Your AI Investment Hasn't Shown a Return Yet. The Challenge Is (Human) Adoption

Almost all the leaders I speak with are deep into how to work with AI, and a striking number of them are doing the same thing: buying the tools and giving someone a title… Head of AI. Director of AI Transformation. AI Lead, et al. Problem solved, job done! It looks great on the org chart, reassures the board, and it can be said out loud on an earnings call. What it very rarely does is change what everybody does every morning.

Seems obvious, that gap between capability acquired and capability being used, is adoption. Adoption is now the entire strategy for most leaders.

Here is why that's hard. The technology clock moves fast; your people clock does not. Most established businesses turn over eight to ten percent of their people a year, which means 90% of your team this year are the same people, doing broadly the same things, in broadly the same way. That is your real metabolic rate. Plus, as we all know “people don't like change”!

Two companies show examples of very different adoption strategies and outcomes.

Ford leaned heavily on AI for design and quality control and separately, it shed around 5,300 salaried roles from its 2020 peak. On a spreadsheet the two decisions look unrelated, they weren't. The automated systems didn't catch what experienced engineers used to catch, because the people whose judgement should have trained them had already gone. As Ford's vice president of vehicle hardware engineering put it, the company assumed that feeding AI the existing design requirements would produce a high-quality product. AI is only as good as the information used to train it.

Ford hired back roughly 300 veteran engineers, known internally as the “grey beards”, to run design reviews and train the very systems that were meant to replace them. It worked, Ford topped the JD Power 2026 US Initial Quality Study among mainstream brands for the first time in sixteen years, with the warranty and recall benefit put in the hundreds of millions. The cameras and the models stayed, what changed is who supervises them.

Institutional knowledge is not an overhead; it is the training data. The most expensive moment in any transformation is the one where nobody left in the building can tell the machine it's wrong.

In 2021 Ingka Group, IKEA's parent, launched an AI assistant called Billie to handle routine queries, order status, delivery windows, returns. Within two years it was resolving roughly 47% of all enquiries, some 3.2 million interactions, saving around €13 million. The conventional next step writes itself. Half the volume had gone, so half the headcount is surplus?

Ingka asked a different question, not “how many people can we remove?” but “what can these people do that the machine cannot?”. They studied the enquiries Billie couldn't resolve and found a pattern hiding in there: people wanted help planning their homes. Consultative conversations, not transactional ones. They retrained roughly 8,500 call centre colleagues as remote interior design advisers. That channel generated €1.3 billion in the 2022 financial year, around 3.3% of group revenue, with a stated ambition to reach 10% by 2028.

Told honestly, the €1.3 billion is the whole remote channel, not incremental revenue from the reskilling alone, and it took years, not quarters, but the decision underneath it is the lesson. They didn't optimise for the automation rate, they found the limits of the machine, then moved their people up into the work worth paying humans for, rather than out of the door.

Compare that with a manager I heard recently at a large professional services firm, describing AI cutting delivery time by around seventy percent while clients were still being billed at the old rates. The work had transformed; the business model hadn't.

Both stories turn on the same thing, someone inside the building owned and drove the change.

Plenty of companies bring in consultancies, nothing kills adoption faster than an outsider who is too good at their job. The organisations that make change stick transfer the status of expert to someone inside. Not just the knowledge, the status and the ownership, someone (or an adoption team) whose name people say when they have a question, who is publicly backed by leadership, and who is measured on the adoption rather than on the rollout. The key is to name the behaviour you are trying to change, not the tool.

Nearly every serious conversation about AI right now is a conversation about the tools and use cases. Almost none of them are conversations about human adoption, which seems remarkable, when adoption is where the entire return lives.

Here's the question worth taking into your next leadership meeting: of everything you have invested in this year, what percentage of your people have genuinely changed what they do every day, and who inside this business owns driving that number? If nobody can answer the second part, you don't have an AI problem, you have an adoption problem, and it is entirely solvable. So, who is your driver of adoption?