How Artificial Intelligence Reduces Hiring Mistakes

In this two-part article series I’ll discuss a leading-edge hiring tool that can have a major impact on hiring: artificial intelligence. In Part One, I’ll cover the basics. In Part Two, I’ll discuss how AI was used to solve a client problem. What are artificial intelligence (AI) programs? Artificial intelligence (AI) programs are special types of computer programs that mimic the function of the human brain. Without going into too many technical details, AI excels at finding patterns among information. How are AI programs now used? AI is a relatively new field and its uses are still being discovered. At present, AI applications include predicting stock market movements, identifying credit card fraud, evaluating loan applications, planning airline schedules, diagnosing cancer, day trading, sales forecasting, financial forecasting, product design, fault tracing, and quality control, among others. Most AI applications work like this:

  1. AI finds hidden patterns associated with performance.
  2. The patterns are stored.
  3. the stored patterns are used to make predictions.

Why be concerned? Hiring research consistently shows the “resume, interview, and background-check” process is about 50/50 accurate. In other words, once a recruiter has “screened out” blatantly unqualified people, he or she would be just as correct flipping a coin for the rest. This practice has some major consequences:

  • The cost of recruiting, training, and developing employees drains cash reserves.
  • Organizational training programs that were intended to enhance employee skills, not repair hiring mistakes, waste money.
  • A manager’s major responsibility is production, coaching weak employees to perform hampers productivity.
  • Financial differences in personal productivity are astonishingly high. They are estimated to be 19% of average salary for semi-skilled workers, 32% for skilled workers and 48% for managers/professionals. Let’s translate these percentages into dollars. If you have twenty $60,000/year managers, a 48% productivity difference would amount to an estimated $576,000 per year; twenty skilled workers paid an average of $40,000 would be $152,000; and, twenty semi-skilled workers at $50,000 would be $320,000 annually.

Hiring professionals who want to compute their own differences in productivity can use the following chart:

Position Number of People x Average Salary x Productivity Difference = Position Totals
Semi-skilled? ? ? 19% ?
Skilled? ? ? 32% ?
Professional? ? ? 48% ?
Managerial? ? ? 48% ?
Overall Total:? ?

How can AI be used to improve hiring? To answer that question, we need to break down hiring into two parts: 1) applicant information, 2) performance data. For example: Part 1. Applicant Information Organizations typically use a wide range of hiring tools to decide whether an applicant is qualified for the job, such as:

Article Continues Below
  • Application forms
  • Resumes
  • Personal interests
  • Amount of education
  • Job experience
  • Reference checks
  • Conscientious tests
  • Biographical data
  • Assessment centers
  • Traditional interviews (unstructured)
  • Integrity tests
  • Job tryouts
  • Prior job knowledge
  • Structured interviews (behavioral or situational)
  • Mental ability tests
  • Work samples

Part 2. Performance Data Organizations usually have a wide range of performance measures that indicate how well their current employees are doing, like:

  • Personal productivity
  • Sales
  • Turnover
  • Job accuracy
  • Teamwork
  • Integrity
  • Customer service
  • Problem solving
  • Project planning

Our objective using AI is discovering patterns in the hiring data that predict turnover. We begin by collecting actual turnover data from current employees. That becomes our “target.” Then, we collect significant pre-hire information gathered from interviews, application forms, tests, etc. We examine all this data using some AI algorithms. When we find the pattern associated with employee turnover, we store it. Whenever a new candidate applies for a job, we use the stored pattern to predict the probability the applicant will quit prematurely. We could do the same with any other measures of performance such as sales, cold calling, job accuracy, personal productivity, integrity, customer service, problem solving, project planning, or any other measure an organization considered important. What if there are no patterns? If there were no patterns associated with success or failure, then we would quickly learn that hiring data had nothing to do with job performance. This would start an investigation of other factors that might affect performance such as economic conditions, management practices, or the quality of data used in the analysis. We might even want to replace low-accuracy hiring tools with high-accuracy ones. What does an AI prediction look like? AI predicts the probability an applicant will fall into one of several groups. If, for example, we built an AI algorithm that predicted “High,” “Medium,” or “Low” performance, AI would predict an applicant’s “probability” as something like “10% High, 35% Medium, 55% Low”. Should AI be applied at the “job level” or the “organization level”? You conduct a “job-level” study when you want to predict job-specific performance such as sales or problem-solving ability. You analyze data at the organization-level if you are looking for generic performance factors like turnover, teamwork or customer service attitudes. Can AI be used with resumes? You can use AI with any kind of applicant data; however, AI predictions are only as accurate as the data they use. The universal rule of “garbage in, garbage out” applies to any analytical technique. Resumes are among the best examples of mixed data ? some good and some garbage. Resumes, for example, often include verifiable names, dates, and duration of employment and education received; however, they seldom contain reliable data about an applicant’s ability to do a specific job. Using all the information from a resume would mix “garbage” data with “good” data ? something that should be avoided. The more accurate and more predictive the hiring tool, the better AI will be at predicting performance. Benefits of Using AI as a Hiring Tool There are several major benefits from using AI in the hiring process. It allows you to:

  • Reduce long application forms and processes to a few critical questions
  • Use a whole-person approach to predict performance
  • Identify both effective and ineffective hiring tools
  • Base predictions on current employee performance
  • Conforms with the intent of the “Uniform Guidelines” for conducting validity studies
  • Reduce hiring mistakes
  • Reduce hiring manager guesswork
  • Give important information more weight in the hiring decision
  • Improve overall individual productivity
  • Reduce training time and training expense
  • Reduce manager “coaching for improvement

In Part Two, we’ll discuss in detail an example of how AI was used to improve hiring quality in an organization that supplies Medical Technician services.

Topics

4 Comments on “How Artificial Intelligence Reduces Hiring Mistakes

  1. I’m familiar with artificial intelligence but could not determine from this article exactly how AI is being used. There are many applications touting the banner “AI” ranging from statistical classifiers to expert systems to natural language processors. I’m curious as to why the author doesn’t specify the general domain of AI deployed and why that technology in particular is suited for behavior prediction.

    You can read the original article at:
    http://www.erexchange.com/a/d.asp?cid=023358EE9254424790BA61022DF7AF3D

    Post your own Article Review
    http://www.erexchange.com/p/g.asp?d=M&cid=023358EE9254424790BA61022DF7AF3D

  2. Good question, David…I did not go into details because that the intent of the article was to raise awareness of the application, not explain details of performing this type of analysis.
    Since you asked, the article refers to using group classification techniques that are based on output from cascading probablistic networks. The intent of this analysis would be to find a pattern among the independent variables which could be used to predict group membership (e.g., turnover). A second phase of the analysis would employ genetic AI algorithms to determine the loading characteristics of each independent variable, then use this data to build a “smart” application blank that would only include highly predictive items. Free free to email me if you have any further questions.

    Wendell

    You can read the original article at:
    http://www.erexchange.com/a/d.asp?cid=023358EE9254424790BA61022DF7AF3D

    Post your own Article Review
    http://www.erexchange.com/p/g.asp?d=M&cid=023358EE9254424790BA61022DF7AF3D

  3. Having read earlier articles by Dr Williams and considering myself a ‘fan’ I was particularly interested in this article as it related to an area of IT in which I worked prior to moving into recruitment.

    At that time I was worked on neural networks in areas of medical diagnosis and IT fault diagnosis.

    I believe expert systems have a lot to offer, and I’m sure the technology has made great leaps in the last ten years or so. I share the concerns raised by David. His question also heightens my interest in waiting for part two when I hope the picture will be complete.

    Dr Williams could be introducing us to a major step forward in recruiting technology. Hurry on part two please.

    You can read the original article at:
    http://www.erexchange.com/a/d.asp?cid=023358EE9254424790BA61022DF7AF3D

    Post your own Article Review
    http://www.erexchange.com/p/g.asp?d=M&cid=023358EE9254424790BA61022DF7AF3D

  4. Ok. Ok. ..You asked. There are many benefits to using analytical tools in hiring applications. For one thing, analysis tells us if hiring data has any relationship with performance (something recruiters and hiring mangers would like to know). It also allows us to eliminate the extraneous nonsense like silly tests, foolish questions and lengthy application blanks that creep into hiring.

    The goal of analytical tools is to build a scientifically-based “hiring formula” that will make better hiring decisions than people can. Easier said than done.

    Building this fomula requires two things: First, we must have some kind of “solid” dependent variable like turnover, sales, new accounts, mistakes, etc. We also have to be careful about using supervisory ratings because they tend to be based more on global personality than hard skills.

    Once we have something to predict, we need to look at all the independent data gathered during the hiring phase. This might include number of jobs, titles, years experience, education, test scores, etc. The secret here is to use data that is verifiable and trustworthy. This is the whole idea of building a predictive model – trustworthy hiring data at the “front” and measurable performance and the “back”.

    Now that we have a trustworthy list of hiring predictors and job performance, it is time to assemble the analytical tools. Traditional statistics makes some serious assumptions about data that are often violated, such as data being normally distributed, variances are equal, relationships are linear, and the data are continuous. (This probably sounds like technobabble, but, remember, you asked). The primary purpose of traditional statistics is to find and explain the strength of relationships.

    For example, we know there is a very strong linear relationship between mental ability, conscientiousness, extraversion, mental stability and performance in many kinds of jobs. We know this because statistical analyses have consistently shown the presence of strong relationships that are unikely to occur by chance. Traditonal stats are primarily a research tool.

    AI tools, on the other hand, are primarily for the practitioner (not that they are any less complicated that traditional stats). AI does not try to explain the strength of relationships. It just tries to find useful patterns. It is non-linear, does not assume normality (especially from analysts who tend to be slightly weird), nor is it concerned with equal variance. It works especially well with binary categorical data (sorry, again about the technobabble).

    For example, AI has no trouble mixing and matching “years of experience” with “schools attended” and “awards achieved”, then using this data to predict “turnover” (assuming, of course, that there really is a relationship with turnover).

    AI is like your brain without the weekend hangover. Hiring managers and recruiters who say they “know-em when the see-em”, are using their own form of AI to “identify” future “winners”. The only differnce between “wetware” patterns and AI patterns is that people make decisions based on emotions and AI makes decisions based on trustworthy historical data.

    Some applications of AI include building a “smart” application form that is fast and easy to complete; taking a collection of mixed and matched data to make a holistically weighted prediction about future turnover, sales success, or job potential; help eliminate worthless data (i.e., EEOC bait)collected during the hiring process; and, best of all it can minimize hiring mistakes by developing a “formula for success”.

    If there are any readers who are frustrated by their hiring results (or if they are an AI skeptic) AND have a large employee database in a spreadsheet format ready to analyze, email me at rww@ScientificSelection.com and I’ll do the analysis for free!

    You can read the original article at:
    http://www.erexchange.com/a/d.asp?cid=023358EE9254424790BA61022DF7AF3D

    Post your own Article Review
    http://www.erexchange.com/p/g.asp?d=M&cid=023358EE9254424790BA61022DF7AF3D

Leave a Comment

Your email address will not be published. Required fields are marked *