Artificial Intelligence: Making Hiring Tools Smarter, or Not

You have probably seen film trailers promoting the movie “AI,” if not the movie itself by now. AI stands for “artificial intelligence.” The movie is set in the future and involves robots powered by computer programs using “artificial intelligence” software. Well, you might not know it, but artificial intelligence is not science fiction. We have been using AI in our products for years. AI works like your brain by finding hidden patterns among data, and it is already hard at work diagnosing cancers, planning airline schedules, trading stocks, assessing credit strength, and determining quickly whether someone stole your credit card, just to name a few of its applications. Of course, the big question is, when will you harness the power of AI in your organization? What’s the benefit, you say? Well for starters, consider what it would it be worth if you could…

Predict performance ratings by analyzing attitudes, interests and motivations at the hiring phase

Improve sales productivity by understanding the relationships between prospecting, customer retention, economic trends, and the Fed’s discount rate
Increase retention in a high turnover environment by determining critical patterns in application form data
Develop a weighted-data application blank for different jobs that accurately identifies highly productive applicants
Streamline your website recruiting application to minimize qualified applicant “fall out” and increase applicant quality
Develop a virtual “model” of your organization for managers to develop their decision making skills
Reduce customer turnover by detecting early trends and acting quickly to correct them
Predict staffing needs based on economic markers, marketing and advertising efforts

Which application is more important to you? I guess it depends on where you feel the most “pain.” If you have a simple job (retail, warehousing, fast food, etc.) you might want to target factors associated with short-term voluntary turnover. If your pain centers on selling, you might want to target cold calling, prospecting or sales expansion. If your pain centers on customer retention, you might want to target customer satisfaction, purchasing patterns, retention, etc. And if you are one of those lucky employers who have people queuing up at the door to work for your organization, you can use AI to reduce application forms to a few critical “predictor” questions and cut out irrelevant items. How Do You Get Started? Using AI effectively requires both careful preparation and knowing what you are doing. If you are working on solving an employee problem, for example, it does not mean studying a job description. That’s silly. It means getting in the trenches and learning from people who do the job every day. Current employees are a treasure-trove of overlooked information. For example, let’s say you want to predict turnover. Obviously, knowing this kind of information could save a lot of time and hiring effort. Furthermore, you probably have a few years’ worth of turnover numbers and some old application forms. That’s a start. How do you get AI to work its magic? First, you carefully choose the most likely data associated with turnover, being careful to leave out “garbage” (i.e. shoe size, height, mother’s maiden name, eyeball distance, etc). Then you gather a few hundred test cases that contain both predictive factors and turnover data. Plug all this data into an AI program and examine the contribution of each predictor to turnover, select the best predictor, calculate a turnover “guesstimate,” and examine how close your guess was to the real turnover value. Then, add another “layer” of non-linear calculations to the formula and repeat the process. Continue the process until the difference between calculated values and actual values is as small as possible. (If you are still awake, you might suspect that I left out a “few” technical details here). Anyway, when AI finds a pattern of factors that accurately predict turnover, you save the formula and add it to your arsenal of hiring tools. When a new applicant knocks on the door, you feed their individual data into your AI formula and see if they become an instant winner! Easy as pie? Not quite. Garbage In, Garbage Out If you don’t know what you are doing, it is easy to misuse this powerful tool. For example, if you feed AI with garbage data or run the analysis too many times, you tend to end up with teensy-weensy patterns that only occur by chance. Basically, you might think you have a valid and reliable pattern, but it would only be right occasionally. Too bad, it would have been easier to flip a coin. A company I learned about a few days ago illustrates the pitfall of using bad data. They use artificial intelligence algorithms to scan resumes. They seemed quite proud of their application – 200,000 resumes in the database they claimed and growing! Too bad for them and their unfortunate users, they might have mastered using AI programs, but they completely missed the mark when it came to good hiring practices. Assuming the ultimate goal of a recruiter is to hire only job-qualified people, it doesn’t take a math whiz to realize this can only be done using a clear job-target and a solid set of measurement tools – and resumes fall short of this goal. In fact, anyone who has ever reviewed or scanned a resume will know that:

Resumes are part fact and part creative writing
People with the best-looking resumes often have the poorest skills
People with poor resumes can have great job skills
You can often outsmart a scanner by seeding a resume with a few key words and phrases
Scanners cannot differentiate between “Thought about going to Harvard 1990 to 1994,” and “Attended Harvard 1990 to 1994”
Smart applicants know how to “write to the job” independent of actual skills
Resumes are often poor indicators of job skills – a few percent at best
Applicants “skip over” undesirable job data in their resume
Aside from technical exposure, resume data often has nothing to do with job requirements
Two people scanning the same resume can arrive at totally different conclusions
You don’t need fancy mathematics to identify job gaps, lack of experience or job hopping

Got the picture? Here is an organization attempting to make valid predictions based on data that is part fact and part fiction. Now assume you want to automate this mess. No matter how sophisticated your analytical engine, the best you have to work with is an inseparable mix of facts and garbage. AI doesn’t care whether you feed it good data or junk food – it will find a pattern just the same. Scan resumes using more powerful analytical techniques? Based on what? Finding better patterns among the trash? Mental Health Week Insanity is often defined as repeating the same behavior over and over again with the same result. If that is true, what assumption can you make about organizations that do not learn from their mistakes and correct their behavior accordingly? Easier said than done, you say? Your organization is organized into non-communicating silos? Your departments seldom get feedback on what works and what does not? Your budget is often based more on maintaining current processes than improving systems? Your people want others to change, but resist change themselves? Change It! Find the hidden patterns in all this mess and use AI techniques to help break out of the insanity cycle by putting that knowledge to productive use. This will make a significant difference between future excellence and current mediocrity. <*SPONSORMESSAGE*>