Analytics has come to HR. It’s partly Orwellian, tracking what employees do on the computer, and partly warm and fuzzy, leveraging the true informal organizational structure via network analysis (jump into Friday’s Network Analysis course to learn the basics).
One dimension assumes the worst about employees, and gives bosses extra powers to keep tabs on them. Interguard provides software that will alert managers when an employee spends too much time on Facebook, or other “unproductive” sites. It will also monitor employees’ computers for suspicious data exchanges. According to a survey by the American Management Association, 45% of employers use software to track content, keystrokes, and time spent at the keyboard. Much of this is rule-based counting, a machine-based extension of existing manual tracking methods (“timecards on steroids”). More sophisticated analytics is required for some tasks, such as analyzing content (natural language processing) and flagging suspicious data activity (predictive modeling).
Another dimension seeks to harness analytics in a more holistic direction, using network analysis to understand the “true org chart” in an organization. The informal connections and communications among employees that can help (or hinder) an organization. In an earlier blog, I reported on a study of interactions between shifts of nurses, and how different types of networks seemed to be related to patient outcomes. That study dealt with a very small sample of data, by today’s organizational standards. Seeking to extend this analysis to an enterprise scale is a big data task. One vendor, Trustsphere, builds up the informal “org chart” with metadata from employee interactions:
- Time/data stamp
- Interaction type (email, SMS, phone, and even personal meeting)
- Message size
- Message subject (presumably learned/classified via NLP)
Trustsphere then calculates the strength and direction of each employee pair relationship, based on the volume and directional balance of communications, duration of the relationship, lag times, and what it terms “some secret sauce.” From this it builds the network structure of the organization, and allows the organization, among other things, to
- Identify hidden influencers
- Measure the effectiveness of leadership development and diversity programs
- Improves sales efforts
The ROI on these benefits can be hard to measure, though, and Trustsphere also touts the harder edge to its capabilities, such as the ability to “pre-empt risky and/or suspicious behaviour.”
But…McKesson, the pharmaceutical company, worked with Trustsphere to understand better why some sales teams did better than others.
It found that teams that had strong external networks, and weaker networks within the company, sold more. By contrast, the teams that did not do as well tended to have a more diverse set of connections both within the company and externally. The lesson seemed clear – concentrate on interacting with the customer, not with your colleagues.
McKesson has also made news on the front pages – for different reasons. Recently released data troves showed McKesson become the #1 distributor of opiods in the 2006-2012 period, helping to saturate some localities with as much as 300 pills per capita per year. Is there a connection? McKesson’s external-facing orientation, giving free range to sales reps and devaluing a nexis of possibly constraining ties inside the company, probably helped it achieve its dominance in the opiod market. As the Washington Post described the practices of the main opiod sellers,
“In case after case, the companies allowed the drugs to reach the streets of communities large and small, despite persistent red flags that those pills were being sold in apparent violation of federal law and diverted to the black market…”
Perhaps if the high-performing McKesson sales teams had had stronger internal connections within the company, supervision would have been more effective and the sales engine would not have gotten out of control.
In 2017, McKesson paid a $150 million fine for violations of the Controlled Substances Act (CSA), and in May of this year settled an opiod-related lawsuit with West Virginia for $37 million.
Of course, these missteps did not result from faulty analytics. The analytics did their job in maximizing sales; it was management’s single-minded focus on sales, while ignoring correlated risk factors, that led to disaster.
A work in progress…network analysis is an exploratory task
Unlike, say, credit scoring, network analysis is an exploratory task. Hence it is useful for generating ideas and hypotheses, but not a definitive tool for reaching conclusions in a first pass. The network structure of an organization has many different attributes (number and sizes of sub-networks, presence or absence of cliques and singletons, overall density, etc.) that might appear to be associated with a number of different organizational goals and behaviors. This allows scope for discovering patterns that are apparently meaningful, but really the product of chance. In the nurses study, for example, we saw this seemingly puzzling contradiction:
- Strongly connected nursing networks fostered reduced falls
- Just the opposite – the presence of cliques and isolates – seemed to prevent adverse drug events
This might be for real, or it might be a chance product of the many different angles the study looked at. You wouldn’t know more definitively without doing a study to collect data specifically on that question, and even then the implications may remain unclear – you’d need to consider the unintended consequences of fostering one particular type of network structure.