GenAI pilots, productivity gains and the bottom line

In July 2025 MIT NANDA put out a figure that has turned up in every boardroom presentation since: 95 per cent of organisations see no measurable return on generative AI. The report looked at over 300 public AI initiatives, spoke to 52 organisations and surveyed 153 executives. General-purpose tools such as ChatGPT and Copilot have been explored by more than 80 per cent of organisations and rolled out by nearly 40 per cent, but they mainly help the individual employee. For custom enterprise tools the funnel is steep: 60 per cent evaluate them, 20 per cent reach a pilot, 5 per cent make it into production. The figure has since been disputed. The sample is small, and the report calls itself preliminary and its own figures directional. The bar for success is also high: in production, with a measurable effect on profit. So the figure says something about a strict definition of success in a small group, and less about the average organisation.

A year later McKinsey finds the same pattern in a larger survey. Almost nine in ten organisations regularly use AI in at least one function, and 44 per cent say AI is now being scaled across the organisation, up from 38 per cent a year earlier. Eight in ten respondents say AI has improved their own productivity. Yet the share of organisations attributing part of their profit to AI remains stuck at 37 per cent, the same as in 2025. The group of front-runners attributing at least 5 per cent of profit to AI is hovering at around 6 per cent. According to McKinsey, these organisations do one thing differently: they redesign the workflow around AI instead of slotting AI into the existing workflow. One in five organisations is now scaling back its use because of the running costs.

In spring 2026 BCG surveyed almost twelve thousand employees, managers and leaders in more than twelve countries. Three quarters use AI daily or a few times a week, 23 percentage points more than in 2025. Of regular users on the front line, 42 per cent say they save about eight hours a week, and two thirds of them get little or no guidance on what to do with that time. Employees with clear guidance perform better than employees with more tools. The share of organisations using AI to redesign workflows end to end rose from 22 to 42 per cent.

In November 2025 the Capgemini Research Institute surveyed more than 1,500 executives at companies with revenues of over a billion. Of these, 38 per cent say they have brought generative AI into day-to-day operations, and nearly two thirds have started pausing low-value projects to concentrate the money on a few applications. Meanwhile the AI budget is growing from 3 to 5 per cent of the company budget. So the clear-out of pilots is already under way, before the return has been measured everywhere.

Ethan Mollick describes the same gap from the technology side. In September 2026 Mollick writes that current models can handle work that would take a person weeks, provided they are well directed, while most use falls far short of that. People who bring deep expertise, broad knowledge, taste and initiative of their own get much more out of them, according to Mollick. Organisations’ systems and processes are adapting more slowly than the models.

In the Netherlands the starting point is more modest. Statistics Netherlands (CBS) finds that one in six companies with two or more people employed used AI in 2025, double the share in 2023. Among companies with 250 or more people employed it is 66 per cent, among the smallest 14 per cent. Use is concentrated in marketing and sales (35 per cent) and in administration and management tasks (32 per cent). Three quarters of companies have not even considered AI, and of those that considered it and decided against, 73 per cent cite lack of experience as the reason.

How solid is this? Three independent institutes measured the same gap between individual gains and business results in 2025 and 2026, and McKinsey has added a second year of data that confirms the pattern. The series is still short and the best-known figure is disputed, so the picture is not yet settled.