BETA
This is a BETA experience. You may opt-out by clicking here

More From Forbes

Edit Story

Data-Minded Or Data-Blinded?

This article is more than 4 years old.

Artificial Intelligence (AI) and HR have become closely acquainted in recent years but as AI penetrates deeper into organizations, there are challenges when it comes to applying AI into analyzing employee networks.

Artificial Intelligence is being applied to almost anything and everything, with even the most banal information delivering telling revelations once the data is collated and interrogated. Anyone in doubt should consider the last time they visited a favorite online shopping store and glanced at some of the ads being offered up; the techniques used to analyze what you have bought and rejected to second-guess what you might buy can produce startlingly accurate results.

This doesn’t just happen with online shopping, our social feeds are subject to network and sentiment analysis to make all kinds of "educated guesses" about us from what brand of shoes we might like best to what political party we might support. It’s not a big leap from there into using network analysis to get under the skin of employees, with a goal of improving employee engagement, productivity, creativity and innovation at work. HR has many use cases for AI, the most common being in the recruitment and hiring phases, but one of the "killer apps" of AI in HR is believed to be AI's ability to look at organizational network data.

But this approach to HR should be handled with care. As with much of AI, it's important to consider the ethics of what you're doing and the implications of the technology. Using AI to make shopping suggestions or recommending a cat video in your social network is a very long way from making decisions or suggestions relating to someone’s career path.

Collaboration and networks are about people not data. People come first, then data. Not the other way around.

Laura Weis

As part of my focus on implementing cutting-edge AI solutions, I spoke to Dr Andrés Cardona & Dr Laura Weis from within the Brainpool.ai network about this matter. They are both experts in the uses, potential and pitfalls of network analysis in organizations. Laura has a PhD in business psychology from UCL and is a London-based organizational psychologist and consultant in people analytics, data-driven HR and social network analysis. Andrés is based in Berlin and is a data scientist and economist with a PhD in social sciences working as an independent consultant in people analytics, network analysis and cloud-based solutions for advanced analytics and machine learning. Here is what I learned from them about the opportunities and considerations in using Social Network Analysis in organizations.

Borowska: Can you put the use of network analysis inside organizations into a broader context?

Laura & Andrés (L&A): Attempts to re-engineer organizations using data from social networks at work have fallen short of their promise. Instead of creating increased employee engagement, productivity, creativity and innovation, Social Network Analysis in organizations has spread confusion and distrust in data-driven HR-management. Useful people analytics start with people and a clear purpose in mind, not with a data-fishing trip with an uncertain catch.

Work in organizations has become increasingly agile, collaborative and digital. Flat hierarchies and project teams are no longer exclusive to high-paced digital start-ups. Emails, chats, and apps like Slack, Yammer or Skype dominate our communication at work. In today's on-the-go business culture, real-time digital collaboration tools make collaborative working seem infinitely easier than ever before, allowing people to work seamlessly despite being located around the globe.

Not long ago, quantifying and understanding collaboration and the networks of relationships among employees it creates was confined to the pristine world of academia. Today, under the banner of "People Analytics" and "Organizational Network Analysis (ONA)," data-savvy HR-departments, and data-hungry consultants, scientists, and app developers alike are trying to use the stream of data produced by collaborative work to re-engineer organizations.

Borowska: Why are organizations so drawn to network analysis?

L&A: The idea is simple and intuitive: data about the networks people build when they work together can be used to increase engagement, productivity, creativity, innovation and boost organizational performance.

Borowska: What type of data is collected?

L&A: Data about who talks to whom, who asks questions and who answers them, who stands in the center of attention and who prefers to wait on the sidelines, data about popularity of team members and connections among teams, data about frequency and content of chat messages and emails. In short, data about the daily activities of a modern knowledge worker in a digitally-connected workplace.

Borowska: This goal is so difficult to achieve, but why is that?

L&A: As appealing as it sounds, the effective use of organizational networks’ data to improve business outcomes is a complex enterprise. It requires more than simply setting up a system with state-of-the-art-looking analytics and visuals. Data scientists and managers alike are often blinded by flashy data and fail to put themselves into the shoes of the people whose work environment they claim to want to improve. To make a long story short, and running the risk of stating the obvious: collaboration and networks are about people not data. People come first, then data. Not the other way around.

The flow of ideas and information doesn’t occur in a vacuum. Effective and healthy collaboration, especially among knowledge workers, requires trust and what psychologists refer to as Psychological Safety.  If people feel comfortable to speak their mind, take risks, make mistakes and give and receive feedback, they are much more likely to contribute to the full of their capacities. Key conditions for Psychological Safety are interpersonal trust and emotional predictability. Flesh and bone social relationships are at the core of meaningful and authentic communication. Unfortunately, digital traces available today fail to capture these qualities and can merely provide some proxy for social connectivity. Consequently, there is a danger of conveniently using reflections of digital realities, in the form of social network data, as undeniable indicators of how people interact.

Borowska: And I guess there are many other social factors that come into play as well?

L&A: Focusing on digital traces of collaboration such as email, chat, or comments in forums as measures of network quality is misleading. The number of emails or messages among two people does not reflect their level of trust, safety or collaborative effectiveness. On top of that, personal preferences in the use of digital communication vary, with some people using digital channels for 90% of their interaction at work and some only using it for 10%. Yet, these and other individual differences tend to be ignored when deriving social network insights.  This leads to ambiguity and data that is prone to misinterpretation, undermining meaningfulness, effectiveness, and viability of network analysis.

Borowska: Are we on the road to handling these factors?

L&A: Some new technologies have been developed to overcome the limitations of traditional digital traces. So-called sociometric badges “sociometers” are a bundle of sensors carried by employees around their necks. They capture in almost real-time location and movement, and collect metadata about conversations such as pitch of voice and turn taking. Data protection issues aside, just knowing that your employer registers literally every step you make and could potentially analyze all conversations with colleagues is quite unsettling. As such, the line between fancy analytics and Big-Brother surveillance is easily crossed causing major privacy concerns that are ultimately known to reduce employee engagement and productivity.

Related to privacy issues is informed consent. Even if employees are informed about what data will be used for, most lack the experience with social network research, and are as such unaware of the consequences of network information. For example, if someone decides not to take actively part in a network analysis, this does not mean that this person won’t appear in the network. One of the main goals of ONA has been to identify informal leaders. These are team members that are frequently sought after for advice or who serve as bridges between teams. When using ONA surveys, identifying central team members requires all participants to name the contacts in their network, not excluding the ones who are not actively taking part. In the end both central and non-central team members are made visible. The results, if not communicated and contextualized right, can be rather disturbing to the people involved. Telling someone that they are in the periphery of the network or that they are not influential can trigger heated responses and the feeling of being betrayed by the data.

Borowska: This sounds dangerous for an employee’s confidence in their workplace. What about others involved in the use of network analysis?

L&A: It is not only employees that are confused by network analysis results. Data scientists, consultants and managers using the data tend to struggle to find meaning in the patterns they see. Network measures are neutral. Things like density of connections or centrality of individuals lack a clear benchmark. High density is not necessarily bad for a team and being central is not necessarily good. But there are all sorts of platforms and tools out there for organizations to buy who claim that these things exist. That, of course, is what makes them easy to market.

To add to this confusion, a crucial aspect of relationships and networks is individual experience of networks: you cannot understand networks without asking people how they perceive and feel about them. This cannot be captured by digital network traces. Why does it matter? The mental monitoring of social relations, the perception and mental processing of social information, is a vital task for individuals. Not only does it determine our participation in teamwork, but it also impacts our success as individuals as we endeavor to get along and get ahead in organizations. Yet research has shown that these mental models tend to majorly diverge from what actually is going on. Humans are known to have major deficits when it comes to accurately perceiving and recalling social relations around them.  For instance, if someone perceives that a high-performing individual in the organization is closely connected to someone else, she/he would assume the two are high performers regardless of whether the connection even exists or whether the second person is really a high performer.  These mental models affect how people relate to each other and navigate their networks.  If it is perceptions, rather than realities predicting how we feel, think and act, the question remains whether digital traces are at all useful to understand employees.

Finally, even if data is properly collected and understood, putting it to use may be unexpectedly difficult. Social relationships and feelings, cannot be easily engineered. Directly rewiring employees as nodes in a computer network is not possible as things like trust, influence or relationship quality cannot be created by decree.

Borowska: Does mean that we should abandon network analytics?

L&A: Working with data about workplace collaboration is full of pitfalls at every step. Attributes like trust or relationship quality cannot be easily measured. If they are measured, with sensors for example, employees privacy may be at stake. Even if privacy is protected, employees may not welcome the insights and feel threatened by them.

As a manager or data scientist, you might not really understand the story the data tells, and if you do, you might not be able to use insights in an actionable way.

Andrés Cardona

Should we forget about Network Analytics altogether? According to Laura and Andrés, no, not quite yet, but we need to try harder. The value certainly does not lie in mechanically extracting insights from large amounts of data.  (Big) Data alone will not magically solve organizational problems. In the case of ONA it actually might create quite a few problems of its own. What is needed is a participative bottom-up approach that treats employees not as objects of study, but as primary clients of data.

It is a topic I will revisit soon to consider how AI should be applied to network analytics and some key pointers on how to avoid the misuse of this potentially useful method.

Follow me on LinkedInCheck out my website