In February 2015, Bradford Gray, the lead author of a case study for the Commonwealth Fund, visited the offices of Buurtzorg, a home-care company in the small Dutch city of Almelo. On the table lay a large stack of letters, each sending an employment contract to a new nurse, all waiting for the chief executive’s signature. Jos de Blok told him the organization hired about 150 new nurses a month on average.
Buurtzorg had started with one team in 2007. By 2015 it had 8,000 nurses in 700 teams. Between those nurses and de Blok, there was no line of managers. “Nurses do not report to managers,” the case study notes, “though their work hours are tracked.” The 700 teams shared 15 coaches. The back office had fewer than 50 people.
Somebody still had to do what managers and planners do in other home-care companies: decide who visits whom, keep the records, send the bills. Where had that work gone?
A drawing with five parts
In The Structuring of Organizations, in 1979, the management scholar Henry Mintzberg proposed a way to draw any organization. At the base sits the operating core, the people who do the basic work. At the top is the strategic apex, the people in charge of the whole. Between them runs the middle line of managers. To one side is the technostructure: analysts who serve the organization, in his words, “by affecting the work of others,” such as planners and the people who write standards. To the other side are the support staff, internal services “from a cafeteria or mailroom to a legal counsel or public relations office.”
An organization can be drawn as five parts. The biggest part says a lot about how it runs.
Mintzberg’s point was that the shapes differ, and have to fit the work. In 1981, in Harvard Business Review, he blamed many design problems on “the assumption that organizations are all alike: mere collections of component parts to which elements of structure can be added and deleted at will.” He drew the professional organization, for example, as flat, with a thin middle line, “a tiny technostructure, but a fully elaborated support staff.”
Draw Buurtzorg and the picture is lopsided in a way that answers the opening question. The operating core is almost everything. The middle line is missing. The support staff is small, not fully elaborated. The missing work moved down, into the teams, and sideways, into software. Teams of up to 12 nurses do assessments, care plans, scheduling and the documentation that billing depends on, and an IT system handles the scheduling, records and billing.
Why other companies needed more boxes
Most Dutch home-care organizations, the case study explained, sent nurses only for the services that needed a nurse’s training, and less costly staff for the rest. Once work is split like that, someone has to fit the pieces back together. In Mintzberg’s drawing, that someone sits in the middle line and the technostructure. In an article de Blok co-wrote, the industry’s focus had “shifted from care to production, protocols, and administration.” A nurse who does the whole job leaves fewer pieces to fit. (How pieces get fitted is the subject of coordinating mechanisms.)
The drawing shows where the work went. It doesn’t say whether the result is better. When the health ministry commissioned KPMG to compare Buurtzorg with other providers, its 2015 report, as the Commonwealth Fund summarized it, found fewer hours of home care per patient, 108 a year against an average of 168. But its patients’ later physician and hospital costs were higher, and its total cost per client, adjusted for case mix, came out just below the national average.
Mintzberg’s bigger claim has been tested directly. He grouped organizations into five ideal types, each with one part in the lead. In 1993 three researchers, D. Harold Doty, William Glick and George Huber, tested whether organizations that resemble those types are more effective. To their own surprise, they reported, “the results do not support Mintzberg’s theory.” The drawing is a way of looking, not a forecast.
The engineer with the laptops
Imagine you’re the second engineer at a 25-person startup. Nobody does IT, and you know laptops, so every new hire’s laptop lands on your desk, along with the printer and the office Wi-Fi. On the org chart there is no support staff. In your week there is: most of Monday.
Hiring someone for it looks like overhead. Leaving the box empty puts the cost in the operating core, where nobody counts it, until a release slips.
Software in the back office
Buurtzorg leaned on software for much of what back offices do. The 2015 article de Blok co-wrote says it is estimated that using the IT platform for administration saves about 20 percent of a typical home-care agency’s costs.
AI agents are now moving into the same parts of the drawing. IBM’s HR department runs an agent called AskHR. Employees ask it things like “Where can I see my payslip?” By IBM’s account, it handles around 80 automated tasks, has reached a 94 percent “containment rate” on common questions, and helped contribute to a 40 percent cut in the HR team’s operational costs over four years. In five-parts terms, software has taken over a slice of the support staff.
The drawing raises a question the figures don’t answer. A contained question is one the agent kept; the number doesn’t say the answer was right. And the rest still goes somewhere. If you put an agent on an internal help desk, read twenty of the questions it closed each month and check the answers against the policy. Then read twenty it handed on, and time how long they waited.
At Buurtzorg, the work without a box landed on the nurses. On a help desk run by an agent, if nobody owns the questions it hands on, that work lands on the employee who asked about a payslip and is still waiting.