Most organisations ask the wrong question when AI starts handling a significant share of their workload. They ask: how many people can we cut?

IKEA asked a different one. And the answer was worth €1.3 billion a year.

When IKEA deployed its AI chatbot Billie in 2021, the results were straightforward. By 2023, Billie was handling 47% of all inbound customer service inquiries — roughly 3.2 million conversations, nearly €13 million in operating savings. The routine work was being absorbed by a machine. The question of what to do with the 8,500 call-centre workers whose work had shifted was, for many organisations, easy to answer.

IKEA asked instead: what could these people do that the chatbot never can?

The answer turned out to be interior design advice — a service that hadn't previously existed at scale, that customers would pay for, and that required exactly the human judgment AI cannot replicate: taste, empathy, spatial reasoning, and the ability to understand what someone actually wants, not just what they literally asked for. IKEA retrained all 8,500 workers as remote interior design consultants. The new channel generated €1.3 billion in annual revenue within its first year.

The IKEA question is simple. It is not often asked.

A 2026 EDUCAUSE survey of organisations on how they are responding to AI found that 80% are encouraging staff to develop skills on their own. A further 71% are offering in-house training. Both responses are valuable. Neither is job redesign.

What they don't answer is the question that determines whether an AI deployment actually changes outcomes: what does this person's role look like now that the AI is handling the routine work? Without an answer, organisations get efficiency metrics that look good while the organisation doesn't feel different — and eventually, they get the Klarna outcome: a highly publicised AI deployment, a quiet reversal, and a CEO admitting in a Bloomberg interview that "we focused too much on cost."

The clearest counterexample to this pattern that I've found is Tan Tock Seng Hospital in Singapore — and what makes it stand apart is not the technology. It is the numbers, and the process.

Working with their healthcare workers' union through Singapore's NTUC Company Training Committee framework, TTSH deployed a predictive patient monitoring system and an AI scheduling tool. They then measured what most organisations don't bother to track: outcomes that matter to the people doing the work.

The results: a 288% increase in nurse-led medication reviews. A 42% reduction in patient falls. Nurse roster preparation time cut from 90 minutes to 15 minutes. S$1.09 million in annual cost savings reinvested in patient care. Sixteen innovations expanded from one ward to thirty-six across the hospital.

The mechanism was co-design — workers, management, and NTUC brought into the same room before deployment, not after. The CTC framework requires that productivity gains from AI translate into wage increases, career development, or both. That requirement is structural, not voluntary.

The broader pattern, across IKEA, TTSH, and DBS Bank's decade-long measurement discipline, reduces to three things the organisations getting real outcomes are doing differently: they ask what people can do that AI can't, before they deploy; they measure outcomes that matter to people, not just to dashboards; and they design the change with the affected workforce, not for them.

None of these require a specific national framework or industry context. They require a decision to treat job redesign as a deliverable — not the consequence that gets managed when adoption lags.

The full research behind this piece, including verified outcome data for all three cases, is available as a free whitepaper: Download it at Blue Connector

Jon Scheele is the founder of Blue Connector, which helps organisations connect technology strategy with real outcomes. He is based in Singapore and hosts The Loop Asia. More at blueconnector.co