Enterprises are standing on the edge of an AI revolution, racing to gain valuable insights, optimise their processes, and enhance their workforce. But despite going all-in on AI, only three-in-ten (30%) CEOs are confident about revenue growth in 2026, as most struggle to turn their investment into tangible returns.
That gap between promise and impact is now impossible to ignore. Too many businesses have rushed to deploy AI, only to find it much harder to move into production in a way that drives value.
Instead of rolling out AI assistants, driving AI Return on Investment (ROI) demands a more fundamental shift in how organisations think about processes, data, and governance. To really activate an enterprise’s data and extract value from AI, enterprises must follow these three steps:
1. Rethink processes, not productivity
One of the biggest mistakes organisations make is applying AI to existing processes and expecting it to drive transformation. True impact comes from rethinking workflows to harness AI’s distinct strengths, rather than using it to do the same work a little faster.
To generate meaningful return on investment, organisations need to rethink work at the level of roles and outcomes. Instead of asking how AI can assist an existing function, enterprises should ask where it can own or orchestrate parts of a workflow altogether.
Take contract analysis and negotiation: AI can give a great overview of the content and highlight potential areas to update. But a human still needs to verify the output, cross-check related records, and ensure any changes are accurate and compliant.
For AI to become more than a summarisation tool, it must be connected to the broader flow of contract data: legal terms, procurement records, approval systems, obligations, and related APIs. Only then can AI agents validate context, trigger next steps, and support decision-making. But this level of independence requires orchestration. And orchestration needs data.
That is where many organisations hit a wall. They have data sources everywhere, but not enough visibility into what data they actually have, where it lives, or how it connects across the business. Without that, it is difficult to even begin thinking seriously about enterprise-scale AI.
2. Prepare data for action, but do it securely
Many organisations are rich in data but poor in data accessibility. Information is spread across cloud platforms, SaaS applications, legacy systems, and business units. Without a clear way to view, access, and connect that data, it becomes almost impossible to identify where AI can drive value, let alone scale it safely.
Before enterprises can get meaningful returns from AI, they must activate their data. This means building a foundation that helps enterprises to connect their data with APIs, understand how systems interact, and ensure data can be utilised by AI systems so they can unlock new value streams for the business.
Of course, connections come with risks too. Data and its connections need to be monitored, governed, and protected. This is best achieved by creating a control plane for APIs and AI agents – a governance layer that can be reviewed and inspected regularly, giving a clear audit trail and traceability for data.
The enterprise should be able to understand basic but increasingly urgent questions: Where is our sensitive data? Has it been tagged correctly? Can we still identify it as it moves through systems? Are downstream applications and AI agents using it appropriately? Do we know if it crosses a boundary it should not cross?
These are no longer theoretical questions. In highly dynamic environments, where data, APIs, and AI agents are all moving faster and faster, organisations need the ability to surface insights in real time and manage all of those moving parts with confidence.
3. Start small and experiment fast
The other key shift is cultural. Enterprises need to embrace this era of experimentation, but in a practical way. Enterprises can’t wait for the perfect, all-encompassing AI use case before getting started. In fact, some of the most valuable AI projects are the least glamorous.
Too many organisations jump straight into highly complex scenarios that take months to design and deploy, only to discover major gaps when testing begins. A better approach is to start with low-effort, high-impact use cases. By rethinking the processes behind these use cases from the ground up and with AI in mind, enterprises can solve real operational pain points, deliver visible value, and create momentum for broader adoption.
Invoice reconciliation is a good example. It may not sound exciting, but for finance teams it can be repetitive, time-consuming, and costly. Automating even part of that workflow with AI can save substantial time, reduce errors, and free employees from deeply manual tasks. For a CFO looking at an annual cost burden in the millions, that kind of use case can generate a compelling return very quickly.
The opportunity, then, is not just to have more data, but to become better at identifying what is valuable, activating it quickly, and applying it where it can improve outcomes. This needs to happen fast, because the pace of change in AI is already extraordinary, and the volume of enterprise data continues to grow at the same time.
Setting data up for success
Every organisation understands that data equals value. That is why they protect it. But if a business is spending more money to extract, process, protect, store, and archive data than the value they actually gain from it, that data starts to look more like a liability.
To keep innovating and meet rising expectations, organisations must move away from rigid legacy tooling and toward a more dynamic, data-driven approach to governance. This includes live observability across data and APIs, policy enforcement applied at runtime as conditions change, and federated governance that ensures business rules and compliance requirements are consistently applied across different domains. These capabilities create the trusted data foundation that AI agents need to perform tasks and transform data into value. And for enterprises looking to finally turn AI ambition into real business return, getting data in order is not just a technical prerequisite. It’s the starting point for everything that comes next.
