RIZWAN FAROOQUI - Data Strategy & Transformation Leader & RICHARD HENRY - Commercial Director - BTP,Innovation & Data Management

Chasing the Golden Record: What Enterprise Data Gets Wrong, and Why AI Is Not the Easy Fix

Rizwan Farooqui and Richard Henry have spent careers watching organisations pour money into data and walk away with surprisingly little. In conversation with Digital Edge, they explain why the problem persists, where AI genuinely earns its place, and what it would actually take to build a business that trusts its own data.

here is a number Rizwan Farooqui keeps in his head. Not a revenue figure or a project budget, but a failure rate. Roughly 85 per cent of AI projects fail. That is not a fringe estimate. It comes from Gartner, and for Rizwan it reads less as a warning than as a diagnosis: a precise expression of what he has watched play out across twenty years of working with enterprise data, split between consulting at SAP and Accenture and end-client roles at organisations including Coca-Cola and NTT Data Inc. "I have experienced data from both sides," he says. "The ones who feel the pain on a day-to-day basis, and those solving it through strategy, quality, and governance." That dual perspective is what makes him unusually clear-eyed. He is not a vendor with a pitch deck. He is someone who has lived inside the problem and who has spent two decades watching well-intentioned organisations make the same mistakes in slightly different configurations.

Richard Henry, Head of Growth for Data, AI and Business Transformation at Bluestonex, comes at it from another angle and lands in much the same place. Bluestonex is a specialist SAP business technology platform provider, now evolving into SAP's AI platform, and Richard carries the instincts of someone who was an enterprise SAP user before he crossed to the partner side. He has watched implementations go live. He has also watched what happens years later, when the assumptions behind them prove optimistic.

They have worked together for the better part of a decade. What connects them is not enthusiasm for tooling, though both believe in what well-deployed technology can do. It is a shared frustration with the gap between what organisations spend on data and what they actually get in return. That distance, they argue, has a consistent cause. And it is rarely the technology.

The Platform Trap

Most enterprise data conversations begin at the same point: the investment decision. A new platform. A cloud migration. A modern data architecture. The business case is compelling, the demo is impressive, and the logic feels sound. Better infrastructure should produce better decisions. Rizwan has seen this play out too many times to hesitate over the flaw. "A platform without a process, without accountability, without the right rules will never work," he says. "It does not matter how sophisticated the technology is. You are just automating a mess at greater expense." The trap is not that the platform is bad. Often, the technology is excellent. The trap is the assumption buried within the purchase: that the platform will do the work of governance, discipline, and cultural change that the organisation has not yet done itself. It will not. No tool has ever replaced the clarity that has to exist before the tool arrives.

Richard frames it from the other end. If the data going into these systems was unreliable before the investment, the output is just as unreliable afterwards, often worse, because it is now driving decisions at speed and at scale. "If bad data was going into these systems," he says, "the output is equally as bad, if not worse." What has changed is the stakes. There was a time when 80 per cent-accurate data was good enough to spot directional trends and broad patterns. "Those days are gone," Richard says. Decisions are now made in real time, at scale, on that data: credit assessments, customer communications, supply chain movements, fraud detection. The tolerance for error has narrowed sharply. The data practices of most large organisations have not kept pace.

The Question of Ownership

If the platform is not the answer on its own, the next question follows immediately: Who is responsible? The instinct is to look upward, and Rizwan is direct about it. "Gone are the days when data was managed by somebody in a dark room somewhere, isolated, treated as a back-office function. Data governance, data quality, data strategy: these need to start at the highest level of the organisation." If leadership cannot see the value, or the case is not put in terms that register, nothing else moves.

But Richard adds a nuance that shifts the picture. A board-level mandate is necessary. It is not sufficient. "Ownership needs to be pivoted toward the business users, the people who actually rely on the data to do their jobs." If the people entering, managing, and acting on data do not feel personally accountable for its quality, no amount of executive sponsorship will change the output. The data stays someone else's problem right up until it causes a failure too visible to ignore.

The two settle on a working position. Leadership must drive it; leadership cannot do it alone. What is needed is a model where strategic accountability sits at the top and operational accountability lives with the people closest to the data. That is a cultural shift as much as a structural one, and it is harder than most organisations expect. It asks people to own something they have long been able to deflect. Removing that deflection takes sustained commitment and, often, a willingness to surface problems that have been quietly accumulating for years. "The organisations that succeed are willing to look at where their data actually is rather than where they wish it were," Rizwan says. "That requires a culture where surfacing problems is rewarded, not punished. Because if people are afraid to say the data is wrong, it never gets corrected. It just gets hidden until it causes a bigger problem downstream."

Is There a Blueprint?

There is a version of this conversation that hunts for a transferable model: a standard framework an organisation can adopt, implement, and run to fix its data in a defined sequence. I gave them an analogy from the Michael Keaton film The Founder, the story of how Ray Kroc built McDonald's on exactly that logic: a replicable process that could be dropped anywhere and produce the same result. Can data management work the same way?

Both are sceptical, and the reasons are worth following carefully. "Every business exists because it is unique," Richard says. "And those that are not unique will ultimately fail." The specific rules, the governance model, the definition of what a trusted record means in a given context: those have to be built by people who understand the business deeply. You cannot lift a data framework from a global retailer, drop it into a financial services firm and expect it to hold. The underlying assets, the regulatory environment, the relationships between systems and the tolerance for ambiguity are all different.

Rizwan agrees, then supplies the other half of the argument. "There are principles. The mistake organisations make is looking for a complete blueprint when what they actually need is a set of disciplines." Those disciplines are consistent and identifiable: clarity on ownership, agreed definitions of key entities, consistent quality standards, and a feedback loop that catches problems before they compound. The implementation varies enormously. The disciplines do not.

In practice, that means the thinking cannot be outsourced. Organisations can hire consultants, buy platforms and adopt frameworks. Translating any of it into something that works within their own business requires knowledge only insiders possess. The ones that succeed do so deliberately and early.

Where the Golden Record Lives, and Does Not

The golden record sits at the centre of all this. It is the idea that somewhere an organisation holds a single, unified, trusted view of a customer, a product, a supplier: one record that resolves every competing version scattered across systems, teams and processes. For most large enterprises, it remains an aspiration rather than a reality.

Richard describes the fragmentation precisely. A CRM system maintains a single version of customer data. An SAP system holds another. Operational, transactional, and third-party data each sit in separate repositories, built independently, none designed to communicate fluently with the others. The question of where the golden record lives has no clean answer because, in most organisations, it does not yet exist. What exists is a collection of partial truths, each broadly accurate from the vantage point of the system that holds it, each subtly at odds with the others.

Reconciling those fragments is not primarily a technical exercise. The technology can surface the discrepancies; it cannot always resolve them. The same customer appears in three systems with slightly different names, two addresses, one account marked active, and another closed. Resolving that takes someone who understands why the discrepancy exists. Which system should be trusted? Do the two accounts represent the same relationship or different ones? What does the right answer even look like in the context of how this business runs? "That judgment requires someone who understands the business context," Rizwan says. "It is not something you can automate away."

The fragmentation runs deeper than most organisations realise until they go looking. Product data is held in one format for logistics and another for marketing. Supplier records that diverge across procurement, finance and operations. Customer identifiers that do not match across CRM, billing and service. Each inconsistency looks manageable on its own. Together, they form a structural problem that compounds over time and makes reliable analysis steadily harder.

What AI Can Genuinely Do

This is where AI enters, and where both men draw careful distinctions. The promise is real. Intelligent systems can accelerate reconciliation, automate high-volume processing, surface patterns that would take analysts months to find by hand, and compress the time it takes to launch products, onboard customers or respond to the market. Rizwan and Richard believe all of it. They are equally specific about what has to be true first.

Richard offers the clearest example of AI delivering without overpromising. RS Group, a distributor of industrial components, used Maextro AI agents to automate its new product introduction process. Taking a new product from initial data submission to live availability had previously run to around 30 days. With agents handling the automation, it collapsed to minutes. The scale shift was just as striking: where a few hundred new products a week had been the ceiling, the business was suddenly processing several thousand in a day. The outcomes were direct and measurable: more market share, additional revenue, and faster response to supplier and customer demand. "It is not the sexiest AI innovation," Richard says. "It did not create a striking image or generate viral content. But it automated a process that was genuinely painful and time-consuming, and the business outcome was unambiguous." Why it worked deserves as much attention as the result. The problem was well-defined. The data was clean and governed. The rules for a valid product record were clear, agreed and consistently applied. Those conditions made AI the right tool. Remove any one of them, and the picture changes.

Rizwan is clear on where AI is heading, and equally clear on where it stands now. "The direction of travel is exciting," he says. "Agentic AI that continuously monitors data quality, proposes corrections, and proactively manages master data with minimal human intervention: that future is nearer than most organisations think. But are we at the stage where I would let AI do the complete end-to-end onboarding of a new record without human intervention? Not yet. And the reason is not a lack of confidence in the technology. It is a question of trust." The judgment calls that surface during reconciliation are precisely the ones that demand genuine business knowledge: why two records look alike but represent different entities, which source to trust when systems disagree, what an anomaly means in the context of how the business runs. That knowledge does not live in training data. It belongs to the people who understand the business, and until AI can reliably show it has learned the same, the human in the loop stays essential.

The Risk Nobody Is Talking About, At Least Not Loudly Enough

Most AI conversations focus on what AI can do. Richard redirects it to a quieter question: what AI does when the conditions for success are missing. "If the AI is using bad data or ungoverned data to come up with a resolution, is that data now accurate? Did it figure out the right answer because it reasoned correctly, or because it pattern-matched to something that happened to be wrong?" A plausible-looking golden record built on shaky foundations is, in his view, more dangerous than an acknowledged gap. It manufactures false confidence. Decisions get made on a record that looks authoritative but isn't. The failure stays invisible until, suddenly, it is not.

Rizwan reinforces the point from the strategic level. "We need to be wary about throwing AI at every problem as though it is universally applicable. The failure rate in AI projects is high precisely because organisations deploy AI before the foundational conditions are in place." Sequence is everything. Data quality, governance and agreed definitions have to come before AI, not after. When they arrive after, or when organisations assume the AI will generate them, the result is a system operating at great speed and great confidence on fundamentally weak ground.

Then there is what AI lacks and cannot acquire. "The functional and business knowledge that AI does not possess," Richard says, "that is your intellectual property." The understanding of what makes your organisation distinct, what your customers need in a particular context, what a good outcome actually looks like in a given situation: that is irreplaceable. Organisations that deploy AI well use it to extend human judgment. They do not use it to replace judgments they have not yet built.

Accountability in an Autonomous World

The conversation about AI autonomy tends to run hotter than it runs rigorous. Rizwan and Richard are clear about where the limits of automation must be set, and their reasoning is simple and non-negotiable. "Accountability has to sit with a named individual at every point where a decision matters," Rizwan says. "You can automate the process. You cannot automate the responsibility. And the organisations that try to are the ones that discover too late what they have given up." "Someone needs to take ownership," Richard adds. "Always. Because you cannot fire an AI." When something goes wrong, a bad credit decision, a regulatory breach, or a customer record that produces the wrong outcome, accountability has to land on a person. Autonomous systems can operate with a speed and consistency that no human matches. But where that automation runs, what guardrails contain it and when a human steps in remain human responsibilities. They have to.

For any of it to work, the people in the business need to understand enough about how the AI operates to recognise when something looks wrong. That is not chiefly a technical skill. It is a business one. The people closest to a customer relationship, a product category or an operational process are best placed to notice when an automated output does not match the reality they know. Protecting that judgment and building the structures that feed it back into the system matters as much as the deployment itself.

What Actually Separates the Organisations That Succeed

By the end of the conversation, a pattern has hardened in how the two describe the enterprises that get this right. It is not the size of the technology budget. It is not vendor selection, platform sophistication or the scale of the deployment. It comes down to something harder to buy and easier to lose. "The organisations that succeed are the ones that did the thinking first," Richard says. Clarity on what they are trying to achieve. Clarity on who owns what. Clarity on what a good outcome looks like before the implementation begins. The common failure mode is the reverse: a platform decision taken before the strategic questions are answered, then years spent retrofitting governance onto a foundation never designed to carry it.

Rizwan adds the internal dimension. Data has been called the new oil and the new gold long enough that the phrases have faded into background noise. "A lot of organisations realise its value," he says. "But how many are actually taking the right steps forward?" The gap between knowing data is strategically important and treating it with the discipline that implies is still, after all these years, the thing worth closing.

There is no elegant shortcut. The blueprint every organisation wants does not exist in transferable form, because the rules, the governance model and the definition of a trusted record have to be built from the inside. What is transferred is the set of disciplines that make it possible: ownership at every level, agreed standards, a culture that surfaces problems rather than hides them, and the rigour to get those foundations right before reaching for the tools that depend on them.

That is not a data problem. It is not a technology problem. It is the kind of organisational problem that has always been hardest to solve, because it demands sustained commitment with no visible finish line. Rizwan and Richard have each spent twenty years on it. Neither thinks it is close to solved.

What they do believe is that the organisations willing to face it honestly, to look at what their data actually is rather than what they wish it were, are the ones that will find AI genuinely transformative rather than quietly disappointing.

The technology is ready. The question is whether the organisations are.

Rizwan Farooqui is a Data Strategy and Transformation Lead with 20 years of experience in enterprise data management, MDM, and data governance across global organisations. Richard Henry is Head of Growth for Data, AI and Business Transformation at Bluestonex.