Most companies have tried AI; fewer can point to what it changed. In McKinsey's 2025 State of AI survey, 88% of respondents said their organisation used AI regularly in at least one business function, but only 39% reported any effect on EBIT at the enterprise level. We close that gap: find the work where AI pays, prove it on real tasks, measure it honestly, and roll it out with the rules and training that make it stick.

Our guide From AI pilots to real gains covers the approach in detail.

We start where the time goes. Workshops with each team collect the tasks that repeat, the documents people re-key and the questions they answer again and again. Each idea is scored for value, the hours saved multiplied by how often it happens, and for feasibility: whether the data is ready, how costly a mistake would be and how much effort it takes. The top-right corner becomes the first pilots.

AI pays when the workflow changes around it, not when a chatbot is added to the old one. In the claims example, AI reads the email and attachments and fills in the claim, a handler checks it, and the acknowledgement is drafted: 25 minutes becomes 6, with a person still deciding.

A pilot is a working prototype in real hands, on real work, measured against a baseline taken before it starts. Each week we track how many people use it, how much time it saves and how often its output is accepted without changes.

People's sense of speed can mislead. In METR's 2025 study, experienced developers expected AI tools to make them 24% faster and turned out 19% slower, while still believing they had sped up. So we measure the work itself.

Before anything scales, it runs a test set of real cases with known answers. Each model and prompt is scored for accuracy, critical errors, cost and speed, and only the options that clear the gate go forward. The cheapest model is often good enough; sometimes it isn't, and the test set shows which.

Clear rules help people use AI openly instead of quietly. We write a short usage policy, classify your data from public to restricted, match each class to approved tools, and set up a simple way to request new tools and record every use in an AI register.

In the EU, the AI Act has required organisations using AI to support their staff's AI literacy since 2 February 2025. For a recognised structure, ISO/IEC 42001 sets out an AI management system and the NIST AI Risk Management Framework maps the risks.

Adoption is a people project. We train leaders first, then a network of champions who help their colleagues, then each team with examples from its own work, and track weekly active use team by team. The teams whose work changed most use it most.

Map, prove, measure, roll out: a first pilot in real use in about six weeks, then team by team.

An AI opportunity map, working prototypes, an evaluation harness, a usage policy and data classification, a rollout plan and training for every team. You keep the prompts, the code and the test sets.

Enterprise AI assistants for everyday work, models from several providers for custom tools, retrieval over your documents, evaluation and monitoring tooling, and your existing identity and data-protection controls.

Answers to the questions we hear most often about AI programmes.