One question still opens most AI conversations inside organisations. It sounds practical. It builds a clean business case. Everything it produces can be counted.
Efficiency matters, and there is nothing wrong with saved hours. The trouble is the size of the ambition. Drucker said it fifty years ago: there is nothing so useless as doing efficiently that which should not be done at all.
Fig. 1 — The appeal is that everything the question yields can be counted. A clean ledger makes a clean business case — and a small ambition.
As a starting question, “can we automate this?” is too small. It is like buying a faster car and asking only whether it can reach the same old place more cheaply. It skips the larger question: where should we be going, now that the context has changed?
Fig. 2 — Efficiency asks how to arrive more cheaply. Ambition asks whether the destination is still the right one.
AI hands organisations a genuinely new set of capabilities.
Fig. 3 — These are real, new capabilities. The risk is spending every one of them on the work we already do.
The economics agree. A Nobel laureate has a name for where the automation question usually lands: so-so automation — technology just good enough to displace a person, not good enough to make anything better. It saves money on paper and shows up nowhere in productivity. In a recent global survey, one percent of executives said AI was fully integrated and producing measurable results. That is not a technology problem. It is a question problem.
So the better question is about ambition.
Fig. 4 — One question hunts for tasks to remove. The other looks for capabilities to build. They lead to very different work.
Asked that way, the questions begin to multiply, and every one of them points at a capability.
Fig. 5 — Each of these moves the conversation from task removal to capability building. That shift is where service design earns its place.
The cleanest example I know comes from IKEA. Their chatbot — Billie, named after the bookcase — came to handle nearly half of all customer queries. The automation case stops there, declared a success. IKEA looked at the other half, found design questions hiding inside it, and retrained 8,500 call-centre staff as remote interior-design advisers. That service earned €1.3 billion in its first year. The chatbot was the small question, answered well. The retraining was the better one: what could we become better at?
Service design works at that second level. It never looks at a task in isolation; it looks at the system around it.
Fig. 6 — A task is one cell in a much larger system. Automate the cell and the system stays the same; redesign the system and the task can disappear.
When AI enters that system, the question is no longer only what the model can do. It is what the service can become.
Take three familiar cases. In each, the countable question has an easy answer, and a better one hiding just behind it.
Fig. 7 — Same starting point, two different questions. The narrow one yields a feature. The better one yields a redesign.
The narrow framing produces a tool. The better framing produces a redesign — of the service, of the organisation’s memory, of the work itself.
This is why I am wary when AI transformation narrows to automation. Automation language sends organisations hunting for tasks to remove instead of capabilities to strengthen, and it rewards volume over learning. The business case is easier to write. The transformation is far less interesting.
Fig. 8 — Productivity should be the baseline we stand on, not the limit we aim for.
A more capable organisation can see its own work more clearly.
Fig. 9 — The dividend of capability is not speed. It is an organisation that learns.
AI is a material for designing better work, and that begins with better questions.
Fig. 10 — Slower questions at the beginning. Far less wasted building later.
These questions feel slower at the start. They save time later, because they stop us building efficient versions of the wrong thing.
The next time an automation idea lands on your desk, try one move before you count the hours it saves: ask what it could make the organisation better at. IKEA found €1.3 billion in the half of the conversations the chatbot could not handle. Your version of that half is sitting in a backlog somewhere, labelled as cost.
Sources: MIT Technology Review — Acemoglu on the economics of AI, MIT Economics — so-so automation, Ingka Group — AI and remote selling at IKEA, PYMNTS — IKEA's 8,500 design consultants