On 5 November 2026, the Capgemini Research Institute published its World Cloud Report for Financial Services 2027. The findings, drawn from a global survey of 1,100 C-suite executives across banking, insurance and payments, paint a picture of an industry that has moved fast on agentic AI and is now confronting the harder problem of making it work at scale.

Fifty-nine per cent of financial institutions are already piloting or deploying multi-agent AI systems. That figure alone would have been unthinkable eighteen months ago. Yet only one in ten of those same institutions has achieved enterprise-wide orchestration. The other 90% are running agents in pockets, under supervision, or in controlled environments where the stakes are low enough that failure is survivable.

For enterprise buyers, the implications are immediate. The question is no longer whether multi-agent AI belongs in financial services. It is whether your organisation is among the 10% that can run it autonomously, or the 90% that is still figuring out how to connect the pieces.

What the numbers say

The Capgemini survey is one of the larger executive samples published on agentic AI this year, and the granularity of the responses is unusual. Here are the figures that matter most.

59% piloting or deploying multi-agent AI; 10% achieving enterprise-wide orchestration. This is the headline gap, and it is larger than most industry observers expected. Multi-agent AI is not a single technology purchase. It is an operational model in which multiple AI agents coordinate across business lines, hand off tasks, resolve conflicts and report outcomes. Getting from the first deployed agent to the hundredth is not a linear process. It requires orchestration infrastructure, governance frameworks and cross-functional alignment that most organisations have not yet built.

90% using or planning intelligent model routing. As agent deployments multiply, firms are realising that sending every task to the most capable model is economically unsustainable. Model routing, the practice of directing each agent task to the cheapest model that can complete it accurately, is becoming a standard discipline. For enterprises, this means that agent infrastructure is no longer just about model selection. It is about model economics.

52% turning to hyperscalers as preferred cloud partners. Financial institutions are consolidating their multi-agent deployments around the major cloud providers, driven by requirements for regulatory certification, data residency, security frameworks and auditability. The trend suggests that multi-agent AI is becoming a cloud-native workload by default, and that the competitive landscape for agent infrastructure will be shaped by which platforms can satisfy regulated-sector compliance requirements.

Eighty per cent prioritising tech debt reduction. To scale agentic AI, banks and insurers recognise they need to modernise the systems underneath it. Legacy infrastructure, fragmented data and outdated integration layers are now understood to be blockers rather than background noise. The finding is significant because it shows that agentic AI is forcing a broader infrastructure conversation that many firms had been postponing.

Half cannot demonstrate ROI on agentic AI initiatives. This is perhaps the most consequential figure for enterprise buyers. Roughly 50% of financial services firms lack clear frameworks for assessing whether their agentic AI programmes deliver enough value to justify continued investment. Without measurement, programmes become vulnerable to budget cuts, leadership changes and regulatory scrutiny.

Outcomes, not efficiency, are becoming the funding criterion

By 2028, Capgemini projects that the number of executives who prioritise business outcomes as their primary AI funding criterion will increase by 141%. Over the same period, the number focused on efficiency will drop by 42%. The shift signals a maturation in how enterprises evaluate agentic AI. The early phase was about proving that agents could do tasks faster or cheaper. The next phase is about proving that they drive measurable business results.

For buyers, this means that vendor selection criteria are changing. A vendor that demonstrates faster task completion is no longer sufficient. Buyers need vendors that can connect agent performance to revenue, risk reduction, customer satisfaction or compliance outcomes. The suppliers that can build that linkage into their platforms will have a material advantage in enterprise sales cycles.

Only 5% are operating at enterprise scale

The report identifies a small group of financial institutions, representing just 5% of those surveyed, that have already institutionalised the capabilities required to operate agentic AI at enterprise scale. Their edge shows up consistently across three priorities: orchestrating agents across business and technology functions, establishing a trusted cloud foundation for autonomous execution, and ensuring AI investments deliver measurable business outcomes.

The 5% figure is important because it establishes a realistic benchmark. Enterprise-scale agentic AI is not a standard that most organisations have met. It is a standard that a small minority has achieved. For the other 95%, the path is not simply to buy more agents. It is to build the orchestration, governance and measurement infrastructure that makes agents useful at scale.

What it means for enterprise buyers

Do not confuse deployment with orchestration. The fact that your organisation has deployed agents does not mean it can run them autonomously across business lines. Capgemini data suggests that most financial institutions are in the deployment phase, not the orchestration phase. Buyers should audit their current agent estate and ask honestly whether the agents can coordinate without manual intervention.

Build measurement frameworks before you scale. If half the industry cannot demonstrate ROI, the problem is not the technology. It is the absence of metrics. Buyers should establish outcome-based measurement frameworks before expanding agent deployments, not after. This means defining what success looks like in business terms, not technical terms.

Plan for hybrid sourcing. Capgemini projects that more than two-thirds of financial institutions will adopt a hybrid sourcing model within three years, balancing in-house development for differentiation, packaged solutions for speed, and partnerships for ecosystem depth. Buyers should not assume they will build everything internally or buy everything off-the-shelf. The winning approach is likely to be a portfolio.

Consider cloud compliance as a first-order requirement. With 52% of banks and insurers selecting hyperscalers as their preferred cloud partner for multi-agent governance, the competitive landscape for agent platforms is consolidating around providers that can meet regulated-sector compliance requirements. If your cloud partner cannot satisfy your regulators, your agent programme will stall regardless of how good the models are.

What it means for suppliers

Orchestration is the new battleground. With 59% of firms deploying but only 10% orchestrating, the market opportunity is not in selling more agents. It is in connecting the ones that already exist. Suppliers that can demonstrate cross-functional agent orchestration, conflict resolution and unified reporting will find a large market of enterprises that have hit the deployment ceiling and need help moving past it.

ROI tooling is now a competitive requirement. If half of buyers cannot demonstrate ROI, suppliers that build outcome measurement and business-case validation into their platforms will differentiate themselves. Agentic AI procurement is moving from faith-based buying to evidence-based buying. Suppliers need to provide the evidence.

Regulatory credibility matters more than feature lists. Financial services firms are selecting cloud partners based on compliance certifications, data residency guarantees and audit frameworks. Suppliers that can demonstrate regulatory credibility, security certifications and transparent governance will have an easier path into regulated sectors than those that lead with model performance alone.

The Agentic Expo angle

Agentic Expo, at Olympia London on 23 to 24 March 2027, exists because the transformation Capgemini documents is not unique to financial services. It is happening in insurance, asset management, healthcare, manufacturing and logistics. In every sector, enterprises are discovering that deploying agents is the easy part, and orchestrating them is where the real work begins.

The 5% of firms that have achieved enterprise-scale agentic AI will not keep that advantage for long. The tools, platforms and methodologies needed to reach scale are maturing rapidly. The question for buyers is not whether to adopt agentic AI. It is whether to adopt it intelligently, with the orchestration, governance and measurement layers that turn isolated experiments into enterprise-grade operations.

For suppliers, the opportunity is equally clear. The market does not need more agents. It needs better ways to run them. The companies that solve orchestration, connect outcomes to business value, and satisfy regulatory requirements will define the next phase of enterprise AI.

Get Tickets Exhibit at Agentic Expo

Sources: Capgemini Research Institute, "Financial Institutions Accelerate Transition to Multi-Agent AI in Drive for Bottom-Line Value", 5 November 2026. World Cloud Report for Financial Services 2027, methodology: 1,100 C-suite executives surveyed across 14 markets, June to July 2026.