The Data Crisis Behind AI

Why most pilots stall before they scale - and what the firms getting real returns did differently

Most financial institutions have already approved the AI business case. The pilots are running, the budgets are signed, and the ambition is real. Yet for the majority, the results are falling short of what was promised. The reason is rarely the AI itself. It is the data behind it.

AI is only as good as the data behind it

AI is more sensitive to data integrity than any technology that came before it. That is the part most firms underestimate.

Traditional reporting systems could absorb a degree of fragmentation. Siloed architectures, inconsistent definitions, manual reconciliation at month-end - none of it was good, but the systems tolerated it and the numbers still reconciled. AI does not work that way. Feed a model poor-quality data and you do not get a wrong answer. You get a confident, plausible, and wrong one. In financial services, that carries regulatory, financial, and reputational consequences that a dashboard error never did.

Gartner found that 63% of organisations either lack the data management practices needed for AI or are unsure whether they have them, and predicts that through 2026 organisations will abandon 60% of AI projects that are not supported by AI-ready data. The pattern is consistent: the projects that stall are rarely the ones with weak models. They are the ones with weak foundations.

Through 2026, Gartner expects organisations to abandon 60% of AI projects that are not supported by AI-ready data.

The institutions seeing genuine returns treated this as a prerequisite rather than an afterthought. Several have replatformed entire data estates and embedded end-to-end lineage before scaling AI across the business - accepting a slower start in exchange for results that hold up.

The governance gap is bigger than most firms realise

Gartner predicts that by 2027, 80% of data and analytics governance initiatives will fail. Most fail not because the framework is wrong, but because it is never tied to a business outcome urgent enough to sustain it. Governance built as a control exercise loses momentum within months. Governance built to enable a priority outcome survives.

This is the trap. Firms invest in policies, committees, and catalogues, then wonder why nothing changes on the ground. The work was real, but it was disconnected from anything the business needed badly enough to defend.

Fragmentation compounds the problem because it stays invisible until something breaks. The supervisory record is full of cases where unreliable risk data aggregation drove remediation cost and operational complexity, particularly across AML, conduct, and credit. By the time the issue surfaces in a regulatory finding or a failed reconciliation, it is far more expensive to fix than it would have been to prevent.

The regulatory pressure is mounting

The direction of supervisory travel is now unambiguous.

BCBS 239, the Basel principles for risk data aggregation and reporting, set the expectation more than a decade ago. Progress has been judged “unsatisfactory” across Europe ever since. In May 2024 the European Central Bank published its Guide on effective Risk Data Aggregation and Risk Reporting, which made the demand explicit: end-to-end, attribute-level data lineage. The ECB has made RDARR a supervisory priority for 2025 to 2027, is assessing it through SREP and dedicated on-site inspections, and has warned banks to “step up their efforts” or face escalation.

In the UK, the Prudential Regulation Authority’s Future Banking Data programme, established in 2025, is reshaping how the industry reports. The first phase strips out duplicative templates, but the direction is clear: more granular, more dynamic, more traceable data collection, with the burden of proof shifting onto the firm.

The implication is the same on both sides of the Channel. A firm that has not modernised its data foundation is not only capping its AI potential. It is quietly accumulating supervisory risk.

That makes the starting point an honest one. Not where the documentation says the data stands, but what the infrastructure can genuinely support today.

The conclusion is uncomfortable but simple

The organisations that will get the most from AI over the next few years are the ones building the right data foundation now. Skipping that step does not save time. It only delays the moment the gap becomes visible - usually at the point of scale, a regulatory review, or a board asking why the pilots never turned into returns.

The model was never the hard part. The data always was.

What next

At Delta Capita, we run data maturity health checks for financial institutions - an honest read on where your data estate stands today, the gaps that undermine analytics and AI, and what it takes to fix them. We benchmark your current state, surface the gaps that quietly undermine downstream investment, and map a practical route to target-state architecture. If you are weighing where your data estate genuinely stands today, get in touch.

Sources: Gartner, “Lack of AI-Ready Data Puts AI Projects at Risk” (Feb 2025); Gartner, “80% of D&A Governance Initiatives Will Fail by 2027” (Feb 2024); Gartner, “Successful AI Initiatives Invest Up to Four Times More in Data and Analytics Foundations” (Apr 2026); European Central Bank, Guide on effective Risk Data Aggregation and Risk Reporting (May 2024); Basel Committee on Banking Supervision, BCBS 239 (2013); Bank of England / PRA, Future Banking Data programme (2025).