Hiring more people: the team doubles, output does not

Hiring more people adds capacity in a straight line and coordination on a curve. Each new person brings their own hours and creates a link with every colleague they need to align with. In two experiments published in 2012, four-person teams spent about half again as many labor hours as two-person teams to build the same model.
Why doesn't doubling the team double the output?
Because hours add up and links multiply. If each person has to align their work with every other person, the number of pairs is n(n-1)/2: 1 pair in a team of two, 6 in a team of four, 28 in a team of eight. The formula appears in the paper by Staats, Milkman, and Fox, who cite Frederick Brooks for the point that more links take more time to keep everyone informed.
Brooks wrote about software projects in 1975, in The Mythical Man-Month, but the reasoning holds for any work split among people. As the notes from a University of Colorado software engineering course summarize it, people and months are interchangeable only when a task can be divided among workers who do not need to communicate. The examples that follow are ours: picking fruit works that way. Month-end close, a sales proposal, and project delivery do not.
In a company's operation, almost all work that matters crosses more than one desk. The new person does not receive an isolated slice. They join a network where they need to know what the others did, and the others need to know what they did.
What did the experiments measure?
They measured the total labor that teams of different sizes spent on the same task. Bradley Staats (University of North Carolina), Katherine Milkman (Wharton), and Craig Fox (UCLA) published the study in 2012 in Organizational Behavior and Human Decision Processes.
In the first experiment, executive MBA students built a 50-piece LEGO model in 34 two-person teams and 33 four-person teams. The pairs spent 36.33 person-minutes on average. The four-person teams spent 52.64, about 45% more by our arithmetic.
In the second, 80 MBA students built a 188-piece set in 12 two-person teams and 14 four-person teams. The pairs spent 72.39 person-minutes and the four-person teams spent 111.50. In the paper's words, four-person teams required 54% more person-minutes.
One calculation the paper does not make, and which is ours: divide the effort by the number of people, and the pair finished in about 36 minutes of clock time and the team of four in about 28. Doubling the team bought roughly 23% of schedule and cost 54% more labor. The larger team does deliver sooner. Each unit of schedule it gains costs far more than the plan assumes.
The limits of this data need to be clear. These are short assembly tasks, with teams of two and four, in a lab. They show the direction of the effect under controlled conditions. The exact proportion in your operation is different, and only your own numbers can tell you what it is.
Why does the hiring plan always miss in the same direction?
Because the person estimating sees the extra hands and underestimates the extra links. The authors call it the team scaling fallacy: as team size grows, people increasingly underestimate the labor hours required.
In the first experiment, another group, of 178 undergraduates, estimated the effort for the same tasks. For pairs, the average estimate was 22.93 person-minutes against 36.33 actual, a miss of 13.40. For teams of four, 30.07 against 52.64, a miss of about 22.6. The optimism is there in both cases and grows with team size.
In the second, 84% of outside estimators predicted a ratio between the four-person effort and the two-person effort that was lower than the observed one. The researchers also asked what each person considered while estimating: 54.3% mentioned coordination gains and 37.6% mentioned coordination losses.
This is what happens in a headcount spreadsheet. The line "three more analysts" goes in as three times the output of one analyst. No line records the hours the current six will spend answering the new three.
Is coordination the only cost?
No. Jennifer Mueller, then a management professor at Wharton, analyzed 26 corporate design teams working on products and services, 238 people in all. According to the school's own coverage, the larger teams produced more in total, but each member performed below the members of smaller teams.
Mueller names this effect relational loss: in larger teams, a person perceives that less support is available from teammates when they hit a difficulty. In the same coverage, she says she saw some evidence of coordination loss and none of motivation loss. We read the school's article, not the original paper, and it does not say at what team size the effect appears.
In practice, lower individual output in a team that grew does not prove the people got worse. It can be an effect of size.
What does a system change in this math?
It changes the shape of the network. What follows is our reading, built on the arithmetic above. When the status of the work lives in people's heads, each person has to ask the others, and the number of links follows the pair formula. When the status lives in a record everyone consults, each person keeps one link, to the record. Nine people stop having 36 possible pairs and have 9 connections to the same place.
That applies to one part of coordination, the part that is a status lookup. What stage is the order in? Who has it? Which version of the quote is valid? What is missing before release? An internal system answers those questions without interrupting anyone, which is why handoffs between teams need structure.
The other part is judgment, such as deciding an exception or negotiating priority between two customers. That remains a conversation between people, and no system removes it. The gain is in keeping conversation for what requires conversation.
At UniTrust, a US insurance IMO, the platform calculates commissions on every paid policy and gives each agent a portal with their own goals, commissions, and scheduled meetings. It has more than 8,000 active users. By the logic of this article, and the reading is ours, every lookup the portal answers in an operation that size is a question that does not have to reach another person.
A practical example: the installer going from 6 to 9
The scenario below is a hypothesis to illustrate the reasoning, not a measured case. A solar installation company has six people in the office: sales, engineering, permitting, purchasing, installation scheduling, and finance. Every project passes through all of them. The status of each project sits in one spreadsheet per area and in the conversations between them.
Sales grow and the plan is to hire three people, one for engineering, one for permitting, and one for finance. The possible pairs go from 15 to 36. Each new hire has to find out, project by project, what the other eight know. The original six answer more questions than before, in the same period in which they are training the new people.
The alternative keeps the hiring and reverses the order. First, a single project record, with stage, owner, and open items visible to every area. Then, measure how many hours came back to the current team. Only then decide how many of the three roles are still needed. It may be all three. The difference is that the new people start by consulting the record instead of their colleagues.
When is hiring the right answer?
When the work is divisible, when the team lacks knowledge nobody on it has, or when the queue shows a real capacity shortage. A team running close to 100% utilization needs slack, and we have shown that the queue grows faster than the load in that regime.
The test before the role fits in three questions. How many hours a week does the current team spend asking for and giving status information? Can the new person's work be done without consulting the others, or does it depend on them at every step? And what will they need to ask, and whom, in the first month? If the answers point to a lot of asking, the hire will return less than the plan says, and the reason is architecture.
Frequently asked questions
My team sorts everything out by talking and it works. Why change?
With six people, talking works because there are 15 pairs and everyone knows everyone. The problem shows up with growth: nine people make 36 pairs, twelve make 66. The conversation stays. The questions a record can answer leave it, and time is left for decisions.
Isn't hiring an assistant cheaper than building a system?
In the following month, it usually is. The comparison that matters runs over two or three years and adds salary, payroll costs, and the hours the current team starts spending on coordinating the new person, which no job requisition records. Measure those hours before deciding. If they are few, the role is the answer.
I have no technical team. How do I know what the system needs to carry?
That part is not technical. Ask the team to write down, for one week, every status question they asked or answered, such as "where is this order?". The list that comes out is the first specification of the record. Building it comes later and is the job of whoever develops the system.
The next step
If your company is about to open roles to keep up with volume, the Operational Architecture Diagnostic is a 30-minute conversation to separate what is a shortage of people from what is coordination a system should carry. Talk to an architect.
Related case studies

Did you recognize your operation in this article?
Book the Operational Architecture Diagnostic: 30 minutes to map where your operation's bottleneck is. No proposal in the first conversation.
Book a diagnostic