Before the release of the System/360 in 1964, computers were largely single and fixed units, so changing one component meant changing the entire computer. The System/360 family of mainframe computers, in contrast, featured compatible software and standardized hardware that allowed their products to be easily reconfigured or upgraded.
This new system was so different in format that IBM had to restructure its organization to build it.
“When we have a big change in technology,” he says Niko Matusekprofessor of strategy at the Kellogg School, “the question is, ‘How do we think this will change the way businesses are organized?’
In the case of IBM, the company decided that producing computers with this type of design would require a more modular manufacturing process. So it split its engineers into separate teams, with each team designing a different component.
But this introduced a new challenge: communication. With so many teams working separately on their projects that would eventually have to come together, team coordination and communication became that much more important. the company had to ensure that necessary information—and only necessary information—was shared with these fledgling groups.
“Designing optimal communication networks is a very complex problem,” says Matouschek. “If you have hundreds of people and they have to decide who to tell their information to and who to do what next—there are hundreds of different possibilities.”
Matouschek collaborated with Kellogg colleagues Michael Powell and Bryony Reich to develop a mathematical model to shed some light on the problem. They focused specifically on communication in the context of companies that, like IBM, manufacture products by assembling separately produced parts or units.
They found that the optimal way for people in these organizations to communicate probably resembles a “core-periphery structure,” in which the organization has a central group of groups that share almost all information with each other, and a group of peripheral groups that rarely do. their information with each other.
“At the core, we talk to each other a lot,” says Matouschek. “People in the district may talk to us, but they don’t talk much to each other. This is a core-periphery structure.’
He notes that this communication structure is not uncommon even outside of manufacturing. In online social networks like Facebook, for example, a person has their main circle of friends who interact with each other frequently. The person is also occasionally involved with some looser social connections, but these connections almost never interact with each other.
“We see this type of core-periphery communication in all kinds of environments,” he says, “and it’s a nice result that, out of all the crazy different approaches that could be optimal, the one that’s optimal might actually look like this. »
Adapt for change
Beyond computers, companies in a wide range of industries—from housing to telephones and airplanes—make their products by putting together separate, interchangeable parts. This technique, called modular productionit allowed companies to tailor their products to customer preferences, outsource manufacturing, and scale production.
With the revival of this technique in recent decades, Matouschek, Powell and Reich turned to mathematical modeling to understand the optimal way for a company using modular manufacturing to have its teams communicate about their work.
They created a model with a few key features. First, for the many decisions the firm is expected to make, the model assumes that those made within groups require more coordination than they have been done between groups. And second, it assumes that every instance of communication comes at a cost of time and energy.
The model’s challenge, then, is to design a communication network that best balances these trade-offs—ensuring that people can share information as efficiently as possible while expending as few resources as possible. In theory, the right communication network would give a company the best chance of maximizing its profits.
Optimizing communication
The model considered several communication strategies. A popular strategy, for example, is to have a tree-like hierarchy in which people at the bottom report information to their immediate supervisor, who then communicates the information to their boss until all important information reaches the top . Another approach is like a matrix: people communicate information to everyone within their department, as well as everyone in related departments.
For companies using modular production, where separate teams manage different parts of the product, another common approach is for each team to simply adopt a “modular communication” strategy, where it communicates with other teams as needed.
But according to the model, this kind of modular or siloed communication could be problematic.
“The model cautions against the common prescription that companies with modular manufacturing operations must adopt modular communication,” says Matouschek. “The model suggests that this will lead to coordination problems.”
Among all the communication strategies the model examined, Matouschek and his colleagues found that the most effective involved organizing groups into two distinct types: the core and the periphery. In the core, teams share almost all of their information with each other, while in the periphery, teams only communicate with each other and with the core to discuss the most critical information.
“The people who happen to be in the core, they communicate a lot. they tell each other everything,” he says. “But they may or may not tell a lot of information to people who are in these different regions.”
This structure ensures that an organization’s core teams are fully informed of important information, while its peripheral teams have as much time as possible to focus on their core work. In other words, it maximizes resource efficiency while minimizing the risk of communication failures.
While the findings are clear, they remain theoretical, according to Matouschek, who sees them less as practical advice than as a basis for future exploration. “The model is miles away from reality,” he says. “But it does suggest that there are good and bad solutions, and that the good ones look like this.”
Discuss the possibilities
There is an axiom in the world of software programming that says, “Adding human resources to a late software project makes it later.” That is, recruiting more people to help complete a task can sometimes make it take longer because of the extra time people need to coordinate.
This is a lesson that Fred Brooks wrote about years after he led software development for IBM’s System/360. During his tenure, he instituted a rule that required everything to be recorded in a central ledger that was available to everyone on the team. In terms of the communication model, Brooks’ approach was basically to communicate everything to everyone.
“Brooks says something like, ‘That turned out to be a disaster, because within months, the book was taller than the Empire State Building,'” says Matouschek. “It speaks to the idea that there’s a real problem with communication, which is that it takes a lot of resources.”
Indeed, Brooks’s approach, although effective in conveying information, was likely unsustainable over time.
Several years later, another well-known software engineer, David Parnas, came out in favor of a different approach. He suggested dividing developers into separate teams that communicated regularly within teams but rarely between teams. In his view, project coordination could be made more efficient by completely abandoning some communication links.
Sound familiar?
Although not a perfect match, Parnas’ strategy offers a much closer approximation to the core-periphery approach suggested by the model. Between the Brooks approach and the Parnas approach, “our model,” notes Matouschek, “says you want to do the latter rather than the former.”
