Amazon Exposes Fatal Flaws of Using AI for Hiring

Whenever asked about the latest hiring buzz I always respond by saying that while the buzz is louder than ever, when it comes to hiring stronger people little progress has been made. 

For example, Amazon just reported that its AI engine for hiring discriminated against women. But this is likely just the tip of the iceberg. I suspect the AI algorithms used also discriminated against anyone who could do the work but who don’t pass the less than scientific screening criteria listed on the job description. This includes all diverse and high potential talent who accomplish more with a different skillset. 

One reason for lack of progress on the hiring front is that there’s too much focus on being more efficient filling jobs rather than hiring the strongest and most diverse talent possible. Being more efficient using job boards to find stronger talent is unlikely when the strongest people either aren’t looking or not applying.

I contend the problem is primarily a strategic one driven by this simple premise: 

No matter how efficient, a surplus of talent strategy won’t work when a surplus of talent doesn’t exist. 

This concept is demonstrated in the graphic. It segments the hiring process into the four steps defined below. A hiring process based on a surplus of talent strategy (the top arrow) moves left to right and is focused on attracting large numbers of people and weeding out the weakest. A hiring process based on a scarcity of talent strategy (the bottom arrow) moves right to left. This is a high touch process focused on attracting a small number of highly qualified top tier and diverse talent.

While the major segments are the same, the order is reversed, and this is the key difference in the surplus and scarcity of talent models. 

The Major Steps in Most Hiring Processes

HAVE: This is what’s written on the job posting and what a person must have on his/her resume to be considered for the role. The problem is that none of these factors predicts on-the-job performance. 

GET: This is what a candidate will get on the start date (title, compensation, location) and the person must agree to this before knowing much about the job. The problem here is that all of these factors are negotiable if the job offers a better long-term career trajectory.

DO: This is the work the person will actually be doing once on the job. From a practical standpoint if it can be proven that the person can do this work, he/she will HAVE all of the skills and experiences required to be successful. The HAVING is the variable in this case and this is the part missing in most of the AI algorithms for hiring I’ve reviewed. 

BECOME: This represents the career growth opportunity if the person is successful DOING the work. This is one of the primary factors used by top tier candidates to both consider and compare offers. 

The surplus model assumes there are enough talented people who will respond to boring job descriptions that are, at best, ill-defined lateral transfers. Just consider all of the people who get promoted to demonstrate that neither the HAVE nor the GET criteria predicts a person’s on-the-job performance. 

Rethinking the strategy and the underlying process starts by defining the DO criteria as a series of 5-6 KPOs (Key Performance Objectives) to replace the laundry list of “Must-Haves.” This is how you attract people who are motivated to do this work including high potential and diverse talent who can do the work but have a different skillset. In this approach, assessing competency involves having candidates describe their most comparable accomplishments for each of the KPOs. If proven to be competent it’s obvious the person has all of the skills, experiences and credentials needed but likely in a different mix than what was on the original job description. 

Technology can certainly be built to handle this right to left scarcity model but progress has been slow for a variety of reasons. Lack of understanding for one. Stiff resistance to shift to a performance qualified approach for defining work is another. Compliance and legal reasons add to the list of excuses. (This whitepaper from a top labor attorney refutes this.) But perhaps most important of all is that tech vendors have no need to invest in these changes when there’s less risk and more growth opportunity by bundling an old-fashioned job posting process with AI that on the surface appears to be more efficient. 

As I learned long ago strategy drives tactics and process design and without the right strategy nothing will help. This is how you mistake activity for progress.

_____________________________________________________

Lou Adler (@LouA) is the CEO of The Adler Group, a consulting and training firm helping companies implement Performance-based Hiring. He's also a regular columnist for LinkedIn, Inc. Magazine, SHRM and BusinessInsider. His new Performance-based Hiring self-paced learning course - The Hiring Machine - is now available 24/7. His latest book, The Essential Guide for Hiring & Getting Hired (Workbench, 2013) provides hands-on advice for job-seekers, hiring managers and recruiters on how to find the best job and hire the best people.