Compensation Cafe

Algorithms and Formulas Are Us

We are moving rapidly into an increasingly “data-driven” age of Human Resources and Compensation. This data is translated into information. Some of it is quantitative, some of it is visual, all of it is probably biased. Understanding the origin and reasons for potential biases will give the most successful companies an edge that may be impossible to beat.

In the end, HR and Compensation are about winning. Winning the battle for talent. Winning the battle for performance. Winning the battle for consistency. Winning the battle for a pleasant and rewarding place to work. Having an advantage in those battles is what we strive for when we create unique programs, pay differently than our peers and build a culture that inspires. The age of data geeks is upon us and we need to be able to use that to our advantage.

The gold standard of data algorithms may be those used by search engines like Google. Here’s the rub, every piece of data is selected because someone wrote a program that indicated it should be selected. Every data outlier rejected didn’t fit a set of guidelines that was first created by a person. Stanford recently did a study in 2018 that showed a distinct pattern of bias in image-based search results. The results would probably not be categorized as discriminatory, but they also cannot be considered totally accurate.

When the result of an algorithm is a group of photos that appear on your screen, bias may not be that harmful. When an algorithm drives pay decisions or more, we should understand how and why the algorithm works the way it does. A perfect example of this is survey data. The first step in a survey is deciding what data to collect. People make those decisions.

Data that is hard to gather is often made a low priority. For example, we seldom see the details behind incentive plan terms, conditions and features. When data is messy, it may be avoided because the final report is simply too hard. For example, the lack of reward and recognition program details in most pay surveys. Data that is hard to understand may be summarized at a level where it lacks any impact. A good example of this is equity compensation data.

Finally, if the collected data just seems weird, it is often minimized or ignored entirely. Some of the most unique and effective pay programs never seem to appear in survey data, even when we know those companies are participants. We don’t love surprises. We like things to match our expectations. We want smooth transitions. We will modify data, job leveling, structures and more in order to fit the data into our models.

The pitch made by the data pros is that the data will fit an objective model and ignore our biases. But, it’s tough to ignore the biases of the data pros, especially if we don’t know what those biases may be. When data always meets your expectations, it may be due to the fact that your expectations are driving the data, instead of the other way around. In a data-driven age, the art of HR and Compensation is even more important than ever before. Your expertise, insight, and experience may be the secret sauce that gives your company the edge against peers who simply take the data at face value and execute accordingly.

Dan Walter is a CECP and CEP and works as Managing Consultant for FutureSense. He is passionately committed to aligning pay with company strategy and culture. Dan is also a leading expert on equity compensation issues and has written several industry resources including the one-of-a-kind Performance-Based Equity Compensation. He has co-authored ”Everything You Do In Compensation is Communication”, “The Decision Makers Guide to Equity Compensation”, “Equity Alternatives” and other books. Connect with Dan on LinkedIn. Or, follow him on Twitter at @DanFutureSense.

Cafe Classic: All Decisions Are B.A.D. Decisions

Editor's Note: Even in the age of AI, ML and even TMI, SWAGs based on B.A.D. are still necessary.  Jim Brennan shares some Classic pointers!

All decisions are B.A.D. decisions, relying on the Best Available Data known at the time.

Even compensation experts, supposedly the most analytical of HR types, have difficulties with data.  Makes sense to me, because all human decisions are B.A.D. decisions, made according to the “best available data” when action was required.   

When there is no good data, you must rely on the Best Available Data.  The resulting acronym (B.A.D.) is quite meaningful as well as memorable.  The process of making decisions without access to complete and perfect information occurs every day.  The best professional survey analyses involving “counts”, for example, report their reliability statistics.  And guess what they define?  The estimated rate of error contained in their statistical results.  Standard error is a vital measure because it predicts how much of a miss is likely when relying on the normative results derived from a particular observation sample.  It is much like the spread in the shot pattern you get from pellets dispersed by a shotgun; or from sand thrown by hand, where wet sand will be more densely clustered than loose sand. 

Every measurement, even the best, comes with a certain risk of error.  Tight standards of precision offer more hope of being “right” and strenghten the ability to make inferences from studies, but that is not always possible or useful.  The real world is filled with bad information, inadequate facts and missing data points.  And the opposite situation occurs, too, where you are flooded with contradictory information.  In this Internet world, it seems like anyone can find support for almost any contention somewhere.  Too Much Information (TMI) has become a catchphrase acronym. 

Relevance counts, as well.  Despite access to all the information in the world, you can never be sure that one particular answer will prove to be valid in your particular unique situation. 

Sooner or later, you may reach a point where you can only speculate before acting without perfect confidence.  Even in the numbers-oriented world of compensation management, Scientific Wild-Ass Guesses (SWAGs) still are required sometimes.  Operating in circumstances where useful facts may be missing  is a normal part of the total rewards profession.  The more often you enter uncharted waters, the more frequently you will find an absence of robust benchmark studies with relevant useful facts that can guide your choices, suggest options or demonstrate context.  Then, it is time to get B.A.D.!

The best approach when dealing with ambiguity generally is to define your assumptions, explain the risk level and give it your best shot.  For instance, if no one complains over an admittedly arbitrary guess about what you think a specific job should be paid, you are probably offering to pay too high and they don't want you to find out because it makes life easier for them.  When hiring managers grumble but are able to comply easily and operate adequately, then the approximation is probably just about right. 

If any SWAG based on the Best Available Data works out, then your B.A.D. decision will be the best possible outcome.

E. James (Jim) Brennan is an independent compensation advisor with extensive total rewards experience in most industries.  After corporate HR posts and consulting CEO roles, he was Senior Associate of pay surveyor ERI before returning to consulting in 2015. A prolific writer (author of the Performance Management Workbook), speaker and frequent expert witness in reasonable executive compensation court cases, Jim also serves on the Advisory Board of the Compensation and Benefits Review.