Jay Galbraith (1939 – 2014) was a leading scholar and consultant in the field of organization design.

In several of his books, he mentioned the possibility of implementing a “mirror image organizational structure.”

He argued that a mirror image structure would facilitate effective coordination across units. But he also cautioned that it could be inefficient in some situations.

In a research project with my two eminent colleagues, Ali Yassine and Tore Christiansen, we delve deeper into this topic.

In this blog post, I will summarize what we are doing.

But let’s first clarify what the term means:

A mirror-image structure is established when a unit (e.g., department) organizes its sub-units (e.g., teams) in such a way as to correspond to the structure of another unit that it needs to collaborate with.

For example, consider an IT department with several teams responsible for developing and deploying new IT system components.*

Traditionally, it has been organized in a functional manner, as indicated below.

But after considering the need for coordination with its internal clients, the IT director decides to create a team structure that mirrors the business units that the IT department serves:

In our research project, we use a more complex case, but it’s basically this problem that we look at. 

We have a built a data set that represents an international firm with three subsidiaries. The question we ask is how the three subsidiaries should be organized.

One alternative is to let the subsidiaries select the appropriate structure, given their particular local requirements (their work processes, products, customers, culture, etc.). This may result in three different organizing principles being used, as indicated below.

Alternatively, the firm may adopt a consistent (i.e., mirror image) structure, where the three subsidiaries are organized in the same manner:

So how can we determine which alternative is the best one?

We wanted to develop an analytical method to find the answer.

We used a coordination cost approach, similar to that used in Reconfig, the organization design software tool.

So we built an Excel model that represents the two alternatives described above (mirror image versus hybrid) in different scenarios.

Our decision rule is: Choose the alternative that minimizes coordination costs. This is likely to be the alternative with the lowest number of interfaces between units (but we also take into account the size of the units).

For simplicity, let’s use the Sales & Marketing unit as an example.

Let us assume that the three divisions manufacture different types of components, but sometimes combine their efforts to deliver integrated solutions. Hence the Sales & Marketing departments need to collaborate to create client proposals, agree on pricing and delivery schedules, and so on.

With a consistent, mirror image structure, there will be three interfaces between them:

This increases to 10 (!) if the divisions choose different organizing principles. The reason is that ii implies that the Sales & Marketing department is split in Division 2 and 3 when they choose to depart from the functional structure in Division 1.

However, we also have to take into consideration the benefits that Division 2 and 3 obtain by selecting a model that reflects their particular requirements.

If these benefits are significant, it could make it worthwhile to adopt a model that is unique for each division, resulting in a hybrid and inconsistent structure for the organization as a whole.

For example, for Division 2, it may be the case that adopting a geographical (instead of a functional) model will reduce internal coordination costs. This will be the case if the geographical model is better aligned with the work processes in this division, as indicated below (for simplicity, the figure only shows two departments, with four employees each):

In other words, this problem is actually a particular instance of a more general conundrum: A decision to adopt a mirror image structure implies a trade-off between organization-wide versus local (unit-specific) performance.

We have calculated the results for various scenarios for a test data set that we created. 

The scenarios differ in terms of the degree of collaboration that is required across the three divisions and how much local adaptation that is required by each division.

What we find is that the mirror image structure is superior in most situations.

However, the hybrid model – where divisions select a unique structure – is better in situations where there is a strong need for local adaptation, combined with little need for collaboration across divisions.

If the results are confirmed by others, one implication is the following:  Leaders should think twice before they delegate organization design decisions in firms that rely on a lot of cross-unit collaboration. Such firms need a fairly consistent structure across sub-units.

Feel free to leave a comment below if you have questions or comments. 

*This example is inspired by: Vlietland, J. & van Vliet, H. (2015). Towards a governance framework for chains of Scrum teams, Journal of Information and Software Technology, 57, 52-65.

Example of social network mapping from client project*.

As I discussed in my last post, we need to understand the current organization before we can design a new one.

And we can’t rely on the existing organization chart for this.

We certainly need to know the existing units and reporting relationships, but there are many other elements of an organization that the organization chart does not reveal.

In particular, we need data about the interfaces between different roles and units in the organization.

So we need to ask questions like:

• Who is dependent upon whom to reach their key goals?
• Who delivers something to whom?
• Who collaborates with whom?

This information is usually not recorded anywhere, so it needs to be collected somehow.

I have experimented with a couple of different approaches.

One is social network mapping.

Now “social networks” can mean several things.

In some academic research, one has focused on informal relationships. So one has sent out surveys to employees and asked them to indicate who they prefer to socialize with or share information with.

But when it comes to organization design, the most relationships are task-related – that is, relationships that emerge because people work on the same business process or activity, or participate in the same project.

So we are talking about using the same methodology – but for a slightly different purpose.

The first time I tried this approach, I used my consultant colleauges as guinea pigs.

I created an electronic survey with a list of all employees, and asked my colleagues to fill in a questionnaire and indicate who they interacted with most frequently.

I then used an analysis tool to create a diagram that showed all the relationships.

But all I got was a “spaghetti” of links.

There were some clusters – that is, I could see that there were groups of people who were strongly interconnected.

But it was hard to say what a given cluster represented: Was it project-related? Or did the cluster only show that a group of people interacted, because they were members of the planning committee for the summar party?

So I learned that one had to do this in a different way to get useful and usable data.

I have tried to tweak the methodology into something that can work (a colleague of mine in Deloitte, Subhan Shafiq has contributed strongly to this, and so has Tor Kielland in BrightArch, a software firm that we work with).

The most important thing we did was to add some additional questions in the survey that we distribute.

We ask people to indicate which specific work processes they work on, how much time they spend on each process, and the reason they are in contact with each other.

With this additional information, we can put toghether the pieces more easily.

This type of analysis can help us build a pretty deep understanding of the current organization. It can also help us identify improvement areas.

Let me mention some examples:

• We can analyze data at the individual level to understand who a specific person collaborates with inside and outside the formal unit that he/she belongs to. You see an example from a client project in the image above (I have removed the names of individuals so the image only shows circles representing people).We also summarize the individual level data in table format, such as this:

• We can evaluate the “collaboration load” (the number of people a specific person has linkages to) and consider whether individuals have a fairly equal load, or whether some individuals are overloaded (for example, as a result of having a number of direct reports or because of involvement in a lot of projects or initiatives).

• We can filter the data by process/activity and get a sense of which units/roles contribute to which process.  If we ask people whether a person delivers or receives something in the work process, we can also see the direction of the flow. We can isolate the processes that are the most “coordination intensive” (i.e., require most cross-unit collaboration).

• Perhaps most importantly, we can use these data to evaluate the current formal structure in relation to how work is actually performed in the organization. As one example, we sometimes find that a group of people in one unit collaborates more frequently with people in another department than with colleagues in their own department. This may be a sign that roles should be regrouped.

In sum, social network mapping provides a lot of data that can form the basis for a whole range of analyses.

Even though it’s a well established methodology for academic research, I would say, though, that it is still new and experimental when it comes to the purpose I describe here.

There are still elements that need to be developed further before we have a well tested methodology and clear guidelines for interpretation of the data.

But it is an approach that has great potential.

It dramatically improves how we visualize the current organization and provides a foundation for making facts-based decisions about organizational changes.

*Explanation to the image:  The circles represent individuals (the original chart included names for each circle). The distance between two circles represent the frequency of interaction. The colors represent department affiliation. The size of the circles represents the number of incoming linkages to that individual (a larger circle = more people interact with this individual). A dotted line represent a weaker relationship than a solid line.

Related post:

Taking the white space on the organization chart into consideration (video)