Making AI deliver: A benchmarking framework on how leading companies operationalise AI for impact

Published by Databricks on September 9, 2026 · Updated September 9, 2026

• artificial-intelligence
• ai-governance
• data-platform-modernization
• mlops
• digital-transformation

What is Making AI deliver about?

Making AI deliver is an Economist Enterprise report, supported by Databricks, about how firms are turning advanced AI—including generative AI and AI agents—into durable operational capability. The report argues that the gap between AI’s technical potential and what companies actually use it for is still vast, and it uses a benchmarking framework to show how leading companies make AI work in practice. It opens against a productivity backdrop in which America’s Bureau of Labor Statistics found that business-sector labour productivity rose by 2.2% in 2025, only slightly above the long-run average of about 2%.

How was the Economist Enterprise study conducted?

Economist Enterprise developed and conducted a global survey of 1,221 respondents between November 2025 and January 2026. The sample covered large enterprises with annual revenue of US$500m or greater in 18 countries across the Americas, Europe and Asia-Pacific. The research also included in-depth interviews with senior technology and data leaders, a roundtable with additional senior leaders, and desk research on operationalising AI in 2026.

Which industries, countries and job titles were included in the survey?

The survey covered senior leaders in financial services, banking and insurance; healthcare, pharmaceuticals and life sciences; retail and consumer goods; manufacturing and automotive; media and entertainment; energy, oil and gas; telecommunications; and digital natives. Countries included Australia, Brazil, Canada, Denmark, Finland, France, Germany, India, Italy, Japan, Mexico, the Netherlands, Norway, South Korea, Spain, Sweden, the United Kingdom and the United States. Respondents held roles such as CIO, CTO, chief data/analytics/AI officer, chief data scientist, chief enterprise/data architect, and senior vice-president or vice-president of IT, data, engineering or AI.

Which executives contributed insights to the report?

The report includes interview insights from Ashish Agrawal of KONE; Angie Ruan of Nasdaq; Annabelle Gerard and Kaynaz Behdin of Stellantis; Chas Murphy of Disney; Craig Price of Suncorp Group; Daniel Eitler of Mercedes-Benz Group; Gabriele Ricci of Takeda; Jose Manuel Silva of Natura; Karthik Iyer of Albertsons Companies; Maria Macuare of Mondelēz International; Shimizu Seita of Tokyo Gas; Swamy Seetharaman of CRED; Tal Saraf of Atlassian; Vijay Mehta of Experian; Wendy Batchelder of Centene; and Youngjin Kim of Yanolja Consumer Platform and NOL Universe. The roundtable included Brian Bischoff of CapTech Ventures; Sowmya Gottipati of Estée Lauder; Kim Hales of NRG Energy; Anand Mishra of MIO Partners; Jason Sharples of American Express; David Ramirez of Broadridge Financial Solutions; Francis Shanahan of Peloton; and Ben Celebicic of Hinge.

How widespread is advanced AI across business functions?

Advanced AI is already widespread across the enterprise, but breadth alone does not guarantee impact. In the Economist Enterprise survey, between eight and nine in ten firms are deploying AI, or have fully embedded it, across every business function asked about. Information technology leads, while marketing, research and development, and sales are nearly as advanced. Even legal and compliance, the laggard, sees AI used at scale by about two in five firms.

Why are so many companies deploying advanced AI but still not seeing much business impact?

Many firms are deploying advanced AI widely, but they are not measuring or redesigning work deeply enough for the technology to change business results. Economist Enterprise found that more than four in five executives say their AI programmes are beating expectations, yet only about two in five firms formally require teams to track business impact such as cost savings, revenue or efficiency. Jose Manuel Silva, vice-president for technology and chief digital officer at Natura, says the goal is not adoption but ‘kinetic value’—the moment an AI model converts latent capability into measurable business outcomes. Gabriele Ricci, chief data and technology officer at Takeda, says his board is focused on AI’s impact on the profit-and-loss account, not on the number of pilots or prompts.

What do the companies that are pulling ahead with AI do differently?

Companies that are pulling ahead with AI tend to pair clear plans for AI use with stronger digital plumbing and more investment in staff. The report says firms that consolidated their data estates before the generative-AI wave arrived are pulling ahead, while firms still patching fragmented systems discover that every new use case exposes weaknesses they could previously ignore. The most advanced organisations also treat AI as an operating change, not just a technical rollout.

What does the benchmarking framework measure?

The benchmarking framework measures nine distinct capacities that together show whether a company can operationalise AI for impact. It spans strategy, data foundations, governance, work redesign and the demands of autonomous agents. The report’s central point is that AI capability does not accumulate evenly: one company can have strong AI infrastructure and weak change management, while another can ship agents quickly but lack the controls to keep them honest.

What does practical AI strategy look like if the goal is real business value?

A practical AI strategy focuses on measurable business outcomes, not on adoption for its own sake. The report says leaders should set clear plans for where AI will create value, harden the data foundations underneath it, redesign workflows, train staff and track impact in business terms. Natura’s Jose Manuel Silva describes that shift as moving from adoption to ‘kinetic value’, and Takeda’s Gabriele Ricci says the board cares about profit-and-loss impact more than the volume of pilots.

What does disciplined AI delivery mean in practice?

Disciplined AI delivery means moving from experiments to a repeatable development and release process. About three in five firms take between 7 and 12 months to ship an AI project, and a similar share either lack a fully established development life cycle for AI projects or do not apply it consistently. The firms breaking out of pilot purgatory share three features: a structured life cycle, disciplined attrition and design for reuse. The report calls that combination the scaling engine.

How important are data foundations when you’re trying to scale AI across the business?

Data foundations are one of the biggest constraints on scaling AI, and data movement is often the binding cost. Just over half of firms with unified data architectures say data storage, movement and duplication are their biggest ongoing AI expense, and that rises to roughly two-thirds among firms with less integrated environments. In both cases, that share is more than double the proportion pointing to computing infrastructure. One executive in the report says the gap between building a prototype on clean data and running AI at enterprise scale is ‘not even in the same universe’.

How should companies think about governance when they want to move AI into production?

Companies should govern AI across its full life cycle, not only during development. About three in five firms review AI systems during development and before deployment, but fewer than two in five continue that oversight after a system goes live, when models drift, data shifts and edge cases multiply. One in eight firms reviews governance only when something goes wrong. The report says companies that govern AI from inception through live operation suffer fewer failures and scale with greater confidence.

What kind of workforce redesign is needed to make AI actually work at scale?

AI only scales when firms redesign tasks, retrain staff and change incentives around the technology. The executives interviewed for the report repeatedly said the hardest part is not building models but rewiring the organisation around them. The report says task-level job redesign, meaningful training and the right incentives matter more than the sophistication of AI systems. That challenge is visible in the survey: half of firms cite human review as a top ongoing cost, yet only 4% point to employee upskilling.

What does the report say about AI agents?

AI agents can already do real work in many leading adopters, but they make existing weaknesses more obvious. About three in five leading AI adopters now have autonomous systems doing real work, yet fewer than half of firms mandate a formal framework for autonomous systems. The barriers to scaling agents—accuracy and integration with legacy systems among them—look familiar from earlier AI, but they matter more when software acts rather than merely advises. The report says the firms furthest ahead invested in the control layer of AI before they invested in autonomy.

What gaps do AI leaders need to close if they already invest in AI?

AI leaders need to close five gaps: measurement, data plumbing, delivery discipline, governance and workforce redesign. The report shows that many firms are already deploying AI, but far fewer are tracking business outcomes, standardising delivery, governing systems after launch or investing enough in the people and process changes that make AI useful at scale.