How to start an AI programme
AI is no longer an experiment on the edge of the business — it is becoming the operating model. This guide sets out a practical, no-nonsense way to start an AI programme: how to pick the right first use cases, build the foundations of data and governance, and scale from early pilots to agentic systems that deliver real value.
Most organisations don't fail at AI because the technology doesn't work — they fail because they start in the wrong place. They chase a flashy pilot, get stuck in proof-of-concept, and never connect the work to real business value. Starting an AI programme well is less about models and more about focus, foundations and follow-through. Here is a practical way to begin.
1. Start with the business problem, not the technology. Resist the urge to "do something with AI". Instead, identify two or three problems where better decisions, faster processes or lower cost would genuinely move the needle — customer service response times, fraud detection, demand forecasting, document-heavy back-office work. Size the value and the feasibility of each, then pick a small number of high-value, achievable use cases to begin with.
2. Secure sponsorship and set an ambition. AI programmes stall without a senior owner. Appoint an executive sponsor, agree what success looks like in twelve months, and be honest about ambition: are you automating tasks, augmenting your people, or redesigning whole workflows? The best programmes aim to reshape how work is done, not just bolt AI onto the side of it.
3. Get your data and foundations in order. AI is only as good as the data feeding it. You don't need a perfect data lake before you start, but you do need to know where your key data lives, whether it is trustworthy, and how it can be accessed safely. In parallel, choose a pragmatic technology approach — a modern cloud platform, access to leading foundation models, and the ability to ground those models in your own data through retrieval so answers are accurate and specific to your business.
4. Build governance and trust in from day one. Set clear rules for how AI can be used, what data it may touch, and where a human must stay in the loop. Address privacy, security, bias and compliance early — not as an afterthought. Responsible AI isn't a brake on progress; it's what lets you scale with confidence and keeps regulators, customers and your own people on side.
5. Run focused pilots, then prove value. Deliver your first use case as a tightly scoped pilot with a clear owner, a deadline and defined success measures. Keep humans reviewing outputs, gather feedback, and measure the real impact — time saved, errors reduced, revenue influenced. Kill what doesn't work quickly, and double down on what does.
6. Invest in people and change. The hardest part of AI is rarely the model — it's adoption. Train your teams, redesign roles and processes around the new capability, and create champions who help colleagues trust and use the tools. A brilliant AI system that no one uses delivers nothing.
7. Scale deliberately — towards agentic AI. Once a pilot proves value, industrialise it: harden the engineering, connect it to core systems, and reuse the platform and governance for the next use case. The frontier is shifting from copilots that assist people to agentic systems that can carry out multi-step tasks on their own within guardrails. Design your programme so you can move up that curve safely as the technology matures.
Done well, an AI programme is a repeatable engine, not a one-off project: a portfolio of use cases, a reusable platform, strong governance, and a workforce that knows how to work alongside AI.