I have written several blog posts on how Bayes’ Rule can help you make better business decisions and application of this theorem. One of the areas where Bayes’ Rule is most often neglected is in hiring decisions. Often, rational and data driven individuals and organizations abandon the rules of optimal decision-making and rely on intuition.
At its core, Bayes’ Rule shows how you can optimize the chance of a correct decision by looking at previous data points that encompass the decision you are trying to make. In the case of hiring, this analysis would be more effective by looking at the metrics and data that shows who succeeds, looking at what makes someone successful in the position you are hiring for and reducing the impact of data that does not lead to good hiring decisions.
What most companies end up doing is using data as a filter but then hiring based on intuition. If you really want to make good decisions, you need to understand your intuition is only one (weak) data point and base the decision on Bayes’ Theorem, using past data to make the optimal decision.
What has worked for others
First, look at the position you are hiring for and identify the most successful people (at other companies or at your’s) in the field and “reverse engineer” their background. What experience(s) did they have before they were hired? What is their educational background (school, degree, extra curricular activities, etc.)? Using Bayes’ Rule, if you are hiring for a Director of Social Media and find that 90 percent of the top performing Directors of Social Media went to Texas A&M, then the chances of making a good hire from Texas Tech is already at less than 10 percent.
With social media, it is easy now to deconstruct the skill sets and background of the best people in the industry. It is also much easier to identify these people. If company Y has a great social media program, you just do an advanced search on LinkedIn for company Y and keyword social media. You then look for common skills, education and experience from the industry rock stars.
What has worked for you
The second key is determining what the successful employees at your company have in common. If 80 percent came from the telecommunications industry, then you optimize your chances of a good hire by hiring again from that industry. Stack rank your employees and determine what is common among those on top. Also, look at the employees on the bottom of the list and see what they have in common. If 85 percent of the worst performers lived with their parents before you hired them, avoid hiring those who are currently living with their parents..
How the individual has performed
The second crucial piece of data is the person’s previous performance. If you are looking to someone to help you sell your company, look at their results with their past employers and how many of those companies have had sales. What is crucial is to:
- Ensure these achievements are consistent with what you have identified from the analysis described above as to what achievements are likely to lead to success in the position you are hiring for. If you run a soccer (football for my non-US readers) and are hiring a mid-fielder, you may find that speed in the 100-meter is a key variable for success. If the candidate has a great time in the 100, that probably suggests they have a good chance to succeed on your club. Conversely, if they highlight their ability to run a 5K marathon, although quite an achievement it does not provide any indication about their fitness for your position.
- Make sure the achievement is real. People are adept at spinning lack of success into achievements. Drill down into whether what they are describing actually is an indicator of success in the position you are hiring for. If it is a growth position and they talk about 500% growth, that growth may be non-monetizing users. If the data shows that successful people for the position grew monetizing users, then this data does not indicate any increased likeliness for success. Moreover, ensure the data they are providing is accurate. Rather than just spinning the numbers to make them look good, some candidates will “create” achievements they did not actually have. Obviously, such non-achievements do not indicate an increased likelihood of success, to say the least.
The key is to measure the prior performance of the candidate and use that data to predict their likelihood for success in the position you are hiring for.
One common mistake to avoid when building a data-driven approach to hiring is basing criteria on outliers. You may have noticed that the company with the best customer service experience has its CS team run by a 6-foot-7-inch Australian with no CS experience. Some would take this data point to imply they should recruit a tall inexperienced Australian to run their customer service department. What actually occurred was there may have only been a five percent chance for the tall Australian to succeed. Looking at the data, you see that 90 percent of the time people successful in this role have spent ten years in customer service and speak eight languages. If you base your search on the outlier, the tall Australian, you still only have a five percent chance of hiring someone who will be successful. Conversely, if you identify the criteria that leads to a 90 percent chance of success, you increase the chances of a good hire 18X.
A related problem is using an outlier to justify a hire. You may really want to hire someone, you have a personal connection, it would the end the search, etc., but they do not hit the criteria you identified in the previous sections. To rationalize your decision, you find someone like them who has been successful in a similar position. Even with low probability of success, there will be successes (just like there are always mutual funds that outperform the indexes but virtually none that do it over time). The problem is the person you are hiring still has the low chance of success and you are not optimizing the expected performance of your team.
Do not overvalue the interviews
As I said at the beginning, data is usually just the filter and the interview is the decision point. The interview, however, is just one (small) data point and actually provides little evidence whether the person will succeed. Used effectively, you can use an interview to get the data to validate if they have the skills and experience you have identified as critical for success. Unfortunately, most interviews are based on how good a fit a person is (translation, whether they are likable) and not whether they have the skills/education/experience to succeed. This phenomenon is consistent with people over-estimated how well they judge people (probably ninety percent of the people reading this think they are a better judge of character than average, while only 50 percent can be). It reminds me of when George Bush looked into the eyes of Putin and determined that he was somebody he could work with (not to single out President Bush, but 100 percent of politicians also think they are in the top half). Additionally, many job candidates (whether or not they started their career in the KGB) are skilled at doing great during interviews (and may have even took instruction in it), but this skill does not translate to success in the workplace.
Hiring the best
What you are really trying to achieve with all hiring decisions is hiring the person with the highest expected value, which is risk times their expected contribution to your company. There is risk with any hire (no matter how positive you feel about it), but if you use previous data (your data, industry data and the candidates data) you optimize your chance for a good hiring decision.
- Bayes’ Rule shows the most effective way to hire is to base decisions on prior data.
- Look at the skills, education and background of others who are successful in the target position and then match it with candidates.
- Do not over-value the interview. It is just a small piece of data and should be used to validate the criteria you have determined leads to a successful candidate; it should not be the data that drives your decision.