Companies are spending far more on AI than they did two years ago, and few can show a big result. Only 13% of chief information officers, the executives who run a company's technology, say AI has brought them significant value so far, and only one in five new AI projects succeeds. Spending is about 18 times what it was two years ago.
The few that do well share a habit that costs almost nothing to copy. They decide what the people whose work AI takes over will do instead. Companies with a plan to redesign roles around AI are 2.4 times more likely to be in the group that gets the best business results from it, and only 4% of chief information officers have such a plan.
Freed hours pay off only when someone has a job for them
When software answers the routine questions, what a company gets is spare hours, not profit. The simplest use of those hours is to cut the people, and that saves a salary. The other use is to put the hours on work that customers pay for, and that can be worth much more than the salary.
Ikea's retailer Ingka is the clearest example. It counted nearly €13 million in savings from its Billie chatbot over the first two years. Sales through its remote channels reached €1.3 billion in the 2022 financial year alone. The two numbers measure different things, and the chatbot did not produce all of those sales, but the gap in size shows where most of the value was: in what the 8,500 people did next.
The retraining was specific. Staff learned remote interior design, selling online, building a relationship with a customer, and solving the unusual problems a chatbot cannot.
What it looks like on a customer desk
For example, a supplier of building materials has a customer desk of 12 people. Half of what reaches them is "where is my delivery?" and "send me a copy of the invoice." If an assistant connected to the order system settles about half of those, as Billie did at Ikea, the desk gets back the time of about six people. That is roughly 240 working hours every week.
The manager now has a choice. She can cut the desk to six people and save six salaries. Or she can move those hours to calling the contractors who have not ordered in three months, helping new customers pick materials for a job, and answering late-delivery complaints that wait days for a reply. A single reorder from a lapsed contractor often pays for many hours of the caller's time, so the second use usually wins in any business with repeat customers.
That is why the choice has to be made before the tool is bought, not after the hours appear. Before approving the next AI tool, write down what the people whose work it takes will do with the time, and which figure in the accounts will show that it worked.