Data analytics is all the rage. That comment is not to suggest that it is not critical for those who develop and execute workforce management strategies to move aggressively into analytics. In September, 2015 I did a LinkedIn post entitled “Evidence-Based Human Resource Management” and it went viral, with a few thousand hits. I realized the topic was not viewed as esoteric and that there were a lot of people that thought it to be important. Since then several of my posts have focused on some of the tools that are available to inform decision-makers when developing workforce management strategies.
Data analytics can find patterns in data, but only can deal with what is/has been, since there is as of yet no data on the future. Regression analysis can project trend lines into the future but for that to be useful the future must be like the present and the past. Machine learning and AI can make predictions based on models that are created, but they too are based on what is/has been. Again, this is not to minimize the usefulness of using data and the tools to analyze it.
Innovation is valued highly in the dynamic, competitive world of today. Most organizations need to leap ahead of others to gain competitive advantage. Creating a product that has never been or features that have not been available relies on innovation and creative design. And at best analyzing present/past data can provide some insight into what has worked in similar situations.
Another form of evidence is the knowledge possessed by those responsible for creation. The brain is capable of synthesizing past experiences into something that has not been, which suggests that using knowledge to augment data can contribute to projection into a unique future. The Toyota production system utilizes the “5 Whys” approach to explaining what has happened. So if an auto production line is producing a particular type of defect (easily identified using real time data) one can look at correlations and regression analysis to identify potential causes. If one factor seems to be perfectly correlated with the defect it is reasonable to suspect causation. Yet correlation does not establish causation. Even though A is correlated with B it may be that both are determined by a yet undiscovered C. Asking a subsequent “why” can divulge the cause of the cause and if that driving factor is not dealt with the problem may not have been resolved. Doing causal path explorations can be the job of a software model. But identifying the full range of possible contributors to the result takes human knowledge and intuition.
Software systems are available that promise to predict the probability of someone with critical talent leaving the organization. And they can be very helpful. But if a key person leaves because of (at least perceived) mistreatment by a manager and that person to that point had no intention of leaving these models are going to be inadequate. A resourceful analyst can begin a search for unique contributors that had not been present when the predictive model was used. And it takes experience with similar occurrences to know that people exit the rational world if a perceived wrong elevates their emotions beyond tolerable levels. To really explain the loss someone has to go the extra mile to do a wide exploration for contributing factors that were not build into the “likely to leave” model.
Organizations use employee surveys to identify dangerous conditions. A commonly asked question deals with employee satisfaction with their pay. If the survey shows that the vast majority of the population is somewhat dissatisfied with their pay it is reasonable to consider raising pay. But if care is not taken to determine whether this is normal the pay adjustments may do no good. If the analyst makes the effort to search for the normative response it may be discovered that this is exactly the result that could have been predicted (most providers of survey systems have databases with which to compare). Neuroscience and Behavioral Economics research into cognitive biases has shown that we all think we are better than we are, which will certainly impact how we feel about how we should be rewarded. That perception may be incorrect but a person’s perception is their reality. If the median performer in a group believes (s)he is in the 75th or 80th percentile (supported by research) there is apt to be dissatisfaction with the relationship of one’s pay increase to those received by peers. So some of the angst over pay rates may be something an organization has to live with.
Another reason a pay adjustment will not resolve the issue could be that someone feels the pay rate is too low compared to what that person should be paid for what (s)he could do, rather than to low for what is being done. If an analyst goes to the next level to identify that this is an under-utilization problem rather than a pay problem there will be a realization that a pay action will not address the issue. Of course the employee may have an inflated sense of capability but again, perception is reality.
So determining causes within complex contexts may require a Sherlock Holmes guiding the investigation. That detective would be wise to use analytics along the way to identify the cause(s) and perhaps the antidote. But when it is suggested that algorithms will replace people
I respond by suggesting that algorithms, AI, machine learning and big data are ideal for extending people’s reach, not replacing human judgment. It may be that with proper programming these systems could do the whole job but it is going to take skilled, knowledgeable people with extensive experience to guide that programming. People should be increasingly acting as Systems Analysts who turn over the programming to machine learning processes. And by the time we get smart enough to do all that we and probably our children are going to have plenty to keep us busy. And our grandchildren are probably going to be safe too… given the rate of change that is unlikely to slow down there will be work for them.
Robert J. Greene | Reward $ystems, Inc. 