AI is only as good as your data (but it can help you fix that)

As AI moves from experimentation into production, organisations are discovering that the real constraint is not intelligence but data – incomplete, inconsistent, and fragmented. The question every CDO is facing is no longer if this matters, but what to do about it.

Flip the logic

But here is the twist: AI is no longer just a capability that sits on top of a strong data foundation. It is becoming one of the most effective tools for building that foundation from the ground up.

A Gartner survey of 353 data and analytics leaders, published in April 2026, found that organisations with successful AI initiatives invest up to four times more, as a percentage of revenue, in unglamorous foundational areas like data quality, governance, and AI-ready people, compared to those getting poor outcomes. Simply put, the organisations pulling ahead are not the ones with the most sophisticated models but those investing one layer deeper.

This article explores four deliberately provocative questions about how AI can build stronger data foundations.

1. Are you still entering data manually?

Modern AI is opening up a fundamentally different way to handle data operations. Instead of navigating complex forms and validation screens, users can interact with an AI-powered interface that performs create, read, update, and delete operations through natural dialogue. This can take shape as a centralised data management platform — a single conversational layer sitting across the master data landscape — or be embedded directly into the source and production systems where data is created day to day. The integration model matters less than the principle: wherever a record is created or updated, the AI is right there with the user as the intelligent steward.

What makes this practical is that the AI does not need to reinvent the toolkit. The proven techniques behind data quality such as pattern validation, duplicate detection, and data classification remain as relevant as ever. What the AI layer adds is something those techniques have never had on their own: reasoning and user guidance. Take duplicate record creation, one of the most persistent and costly problems in master data management. In a traditional setup, a user creating a new supplier or customer record relies on manual searches and their own judgement to avoid duplicates. An AI-mediated flow changes that equation entirely, and the business impact is immediate and concrete.

Picture a sales rep asking the AI to set up a new customer account. In one case, the AI recognises that the company already exists under a slightly different legal name and asks whether to link the opportunity to the existing record before creating a duplicate. In the other case, when the customer is genuinely new, the AI guides the rep through proper setup, pulling the VAT number from the business register, applying the right industry classification, and flagging anything that deviates from standard terms. In both cases, the system reasons across the full dataset, surfaces what matters, and guides the user towards the right outcome.

There is also a second, less obvious advantage. When a data operation happens through a conversational interface, the interaction itself becomes part of the audit trail. Traditional logs tell you what changed and who made the change. An AI-assisted flow also preserves the why: the reasoning, the alternatives considered, the context behind the decision.

2. If you truly care about your data, why track static data quality metrics?

The AI-based orchestration described above does not stop at guiding users through data operations. It extends into the ongoing work of data stewards, with the potential to increase their effectiveness and productivity significantly. The same principle applies: AI does not replace the traditional data management toolkit. It orchestrates them at a scale and frequency that manual stewardship has never been able to sustain.

In practice, this means the AI operates as an always-available assistant to the data steward, with access to the full range of traditional data management capabilities: profiling, matching, classification, standardisation, and the ability to reason across them. A steward investigating a data quality issue no longer needs to piece together a picture. The steward may ask the AI to assess a segment of the data, and it returns a structured view of results and recommended actions.

Picture a steward looking into why a recent customer-segment analysis produced unexpected results. In a single response, the AI reports that a substantial number of customer records carry industry classifications that no longer appear to match the actual activity, and identifies records last reviewed more than three years ago as the most likely source of the drift. Based on this, it recommends a re-classification pass against the national business register, and flags a small subset of records that appear to have changed legal entity and warrant a closer look. The steward's judgement remains central, but the time from question to informed decision is much shorter.

Beyond on-demand use, the AI can be scheduled to run dynamic assessments on a recurring basis, adapting its analysis to the findings it uncovers. These might generate reports summarising the findings, delivered by email or surfaced in a dashboard. The steward's workflow shifts from reactive, waiting for something to break, to proactive, with a regular cadence of AI-generated intelligence informing where to focus next.

3. Is your view of the outside world as rich as your view of the inside?

Suppliers, customers, and legal entities are often well defined internally yet carry almost no external context. A record can look complete in the system while offering only a narrow view of the entity it represents. AI makes it practical to close that gap by enriching internal master data with relevant external intelligence at scale – news coverage, regulatory filings, company announcements, ownership changes, analyst commentary. Crucially, AI can interpret unstructured external sources and link the intelligence back to the right internal record, even when the connection is not obvious from names or identifiers alone.

To make this concrete, imagine a company enriching its top corporate customers, scoring each against dimensions like financial stability, reputational exposure, strategic direction, and compliance standing. Each score is backed by a short, structured justification drawn from the underlying sources. An approach like this could realistically flag material risks that account teams would not have spotted through internal signals alone.

Caution is warranted. Public sources vary in reliability, and scores are only as good as the dimensions and sources behind them. External enrichment is not a substitute for human judgement on high-stakes decisions. But used with care, external enrichment turns qualitative context into something structured, comparable across entities, and actionable across procurement, sales, and risk functions.

For organisations ready to act on it, external enrichment opens new possibilities. It provides a structured basis for competitive intelligence and adds a richer understanding of the external environment to the data foundation. As AI solutions mature, that kind of curated, entity-level context is exactly what enables those solutions to reason more effectively about the world beyond the organisation's own systems.

4. Are you able to leverage your legacy data, or is it trapped?

For many organisations, some of the most valuable data is locked in legacy systems, archived files, scanned PDFs, and document formats that are difficult to work with. AI offers a practical route to unlock that value by extracting information from semi-structured and unstructured sources and converting it into structured data.

The opportunity is not theoretical. Common use cases include converting historical documents into searchable records, extracting key terms from contracts and forms at scale, recovering product or asset data from technical documentation, and enriching current datasets with information that previously existed only in static files.

Imagine a manufacturer using this approach to recover specifications from 30 years of scanned engineering drawings, feeding structured information back into its product lifecycle management system. Engineers servicing or sourcing parts for ageing equipment could then find the right specification in seconds, instead of searching archives that few people know how to navigate. The value is not in digitising the past for its own sake. It is in making trapped information available for reporting, operations, and the next generation of AI initiatives.

This matters because legacy data is often treated as too expensive, too fragmented, or too inaccessible to use. AI changes that calculation. It reduces the effort required to recover value from old assets and helps organisations extend the reach of their data foundation without rebuilding everything from scratch.

Data foundations and the flywheel that follows

Taken together, these four shifts point to something bigger than each of them in isolation. AI is no longer only a capability built on top of strong data foundations; it is increasingly becoming one of the most effective ways to build them. But the real power is in how these shifts compound.