ORIGINAL ARTICLE PUBLISHED BY · · UPDATED
Hype and expectations for AI are now sky high. We will soon be talking to our computers; drones will deliver our groceries; our cars will drive themselves; and most white collar workers will be monitoring machines. Is all this true and will this really happen?
As an industry analyst and engineer who has followed technology for decades, I’d say we are in interesting phase: on one hand the hype is far ahead of the reality; on the other, the upside could be much bigger than we think. And in HR the opportunity for value is massive.
While nearly every HR vendor is now building an AI team and we all want our systems to be more intelligent and useful, I believe this market is still very young, so I want to point out some things to consider.
Last week I attended a conference on recruitment automation and we listened to Billy Beane, General Manager of the Oakland A’s, talk about Moneyball. After a wonderful discussion of the history of Sabermetrics and how data has changed the game of baseball, he told us that he now has six PhD machine learning engineers on his roster, and “the team with the most PhD’s is really hard to beat.” This is what is happening in business.
Let’s recognize that AI is not some magical computerized persona; it is a wide range of algorithms and machine learning tools that can rapidly injest data, identify patterns, and optimize and predict trends. The systems can understand speech, identify photos, and use pattern matching to pick up signals about mood, honesty, and even personality. These algorithms are not “intuitive” like human beings, but they are fast, so they can analyze millions of pieces of information in seconds and quickly correlate them against patterns.
Statistically AI systems can “predict” and “learn,” by plotting curves of possible outcomes and then optimizing decisions based on many criteria. So you could imagine an AI system that looks at ll the possible demographics, job history, and interview questions with a candidate and then “predicts” how well they will perform on the job. (HiredScore, Pymetrics, HireVue, IBM, and others are working on this.)
While this is more complicated than it sounds, it’s an important and noble effort. When I was asked about this a few weeks ago I answered: “most management decisions we make today are done by the seat of our pants. If these systems make us a little smarter we can possibly improve our operations tremendously.”
Yes there are many risks and obstacles to deal with, but the potential is very big.
Let me list just a few of the many areas we could see breakthrough results.
In recruiting, we make many decisions on gut feel. One study showed that most hiring managers make a decision on a candidate within the first 60 seconds of meeting a candidate, often based on look, handshake, attire, or speech. Does we really know what characteristics, experiences, education, and personality traits guarantee success in a given role? No we don’t. Managers and HR professionals use billions of dollars of assessment, tests, simulations, and games to hire people – yet many tell me they still get 30-40% of their candidates wrong.
Algorithms based on AI can weed through resumes, find good internal candidates, profile high performers, and even decode video interviews and give us signals about who is likely to succeed. One of our clients now uses Pymetrics’ AI-based gamified assessment to screen candidates for its marketing and sales roles and their success rate has gone up by over 30%, while eliminating all the “interview bias” and “educational pedigree bias” inherent in the current process. AI in recruitment will be huge.
By the way, while we are all worried about job skills (software skills, sales skills, math skills, etc.) most research shows that technical skills make up a small part of a person’s success. In our most recent research on High-Impact Talent Acquisition, we found that the level 4 maturity companies, those with the highest financial return from hiring, allocate almost 40% of their hiring criteria to emotional and psychological traits like ambition, learning agility, passion, and sense of purpose. Will AI uncover this too? Perhaps.
(Vendors in this market include LinkedIn, Pymetrics, Entelo, HiredScore, IBM, Textio, Talview, Unitive, PredictiveHire, and more.)
In employee development and learning, we really don’t know how to “train” people perfectly. The global L&D industry is over $200 billion and most learning professionals tell us that at least half this is wasted (forgotten, inappropriately applied, or just wasting peoples’ time). But we don’t know which half this is!
Do you as an individual know what you “need to learn” to be better at your job? We all have a pretty good idea, but what if we had algorithms that monitored and studied the skills, behaviors, and activities of the highest performers in our teams and then just told us how to be more like them? These kinds of “Netflix-like” algorithms are now entering the world of learning platforms, making learning as useful and fun as watching cable TV. Again the market is young, but the opportunity is massive. Our research shows that the average employee has less than 25 minutes a week to train and learn; if we make that time more relevant everyone will perform better.
(Vendors in this market include Degreed, EdCast, Filtered, Volley, Axonify, BetterUp, Clustree, Workday, and more.)
In management and leadership, we operate like Zen masters. We read books, we go to workshops, we copy the bosses we admire, and we glorify the successful leaders of the day. Do we really know the science of leadership? I’d suggest it’s a fleeting topic. This year we are focused on purpose, mission, and followership. Only a few years ago it was “servant leadership” and when I was young it was “execution and financial acumen.” Most studies find that there are dozens of management and leadership traits that define success, and each of us brings a slightly different and unique combination of them.
AI can now help decode this. I know of three vendors who have built “AI-based” coaching tools, systems that request feedback, read comments, and intuit sentiment from employees and teams. They use this data to match these individual and teams’ issues against higher performing teams, and the use that data to give managers and supervisors “nudges” on how to do better. One client told me that within only 3 months of using this tool their leadership teams showed a 25% improvement in corporate values just based on small behavioral nudges.
(Vendors in this space include Reflektiv, BetterWorks, Ultimate Software, Zugata, Humanyze, ADP, Impraise, and more.)
In fraud and compliance, the opportunities are massive. One study found that employees who steal or commit crimes are “contagious” to their peers (people who work with them pick up bad habits). AI can look at organizational network data (email traffic, sentiment of comments) and identify areas of stress, areas of possibly ethic lapses, and many other forms of compliance risk, and the point out the “red areas” to HR or compliance officers so they can intervene before bad behavior occurs.
(Vendors in this space include TrustSphere, Keencorp, Volley, Cornerstone, and more.)
In well-being and employee engagement, AI is now being used to identify behaviors that cause poor work performance. In safety AI can identify behaviors and experiences that lead to accidents. A new breed of survey tools can identify patterns of stress and bad behavior and alert HR or line managers.
(Vendors in this space include Limeaid, VirginPulse, Glint, Ultimate Software, CultureAmp, TinyPulse, Peakon, and more.)
In employee self-service and candidate management, a new breed of intelligent chatbots can make interactions intelligent and easy.
(Vendors in this space include IBM, ServiceNow, Xor, Mya, Ideal, Paradox, and more.)
The list goes on and on.
All these applications are new, and as exciting as they seem, there are plenty of risks to worry about. The biggest is that AI cannot work without “Training data.” In other words, the algorithms learn from the past. If your current management practices are biased, discriminatory, punative, or overly hierarchical, you may just wind up institutionalizing all the things you hate. We need AI that is transparent and “tuneable” so we can inspect the algorthms to make sure they’re doing the right things. Just like the early automobiles didn’t always drive straight, our early algorithms are going to need “bumpers” and “tuning knobs” so we learn how to make them more accurate.
The systems can institutionalize bias. Suppose your company has never hired women in engineering and has very few African American engineers. The AI recruitment system would naturally conclude that women and black engineers are less likely to move into management. This type of bias has to be carefully removed from the algorithms, and it will take time to do this well.
There’s a risk of data exposure and inadvertent misuse as well. Consider a common use of analytics where we try to predict the likelihood of a high-performer leaving the company. If we tell managers “this person has a highly likelihood of leaving” we may in fact create the wrong behavior – the manager may ignore this person, or treat him or her differently. We have to learn how to apply behavioral economics carefully, so we don’t let AI turn into “HAL” (the movie 2000) by accident. AI is a “tool” for suggestion and improvement – not an independent decisonmaking system today.
I spoke with one of the AI leaders at Entelo this week and we discussed the need to create “interpretive” and “transparent” AI systems. In other words, whenever the system makes a decision, it should tell us why it made this decision, so we as humans can decide if the criteria it used are still accurate. He told me this is one of their most important criteria for new tools, and unfortunately today most AI systems are a complete black box.
Consider what happens when an autonomous vehicle has a crash. We spend a lot of time diagnosing how it happened, what visual or algorithmic systems failed, and what conditions could have led to the accident. What if AI makes the wrong recommdnation on a candidate, or a salary adjustment, or a management intervention? Will we find out? Will we diagnose it? Will we even notice until it’s too late? We have lots of work yet to do to instrument and learn how to “train” our management-base AI systems to work well.
Right now the hype around AI is at an all-time high. Every HR software vendor wants you to believe they have a machine learning team and a best-of-breed AI solution. Yes, these capabilities are immensely important to this industry, but don’t believe the hype.
The success of an HR tool will be dependent on many things: the accuracy and completeness of its algorithms, the ease of use of its systems, but more important than all its ability to provide what is called “narrow AI” – or very specific solutions that solve your problems. This can only be done when the vendor has massive amounts of data (to train its system) and they gain lots of feedback on how well it works. So I believe the barriers to entry are going to be focus, business strategy, and client intimacy, not just having great engineers.
And don’t buy a system that’s a black box unless you can really prove it in your company. Every company’s management and people decisions are often culture based, so we’ll have to take time to try these systems in the real world and tune them for best use. IBM, for example, has spent years optimizing its AI-based compensation and career solutions for its company, culture, and business model. They are now bringing these tools to corporate clients and finding that each implementation teaches IBM new things about the algorithms to make them better for that industry, culture, or organizational need.
Despite these challenges and risks, the upside is enormous. Companies spend 40-60% of their revenue on payroll and much of this enormous expense is driven by management decisions we make on gut feel. As AI systems in HR get smarter, more proven, and more focused on specific problems, I believe we will see dramatic improvements in productivity, performance, and employee wellbeing. We just have to be patient, vigilant, and willing to invest.
The original article was published by https://joshbersin.com/2018/06/ai-in-hr-a-real-killer-app/
Do you want to read similar articles? Click Here!
Albert brings global market research experience to help executives harness intelligent technologies, reinvent HR, re-skilling and employee experience across NAR, APAC, LATAM and EMEA regions.
Hello, and welcome back to my Blog! More than ever, the sense of belonging, skill development, role modeling and aligning mindsets and behaviors are critical
The Challenge: Global sport management client was going through a business transformation to streamline operations and reduce cost. As a result, the company decided to
The Challenge: Global chemical client was in the process to spin off and create three brand new companies to increase market focus and business agility.
Hello, and welcome back to my Blog! A new human-machine collaboration is shaping the future of human capital and HR. The last few months we
The Challenge: A global insurance organization with over 100,000 employees is undergoing a significant M&A transformation aimed at streamlining operations and expanding market reach. A
The Challenge: A global retail client with over 30,000 employees was expanding operations and opening new stores, requiring the hiring and onboarding of talent at
The Challenge: A global construction engineering organization with over 10,000 employees was experiencing exponential growth and preparing to enter international markets. However, the client faced
Hello, and welcome back to my Blog! This is a question that I get all the time and thought I would share a perspective based
The Challenge: A Ieading banking organization, with over 200,000 employees, embarked on a strategic investment in talent to address new business requirements and expand to
