Blog

Blog PostsCommunications in Complex Organizations: Are They Helping or Hurting?

Communications in Complex Organizations: Are They Helping or Hurting?

Using hard data behind ‘Soft Skills’ to help improve performance

Good communications are an integral part of any organization. And the more complex the organization, the more complex the network of communications: from face-to-face conversations, cell and radio, to text, chat, and email.  But with so many layers of communication, how do you know you are getting the right information, at the right time, to the right people to ensure the organization’s success?

Imagine a new staffer who is technically proficient, but isn’t adequately sharing information to support team situational awareness. Or a leader who is unknowingly blocking information exchanges with other parts of the organization. Multiply those bottlenecks by tens or hundreds of people and you can quickly see how any number of breakdowns can compromise a mission or lead to its failure.

Global Blockchain Network

As organizations increasingly rely on distributed, multi-team systems across the military and civilian domains, we need to recognize the nature and impact of these communications. After all, communications are the nervous system that enables coordination, collaboration, and cohesion in and amongst teams.

But how do we uncover where and why breakdowns occur so we can correct them and improve organizational performance? Even in controlled training exercises with human observers, it can be near impossible to monitor the face-to-face-interactions of dozens of people occurring at the same time, notwithstanding the chat, phone, and email communications also taking place. It can be a veritable ‘he said, she said’ to uncover and understand the complex dynamics of why something was or wasn’t said, sent, received, acted on, ignored, or other disconnect.

The solution lies in making use of the hard data and science behind what often is regarded as the ‘soft skills’ of communications. In this post, we explore several methods for measuring and analyzing communications throughout organizations, and how those findings – both quantitative and qualitative – can be used to understand and improve organizational performance.

Conceptualizing the team as a network

Team as Network

As part of a team, it’s easy to think only about your direct communications with your colleagues.  In reality, though, the communications within an organization form a complex network. Your immediate actions can have far-reaching consequences across the organization. This graph shows the team as a network, one where numerous parts are interrelated and dependent on one another in order to achieve a common objective.

ACCRUE user interface

In Army staff exercises, for example, different teams organized by specialties must work together by exchanging information. This tool, called ACCRUE, maps the network of communications across groups over time. This command dashboard shows a dense, well-integrated network of face-to-face interactions…but that is not always the case.

Improving the Network

chat and radio exchanges

This example, from the Integrated Combat Operations Training Research Testbed at the Air Force Research Institute, maps chat and radio exchanges. The top graph shows sparser, less regular communications from an earlier stage in training, whereas the network at the bottom, with dense star-shaped cluster, shows a better-integrated team at a later stage.

Trainers and commanders can use this feedback to improve the team’s performance. Able to see their interactions (or lack of), a team member on the periphery can learn how to better integrate within the team network given their role.

Using this data, commanders can also pose what if questions, testing before and after scenarios to find optimal solutions. For example, how rearranging the physical layout in a command post, changing people’s position in the environment might affect the different cells and staff performance.

How to Collect Communications Data?

Examples: chat, face to face, from the sky, tag cloud, charts

Data can be collected from any source of relevant communication. Most commonly from chat, email, and other text-based analyses, but it can also include face-to-face interactions (from wearable sensors), and in the future will include more radio and other machine communication patterns.

Once in a database they can be analyzed to provide a variety of quantitative metrics, most notably:

  • Network metrics, which include frequency, density, and centrality.
  • Timeliness, which includes time between communications, time spent communicating between or within teams, and time to react to events.
  • Resiliency & adaptation, which includes the ability of the communications to withstand perturbations, and the ability of the network to quickly adapt to new situations.

While these traditional metrics tell the who, when, and where of communications…they don’t tell the whole story. For a more complete, holistic picture, we need to understand what was communicated and why. And for that, we need to more fully understand the content of messages.

Communicating…but about what?

Battle scenarios or fast food?

Communications across the network are generally a positive sign, but what if the discussion veers from the mission to something less relevant? There may be nothing wrong with wanting to know what’s for lunch – that’s part of team cohesion – but at the wrong time, the wrong content can undermine a mission. Understanding the content of communications in context to the mission can provide that clarity.

Content Analysis Approaches

Manual Approaches
Text Labeling
Video Tagging
Event Categorization
Entity Analysis
Named Entity Recognition
Entity Disambiguation
Event Extraction
Statistical Bag of Words Approaches
Latent Dirichlet Allocation
Latent Semantic Analysis
Dirichlet Multinomial Regression
Attitudinal Analysis
Sentiment Analysis
Dialogue Analysis
Emotional Analysis
Related Approaches
Speech-to-Text
Temporal Analysis
Chat Threading

There are numerous ways to assess content; each has its place in the world of content analysis depending on the goals. We’ll look at several examples.

Statistical Topic Analysis

Example Topics
“Terrorist Attack” “Elections” “Protests”
Attack Election Protester
Bombing Vote Student
Terrorist Direct Rally
Terrorism Voter Protest
Terror Observer Demonstration
Bali Voting Front
Killed Poll Group
Bomb General Demonstrator
Samudra Presidential Staged
Noordin Direct
presidential
election
Hundred

Statistical Topic Analysis uses machine learning and statistical approaches to understand content in an objective way. By analyzing a body of documents we can extract topics of discussion to distill the gist of conversations.

Dialogue Act Examples

DA Description Example DAs
Appreciation Appreciation of others appreciate, thanks, helpful,
thank you
Certainty Confident or extreme
statements of absolutes
absolutely, everything, nothing,
I know, in no way, necessary
Frustration Discussion of one’s own
or one’s group’s failings
it sucks, unfortunately, I’m unable, angry, screw that
Question Indications of a question being asked about facts ?, does it, is there, requesting information
Uncertainty Indications of uncertainty or lack of confidence about information don’t know, believe, seems, probably, assuming

In a different approach, by looking at word usage to categorize communications, one can better understand the manner of how people speak to each other and its effect on performance. This can provide valuable feedback, for law enforcement for example, in their interactions with the public. In this context, an officer’s word choice, emotional state or the ‘temperature’ of the conversation can escalate or de-escalate a situation, or help gain trust to obtain needed information.

Combining Content Analysis Approaches

chart of topics tweeted about over the course of Indian election

Entity Extraction + Topic Analysis + Sentiment Analysis + Temporal Analysis

Combining different approaches provides more powerful analysis than any one technique alone. In the example above, which examined tweets from an Indian election, the top graph shows how the discussion of candidates (the entity) changed over time, from one topic to another. The bottom reflects polling sentiment, and how positive or negative people felt about an issue.

Army Command Staff Case Study

tight network in planning, looser network in execution

When we bring together the quantitative aspects of the network interactions, and combine them with the qualitative – what was discussed – we get a richer picture. In an Army training exercise, there’s a difference in the communications at different times, but why? When revealed by the content or context of the interactions, it then becomes clear. On the left, we see tighter communications amongst the CHOPS, S3, and XO, a function of the planning-related efforts, while on the right, the looser, more decentralized network is a function of the execution phase, with people performing their assigned tasks.

Perception and reality

While people generally form mental models in their head of what may be taking place in the world, or in teams, that can differ widely from ground truth. And in complex networked organizations, it can take concrete evidence to make the case otherwise.

Most organizations, currently, do not have a way to analyze and visualize their communications. These methods provide a sort of x-ray vision, illuminating the oft-hidden, yet essential communications that must flow throughout the organization. Using the ubiquity of data now available today, these tools can help diagnose root issues, providing the feedback and evidence to improve organizational performance.