The Secrets to Scaling AI Projects from Pilot to Deployment
Many companies are rushing to use AI, but the most successful ones are moving past just testing it and are now finding ways to get real, measurable results.
A new study, Making AI Deliver by Economist Enterprise, reveals exactly what leading organisations are doing differently to achieve the best results.
The research is part of the Tech Frontiers initiative and is supported by Databricks. It spans 18 countries and draws on a survey of 1,221 senior technology leaders conducted between November 2025 and January 2026.
It includes insights from senior executives at Atlassian, Disney, Estée Lauder, KONE, Mercedes-Benz, Nasdaq, Peloton, Stellantis, Takeda, and Tokyo Gas.
While four in five executives claim AI programmes exceed expectations, only two in five firms have formal requirements for teams to track their business impact.
The study goes on to identify core capabilities that separate leaders from slow adopters, ranging from data foundations to the use of autonomous agents.
According to Eddie Milev, who led the research and oversees the Tech Frontiers initiative at Economist Enterprise, achieving meaningful results is not about being a large firm or using the most sophisticated technology.
He says: âWhat leading companies share is a willingness to change how they work after spending time with the technology.â
Benchmarking for the real world
Most frameworks for AI maturity present it as a linear progression from piloting to optimising. In reality, the journey is rarely a straight line.
Making AI Deliver introduces a framework grounded in the capabilities, including strategy, governance and workforce redesign.
It acknowledges the reality that AI capabilities do not accumulate evenly across organisations.
A company might have an amazing data infrastructure but lack the change management skills to implement it effectively.
Rather than smoothing over these gaps, the framework is built to expose them.
This, in turn, helps leaders pinpoint where capability is misaligned and, crucially, what to do next.
Pillars of successful AI strategy
The firms pulling ahead in the game treat data architecture as the primary constraint on AI success.
Research found that 97% of organisations with a unified architecture report progress on ROI that is ahead of plan.
By prioritising a solid foundation, these companies ensure that their models have access to the high-quality, accessible data required to function.
Real value is also being achieved by building a disciplined route from a new idea to live deployment. This matters because only 40% of firms have a fully established AI development life cycle.
Those that do can move faster and retire experiments that do not work, rather than letting them pile up in pilot purgatory.
Governance remains a hurdle, as fewer than two in five firms continue oversight after a system goes live. Nevertheless, pioneering firms understand that governance cannot stop at deployment.
To achieve this, they layer automated monitoring across the full life cycle to detect drift.
This constant oversight ensures AI remains safe and trustworthy for the long run.
According to the study, the hard part of AI is rarely the model, but rather changing the organisation around it.
Successful businesses embed AI into the tools and routines that already shape daily work.
An example of rewiring the organisation is KONE adding an AI assistant to an existing engineer app, after which customer complaints fell by up to 40% in some markets.
About three in five leading AI adopters already use autonomous agents for real-world tasks. The firms moving fastest invested in governance and control before granting autonomy, turning agentic AI from a potential risk into a competitive advantage.
Leading companies also connect AI projects to specific business outcomes and have the discipline to cut underperforming programmes.
Stellantis, for example, cut its portfolio to 20 programmes which are each required to show measurable value within 12 to 18 months. This focus ensures that AI investment directly serves the bottom line.
While the gap between AI’s potential and its practical application remains wide across industries, the research assures that they all can be conquered.
Organisations of any size can bridge this divide by building the capacities internally, as outlined in the Making AI Deliver study.
