Where should a traditional group actually start with AI?

Start with a decision, not a tool. Find the two or three choices the group makes regularly that are slow, expensive or inconsistent today. Get clear on what a better version of that decision looks like, who needs what information to make it, and where the bottleneck actually sits. Only then ask whether AI helps. Sometimes it will. Sometimes a simpler change, a cleaner data feed or a shorter approval chain, does the job. Leading with the tool gets you pilots in search of a purpose. Leading with the decision gets you advantage.

Why the “start with a tool” instinct is so common

It is natural. A CEO sees a demo, reads about a competitor’s chatbot or gets a pitch from a vendor, and the first question becomes: where can we deploy this? The technology is exciting, genuinely capable, and moving fast. But excitement is not a strategy.

McKinsey’s 2025 State of AI survey found that 88 percent of organisations now use AI in at least one function, up from 78 percent a year earlier. Yet only 39 percent report any measurable impact on earnings, and for most of those, the impact is less than five percent of EBIT. The technology is everywhere. The value is not.

Harvard Business Review identified what they call the “AI experimentation trap”: companies launching pilots across departments, chasing quick wins and marginal efficiencies, but never connecting those pilots to real business value. The pattern is familiar to anyone who lived through “digital transformation” a decade ago. Scattered activity, no focus, no returns.

What the companies getting value actually do differently

The pattern among high performers is consistent across the research. They do not start with the technology. They start with the work.

BCG’s 2025 AI research found that companies generating real value from AI share a set of structural accelerators: a multi-year vision sponsored by the CEO, workflow redesign before tool selection, and serious investment in upskilling. The technology choice comes after those decisions, not before.

Bain makes a similar point. In their 2026 outlook, they note that data access and process redesign must be treated as CEO-level problems, not IT problems. When business cases are built on full automation economics but the actual systems still route most decisions to humans, the CFO approved one set of numbers and the organisation lives with another.

A practical starting sequence for a multi-business group

We suggest five steps, none of which require buying anything.

First, map the decisions that matter. In each business, identify the three to five recurring decisions that drive margin, speed or customer retention. Pricing, procurement approval, credit extension, maintenance scheduling. These are usually well known but rarely written down as decision processes.

Second, diagnose where each one breaks. Is the problem bad data, slow escalation, inconsistent judgement across locations, or something else? Be specific.

Third, define what “better” looks like. Faster by how much? More consistent by what standard? Cheaper by what amount? If you cannot quantify the improvement, you cannot evaluate any solution, AI or otherwise.

Fourth, ask whether AI addresses the specific bottleneck. If the problem is that pricing decisions wait three days for a regional manager’s sign-off, an AI model will not help. A delegation framework will. If the problem is that credit analysts spend 80 percent of their time assembling data from four systems, an AI tool that consolidates and summarises might help a great deal.

Fifth, pilot narrow and measure tight. One decision, one business unit, 90 days, clear metrics. Not a “centre of excellence” or an “AI strategy.” A test of whether a specific tool improves a specific decision by a specific amount.

What this means for a group CEO

The temptation is to delegate AI to the CIO or to a newly hired Chief AI Officer. That is fine for implementation. But the question of which decisions matter most, and what “better” looks like, is a business question. It belongs to the people who run the businesses.

Your role is not to pick the technology. It is to insist on the discipline: start with the decision, define the improvement, then evaluate the tool. Everything else is a demo.