path, and the right policy constraints. Master data management is what gives those concepts operational stability. Without it, agents remain articulate but ungrounded.

6. Broken data becomes a risk issue before it becomes a technology issue

The agentic conversation is often dominated by productivity and innovation language. It should also be dominated by risk language. MIT Sloan’s recent coverage of generative artificial intelligence risks highlights that risk enters through multiple components, including training data, foundation models, user prompts, and system prompts. It also notes that embedded risks are shaped by training-data quality and model behaviour, while enacted risks arise from how organisations configure and deploy systems.

In parallel, McKinsey’s security playbook for agentic artificial intelligence flags data corruption propagation as a specific danger: low-quality data can silently affect decisions across agents. That is exactly the point senior leaders need to absorb. In an interconnected agent environment, a poor-quality label, incorrect supplier classification, wrong customer hierarchy, or outdated policy rule does not stay contained. It travels. It contaminates downstream logic. It distorts reporting, action, and oversight at scale.

For highly regulated sectors, the implications are obvious. For boards, the lesson is even broader. The question is no longer whether artificial intelligence can automate a process. The question is whether the enterprise can defend the integrity of the data, rules, and permissions behind that automation. That is why the master data management discussion should now sit with risk, finance, operations, and executive leadership, not only with data teams.

7. South African organisations should treat this as a strategic discipline now

For South African enterprises, the timing is especially important. Many large organisations are operating with layered technology estates, acquired systems, legacy process workarounds, regional complexity, and rising reporting expectations across finance, sustainability, customer experience, and risk. Your uploaded Emergent Africa SEO brief rightly positions master data management, data governance consulting, decision intelligence, and the artificial intelligence data foundation as priority search themes for the local market. That reflects a genuine executive need, not just a marketing opportunity.

Regional evidence points in the same direction. IBM’s Middle East and Africa study found that while leaders are moving quickly towards artificial intelligence outcomes, only 27 per cent are confident in their ability to use unstructured data for business value, and 79 per cent admit they are still early in defining how to scale and govern artificial intelligence agents. The appetite is real. The operating maturity is not yet where it needs to be.

That should prompt a reframing in South African boardrooms. The master data management case is no longer about future readiness in the abstract. It is about present commercial resilience. It affects how confidently a business can automate decisions, how defensibly it can report performance, how quickly it can integrate acquisitions, how effectively it can personalise customer engagement, and how safely it can delegate work to autonomous systems.

8. A real master data management response is organisational, not cosmetic

When executives finally acknowledge the problem, they often ask for a data cleansing exercise. That is too small a response. A serious master data management programme requires decisions about ownership, stewardship, policy, standards, workflow, and accountability. Microsoft’s Responsible Artificial Intelligence Standard requires data governance and management practices tied to intended uses and stakeholders. DAMA makes the same point in broader terms: data management is a discipline with defined principles, roles, and methods, not a one-off intervention.

That means the wake-up call should lead to a sequence of executive choices. Which data domains matter most to commercial value? Which records need to be authoritative? Which policies govern creation, matching, approval, and change? Who owns data quality at the process level? Which exceptions require human escalation? What lineage is required for audit and regulatory defence? Which metrics will prove improvement? These are management questions before they are software questions.

9. The right ambition is not more artificial intelligence. It is more trust

There is a revealing sentence in IBM’s May 2025 announcement with State Street: “Unlocking the full value of AI begins with a strong, enterprise-wide data foundation.” That is the strategic heart of the issue. The winning organisations will not be the ones with the most agent demonstrations. They will be the ones that can trust the identity, status, provenance, meaning, and quality of the data their agents use.

IBM’s chief executive study reinforces that point from another angle: 72 per cent of chief executives say proprietary data is key to unlocking generative artificial intelligence value, yet 50 per cent say their organisations have disconnected technology because of the pace of recent investments. In other words, leaders already understand that data is strategic, but their operating estates often remain fragmented. Master data management is one of the few disciplines specifically designed to close that gap between strategic intent and operational consistency.

10. The immediate executive agenda is clear

The practical agenda now is not mysterious. First, identify the decisions and workflows where agentic artificial intelligence could create material value. Second, map the critical master and reference data those use cases depend on. Third, assess the quality, completeness, lineage, ownership, and accessibility of that data. Fourth, establish governance, stewardship, and control mechanisms before scaling autonomy. Fifth, connect master data management to process redesign so agents work from authoritative records and auditable rules rather than whatever data happens to be reachable.

This is also why so many artificial intelligence pilots stall. McKinsey notes that fewer than 10 per cent of deployed use cases make it past the pilot stage. That is not because the models are incapable. Often it is because the enterprise is not ready to absorb them at workflow level. Broken definitions, fragmented ownership, missing controls, and inconsistent data make industrialisation difficult. Master data management does not guarantee success, but the lack of it dramatically increases the odds of failure.

Conclusion

Agentic artificial intelligence is real. It will reshape workflows, compress cycle times, alter operating models, and change what leaders expect from digital systems. But it will not rescue an organisation from weak data discipline. It will expose that weakness faster, more visibly, and at greater scale. The firms that benefit most from the next wave of artificial intelligence will not be those chasing autonomy for its own sake. They will be those building trusted foundations: clean records, clear definitions, strong governance, observable lineage, disciplined stewardship, and master data management that is treated as a strategic capability rather than deferred housekeeping.

That is the real master data management wake-up call. In the agentic era, the core question is not whether your organisation can deploy artificial intelligence agents. It is whether those agents can act on data your business is prepared to stand behind. If the answer is no, the next investment should not start with another artificial intelligence pilot. It should start with fixing the data foundations that determine whether intelligence can be trusted at all. For organisations serious about decision intelligence, scalable automation, and credible transformation, master data management is no longer optional. It is the prerequisite.

Sources

• Emergent Africa Website SEO Optimisation Specification, March 2026
• National Institute of Standards and Technology, Artificial Intelligence Risk Management Framework (AI RMF 1.0)
• McKinsey, What Is an AI Agent?, March 2025
• McKinsey, Seizing the Agentic AI Advantage, June 2025
• McKinsey, Deploying Agentic AI with Safety and Security: A Playbook for Technology Leaders, October 2025
• IBM Institute for Business Value, 2025 CEO Study: 5 Mindshifts to Supercharge Business Growth
• IBM Newsroom, IBM Study: Chief Data Officers Redefine Strategies as AI Ambitions Outpace Readiness, November 2025
• IBM Middle East and Africa Newsroom, IBM Study: Middle East and Africa Chief Data Officers Redefine Strategies as AI Ambitions Outpace Readiness, November 2025
• IBM Newsroom, IBM Accelerates Enterprise Gen AI Revolution with Hybrid Capabilities, May 2025
• Harvard Business Review, To Create Value with AI, Improve the Quality of Your Unstructured Data, May 2025
• DAMA International, DAMA-DMBOK
• Microsoft Responsible AI Standard v2 General Requirements
• World Economic Forum, Why Data Readiness Is a Strategic Imperative for Businesses, January 2026
• MIT Sloan Management Review, Where to Look for Generative AI Risks, March 2026
• Informatica, How Master Data Management Should Shape Your AI Strategy
• Informatica, What Is Master Data Management?
• SAS, Data Management: What It Is and Why It Matters