The integration gap: where European fintechs and banks are actually losing the agentic transformation.

A pattern worth naming

Across teams we talk with in European fintech, banking, and insurance, the same conversation keeps surfacing — in different rooms, on different agendas, with different vocabularies, but with the same underlying shape.

The platforms are bought. The roles are filled. The strategy is on the table. Six months in, the agents are still in pilot mode and the day-to-day has not changed.

This is not a technology problem. The technology works. It is not a strategy problem. The strategies are sound. It is the gap between the two — between what the strategy describes and what the operating reality looks like — that is where most agentic transformations are stalling.

We have started calling it the integration gap. And looking at the published research from the last twelve months, it appears to be the central pattern of the field, not a peripheral one.

What the research is converging on

Three findings from the past year, from three different vantage points, tell the same story.

McKinsey's 2025 State of AI survey tested 25 attributes across nearly 2,000 organisations to identify which factors actually correlate with EBIT impact from AI deployments. The single strongest correlation was not the technology, the talent, the budget, or the strategy. It was workflow redesign. The high-performing organisations — roughly the 6% reporting meaningful financial returns — were nearly three times more likely to have fundamentally redesigned their workflows around AI rather than layering AI onto existing processes.

BCG's 2025 global study of 1,250 companies (Build for the Future 2025) found the same shape from a different angle. Roughly 60% of organisations generate no material value from their AI investments. Only about 5% create substantial value at scale. The differentiator, BCG concludes, is not the technology or the spend — it is whether the organisation has done the actual work of redesigning how work gets done.

nCino's AI in Banking Benchmark 2026 — focused specifically on the banking sector and presented at nSight 2026 — found that 84% of banking executives are deploying AI at an enterprise level. A significant proportion, however, remain focused on adoption rather than return. nCino's summary of the moment is direct: "AI capability is no longer the limiting factor — execution is."

Three studies. Three methodologies. Three sectors. One conclusion: the gap between agentic AI being deployed and agentic AI creating value is not closing on its own. It is closing only where organisations are doing the structural work of integration.

Why the gap is widening, not narrowing

A reasonable question at this point: if the research has been consistent for at least a year, why is the gap widening rather than closing?

Three observations from the published evidence and the conversations across the field.

First, the speed of agent capability is outpacing the speed of organisational adaptation. A model that was at the frontier in March 2025 was superseded by autumn. Agents that handled three workflows in pilot can now handle thirty. But the average European bank cannot redesign thirty workflows per quarter — and would not be wise to try. The capability curve and the organisational curve are diverging, and the gap between them is where the integration work sits.

Second, the cross-functional nature of agentic deployment is genuinely hard. A traditional digital transformation involves IT, operations, and a sponsor on the executive committee. An agentic transformation involves legal (because of the EU AI Act and DORA), technical (for the deployment), commercial (for the business case), people (for the workforce design), and data (for governance and traceability). Each function does its job well; the integration of all five is what nobody owns by default. This is the pattern that turns up across the European Banking Outlook 2026 (Deloitte / Gambit Finance) and across the McKinsey and BCG findings cited above.

Third, the strategic conversation has run ahead of the operational one. Most senior teams now know what they want from agentic AI. The board decks have been written, the investment cases have been approved, the platforms have been selected. What has not happened, in many organisations, is the slower work of changing how decisions are made, how oversight is designed, how the workforce reorganises around agents acting alongside humans. That work is less visible than the strategic announcement — and it is the work that determines whether the strategy lands.

Reading the gap across five dimensions

When we look at where specific organisations are getting stuck, the pattern is rarely "we deployed the wrong agents." It is more often that one or two of the five dimensions of agentic transformation have moved ahead of the others — and the misalignment between the dimensions is what is producing the stall.

The five dimensions we use to read this:

Legal and regulatory. Where the EU AI Act, DORA, GDPR, and MiCA become operating architecture rather than paperwork. Stalls here typically present as repeated last-minute interventions from legal that delay agent decisions the rest of the organisation thought were ready.

Technical. Where the agent capability, integration, observability, and stack architecture sit. Stalls here typically present as agents that work in isolation but cannot be safely connected to the customer-facing or decision-critical systems that would make them productive.

Commercial. Where the business case, productivity model, partnership structure, and revenue logic live. Stalls here typically present as productivity claims that do not survive contact with how work actually flows through the organisation.

People and organisational. Where the human workforce sits alongside the agentic workforce — decision rights, culture, ethics, change management. Stalls here typically present as low adoption, workarounds that have settled in, and a quiet sense in the team that the agents have not really changed anything.

Data and intelligence. Where the agents participate in the data layer — governance, traceability, accountability, lineage. Stalls here typically present as repeated delays on agentic decisions because the data quality, lineage, or governance is not yet sufficient to satisfy the risk function.

The point is not that any one of these dimensions is the cause of the gap. The point is that the dimensions move at different speeds, and integration is the work of holding them together long enough for all five to land. Most organisations are doing four of the five well and one of them not yet — and the one that is behind sets the pace of the whole transformation.

What this implies for senior teams

Three implications worth thinking about, framed as questions rather than recommendations.

Where is the integration owned? In most organisations we work with, the answer is nowhere specific. Legal owns legal. Technology owns technology. Risk owns risk. People owns people. The integration sits between them, and falls into the cracks between functions. The question for senior teams is not "do we have someone in each function?" — it is "who is responsible for the alignment between functions, and what authority do they have?"

Where is the workflow redesign actually happening? McKinsey's finding suggests that workflow redesign is the variable that distinguishes the 6% who realise value from the 94% who do not. In your organisation, in what specific workflows has the redesign been done, by whom, with what governance? The question is uncomfortable because the honest answer in many organisations is "we have not yet done that work at the depth that the research is pointing to."

Where is the next twelve months pointing? The shift from pilots to production is the defining transition of 2026 for European banking, as the European Banking Outlook 2026 makes explicit. That transition is not a technology event. It is an organisational event. Where will your organisation be on it twelve months from now — and what would have to change in the next quarter for that trajectory to be the one you want?

A note on what comes next

This is the pattern as we are reading it across the work, from inside engagements, and from the published research that supports the reading. It is not a complete picture — the agentic transformation is still being written, and what is true in May 2026 may be partially superseded by November 2026.

What we believe is durable, though, is the diagnosis: the gap between agentic AI being deployed and agentic AI creating value is an integration problem, and integration is structural work that requires more than a strategy and a platform. The organisations that close the gap first will be the ones that have done the structural work first — across all five dimensions, held together by someone who can see across them.

We will continue to publish what we are seeing as the field develops. If you are working through this from the inside, we would like to hear from you.

Sources

• McKinsey & Company. The state of AI 2025: How organizations are rewiring to capture value. March 2025.

• McKinsey & Company. Seizing the agentic AI advantage. June 2025.

• BCG. Build for the Future 2025: The Widening AI Value Gap. Global study, n = 1,250. September 2025.

• nCino. AI in Banking Benchmark 2026. Presented at nSight 2026. Reported in FinTech Global, June 2026.

• Deloitte / Gambit Finance. European Banking Outlook: Entering 2026. January 2026.