The Role Of Generative AI And Large Language Models in HR – JOSH BERSIN

The Role Of Generative AI And Large Language Models in HR – JOSH BERSIN

Human Resources is one of the most complex, imperfect areas of business. Virtually every decision we make about people (who to hire, who to promote, how much to pay someone, how to develop someone) is based on judgment, experience, personal bias, and some amount of data. And since well over 50% of all corporate spending is on salaries (United States payroll is around $15 Trillion), these “judgmental decisions” cost companies a lot of money.

And in my world, where we deeply study every part of management, leadership, and HR, we often try to correlate various “HR practices” against outcomes to figure out what works. Much of our business is based on this work, and we regularly “re-run” most of our analysis every few years as culture, the labor market, and technology changes.

Right now, for example, we know that workplace stress, pay equity, and career growth are among the most important drivers of employee satisfaction and workforce productivity. Only a few years ago it was all about fancy benefits, bonuses, and grandiose titles.

So what I’m essentially saying is that much of HR is based on organizational psychology, many forms of social science research, and never-ending effort to experiment, learn from others, and figure out what works. And it’s difficult, imperfect, and always subject to debate.

The Underlying Data Set In HR Is Textual

While this massive effort has been going on, most of the “hard science” in HR and management has been focused on numbers. We ask people to take tests, we look at people’s “performance ratings” and grade point averages (which are extremely subjective), and we ask people for surveys, feedback, and lots of data to make decisions. And then we correlate business results (sales, profit, market share) against various people metrics, and think “we have the answer.”

For recruiting and selection we look at experience, job-related tests, and opinions and scores from interviewers. Theoretically if we get enough of this data we can make better and better hiring decisions. And the precise same thing happens when we look at who to promote, who to demote, and who should make it to the very top ranks of the company.

The whole premise of promotion is based on old ideas of “promotability” or “potential” rated against “current job performance” (the 9-box grid). That approach, which sounds quantitative, is filled with bias, so we have to “infer” who has high potential from various assessments, observations, and inputs. Again, when we get lots of data (looking at the background and behaviors of many high performers), we can improve the science of promotion. But for the most part this is based on judgement.

The core “science” of HR is often rooted in Industrial Psychology, which is a fascinating domain which studies attributes, behaviors, and psychology at work. And as much as I admire and follow much of this science, most companies don’t use it very much. There is a billion dollar industry of “validated pre-hire assessments” and they are extremely useful. But for many jobs they are misleading and companies have to validate these tests so they don’t get sued for discrimination.

So if you want to really do a “big data” analysis of your workforce’s skills, experience, and suitability for different work, you’re dealing with mountains of “anecdotal data,” much of which is encoded in biographies, work output, company leadership frameworks, assessments, and lots of communications. And of course there are performance appraisals, business results, and more.

Consider the two most common parts of HR: a job requisition (job posting) and a job description. Both these artifacts are “thrown together” by hiring managers or HR professionals, often based on what people think a job is like, a set of company standards, and some “technical skills” we know this person will use. As we all know, these artifacts don’t really predict who will succeed, because so much of “success” is based on ambition, learning agility, culture fit, and alignment with purpose.

In other words, this is one of the most complicated and fascinating “mixed data” problems in the world, and making decisions a few percent better can drive billions of dollars of business value.

How Generative AI and Large Language Models Can Help

Given the complex, important, and messy business we are in, how can Generative AI and Large Language Models help? Well while it’s still early days, let me venture the idea that this new branch of AI has the potential to totally reinvent how much of HR works. And in this disruptive change we will see new platforms, new vendors, and new ways of running our companies.

(For those of you who don’t know what Generative AI and Large Language Models are, let me simply say these AI systems can index, categorize, and cluster billions of “tokens” which include words, phrases, numbers, and even code, to find patterns and predictions you can query. And through English language interface (and other language as well) they can analyze, summarize, and infer meaning from all this mess. Read about the statistics behind it here.)

Let me rattle off a few of the huge use-cases we’ve uncovered in the last few months:

1/ Creating content for job descriptions, competency guides, learning outlines, and onboarding and transition tools.

I’ve always felt that the best way to “describe a job” is to watch what people are doing. If you actually observe, capture, and analyze a few months of work, you could literally “write the job description” based on the actual work. Well Generative AI can do this.

You could use Generative AI to look at “the sales operation in your company” and analyze all the biographies, work histories, sales tools, and various sales materials in your sales organization. And it could likely describe “what sales people in your company do” and help you write realistic job reqs based on real roles.

Then if you want to know how to train sales people, you could ask it “tell me what the top performers do vs. the low performers.” And it would find things you may not know. And then you could ask the Generative AI machine to “read all our sales and product training” and “give me an outline of what people need to learn and know.” It could then build you tests, online learning guides, and eventually become the “sales coach” for your company.  (This is essentially what Salesforce Einstein GPT is trying to do – you don’t need to buy this from Salesforce by the way, you can do it yourself.)

Then you could ask the tool “who are our top accounts measured by total revenue and total profit” and if it has access to financial data it could answer that too.  So not only could it help you improve and rewrite all your job descriptions, it could help you “define the success criteria,” help you “evaluate who is performing well and why,” and then build the killer “sales training materials” you know are badly needed.

Now replicate this idea in manufacturing, marketing, finance, logistics, and even HR.

I’m sure it won’t be perfect at all this, but in a short period of time you’ll learn things you didn’t know and I would not be surprised to see these types of apps come “out of the box” within a year.

2/ Create skills models, experience models, and candidate profiles for recruiting

The second, and probably biggest spending area for improvement is recruiting. You all know how hard it is to find, assess, and select the “right person” for a job. Well right now everyone is gaga about “skills-based hiring.” But what does that really mean? Does it mean this person has passed a test in some tool or programming language? Does it mean they’ve done it 100 times before? Or does it mean they worked in a company that was really good at this so they probably learned a lot about it there?

See, it’s complicated. Supposed you could crawl millions of employee profiles and then look at the “work they did” (ie. scan Github, articles written, legal briefs, etc) and then decide “how good” this person is at this job? That would be almost impossible to do manually, but Generative AI can do this. And it can do much more.

Suppose it looks at this person’s biography and work history and then compares it to other candidates. It could probably tell you which has higher education, which has better spelling, and what other personal characteristics vary. One of the second-generation talent intelligence vendors we’re working with now has a tool that can show you “the leadership profile of company A” compared to “the leadership profile of company B” simply by scanning, analyzing, and deeply understanding the different experiences, language, education, and credentials of leaders from these two companies. Not a bad way to do your competitive analysis or recruiting eh?

I know L&D vendors who have already used ChatGPT to build lesson plans, learning objectives, and skills assessments from existing content. This kind of analysis applied to billions of job candidates can start to show recruiters who the “adjacent skilled” professionals are who could take that hard to fill job. They may have “related experience” that is 100% perfect for the job you’re filling. This is already happening, and it’s going to get better.

And by the way, with tuning these models can remove gender bias, age bias, racial bias, and more. So not only are they potentially more useful, they’re actually likely to be “safer” as well.

3/ Analyze and improve pay, salary benchmarks and rewards

A third massive challenge in HR is “how much to pay people” and “what benefits to provide.” And this is a very tricky subject. More than 95% of companies have pay equity problems already (our new research details this whole area) and as inflation goes up and salaries keep varying based on demand, HR departments are always struggling to keep up.

Generative AI can quickly do salary benchmarking, assess pay levels across millions of open jobs, and analyze external and labor market data to help understand competitive pay, rewards, incentives, and other benefit programs. Most companies try to do this by hiring expensive consultants: these consultants should soon come armed with AI-enabled tools, and then you’ll be able to get the tools yourself. I know of at least five vendors leaning into this today, and it is likely to make all these decisions better.

The whole issue of pay equity is a mess to fix as well. While some AI vendors are starting to focus here, we know from our research that most companies have 5-15% of their total aggregate payroll in some from of “inequitable pay” distribution. People get raises for political reasons and then over time we end up with highly paid, highly tenured people far overpaid based on their market salary or competition with others. I know software engineers who make $500K or more just because they hired into a “hot company at a hot time.” Suddenly a few months later they’re making 1.5-2X more than their peers. Companies hate trying to solve these problems.

4/ Performance management and feedback

One of the most difficult and often despised part of HR is performance management, performance appraisal, and development planning. While there are hundreds of fantastic books and models to define this process, it often comes down to personal judgement. And in most cases the manager gives an appraisal without doing a comprehensive look at an employee’s entire year of work. Imagine if the Generative AI system indexed a year of an employee’s work, hours worked, meetings, and other production and helped managers assess what happened?

Imagine then if the Generative AI took this work effort and perhaps compared it to similar roles to show the manager where this employee was outperforming and perhaps underperforming?  I know the technology can do this to some degree today: I recently asked Bing Chat to tell me how Microsoft’s financial performance varied from 2021 to 2022 and it did a pretty good job. Many of the new models of Generative AI can “learn skills” from this kind analysis and these “skills” can be saved and shared with others. And this leads me to the next use-case: Coaching and Development.

5/ Coaching and leadership development

As most of us know, the most valuable assistance we have in our careers is a “coach.” A Coach is someone who watches what we do, knows how it should be done, and gives us developmental feedback. They coach may or may not be an “experts” (many coaching models are built around the idea of “coach as psychologist”) so the coach may simply be observing us and giving us badly needed support. They may interview our peers and help us see blind spots and understand challenging situations.

Well today this market is explosive. Vendors like BetterUp, CoachHub, Torch, SoundingBoard, Skillsoft, and many others have created nearly a $billion dollar market for “coaching on demand.” Well what if this coaching came from an intelligent bot? Medical providers have built these systems for suicide prevention, medical intervention, and other support needs and they work quite well. In the business world this is an enormous area of “low hanging fruit.”

Imagine, for example, if I have to lay someone off. I could easily ask the ChatBot (which may have access to many guides, books, and videos from our company and experts): “how should I approach the layoff conversation?” Or “what is the best way to coach someone who keeps coming late to meetings? Or even “how can I have greater impact on my team” or even “how can I make my meetings more effective?”

These types of questions have been asked millions of time by millions of leaders, so there are well honed answers, suggestions, and tips for all of them. And most companies have license to leadership development content, compliance content, and all sorts of “difficult conversation” content now. The Generative AI system can easily find this, interpret it, and make it easy for managers to use.

And it will get better. Imagine, as I described above, if you put your own particular leadership model and approach to management into this system. You would get “the Starbucks store manager coach” or “the Fiat Chrysler manufacturing leader coach” and so on. My friends in the leadership development and coaching industry are probably excited (and nervous as well). This is coming fast.

6/ Individual Coaching, Mental Health, and Wellbeing

Perhaps one of the biggest successes in Generational AI has been the success of tools like “Woebot” which help treat mental health, stress, and suicide. This tool, which was launched in 2017, has reduced stress, anxiety, and suicide with almost twice the effectiveness of therapy. How could it be so good? By using the feedback loops in Generative AI (with human training), the system can quickly identify a user who is considering suicide and just by listening, help them relax and move forward.

I strongly recommend the story in the New Yorker this week (Can AI Treat Mental Illness) which convincingly explains how this technology has become so successful. These tools were not trained for work-related stress, but the problem is very similar. Over the last five years the Workplace Wellbeing market has grown to over $50 Billion in size and our research on The Healthy Organization found that the typical solutions (EAP programs, online coaches, training, mindfulness) have less impact than we expected. Witness the fact that most statistics on workplace mental health show that it continues to be a problem, even after billions of dollars have been invested.

This particular use case, which every company needs, could end up being pretty important. So we can expect healthcare providers, insurance companies, and forward-thinking vendors like (who now owns Headspace) to jump into this market.

7/ HR self service and knowledge management

The final use-case I will mention is self-service and knowledge management, perhaps the “lowest hanging fruit” of all. We all have thousands of documents, compliance books, diversity guidelines, safety rules, process maps, and help systems to aid employees in selecting benefits, understanding company policies, and even just resetting our password. And things like “figuring out what button to push in Workday or SAP” goes into this category as well.

All this complicated “knowledge enablement” and self-service stuff is perfect for Generative AI. Microsoft’s new Power Platform interface to OpenAI lets companies embed workflows into the system, so you could tell the chatbot “please apply for family leave and ask my manager for approval” or “please put a case into IT for me to upgrade my laptop.” And the use cases will go wild. Many of you who work in HR operations, call centers, and service delivery centers will be investing in this almost immediately. And that means every HR Tech vendor from Oracle to Workday to ServiceNow and ADP will embed this technology into its platforms.

Bottom Line: This Technology Will Make Work Better

Let me make one final point. Despite the fears and inflammatory headlines you read in the New York Times (the NYT seems particularly unhappy about this technology), I want you to remember that this technology will be a massive step forward in business. Last week I published an article by two MIT PhD students who analyzed the use of ChatGPT on 400+ business professionals and the productivity improvements were stunning. This will start to happen in all these other areas as well.

Will it be perfect? Of course not. But today, as I touch on above, we make thousands of critical decisions with poor data, uneducated judgement, and often just not enough internal research. I believe Generative AI and all its variations will be a total gamechanger in HR. And for everything we do just a little bit better, our employees end up with a better work experience and our companies perform at a higher level.

Stay tuned, there is much more to come.

How To Learn More: Come To Irresistible 2023

How can you learn more?  Here are a few big resources.

First, come to Irresistible 2023 – our big annual conference held at USC on June 20-22. We are bringing in several famous experts in Generative AI for HR to that event and we are going to have a big discussion about how it will impact HR. I promise you this will be well worth your time.

Second, if you’re a geek like me, read Steven Wolfram’s article “What is ChatGPT and How Does it Work.” And I will post some other cool articles below.

Finally, if you are doing something exciting and want to share, please let us know. We are talking with hundreds of companies about these tools and I promise we will share more as time goes on. I will be presenting some of our newest research at the HR TechFest Singapore in mid-May and then much more at the Irresistible Conference in June.

Onward and upward!

Additional Resources

New MIT Research Shows Spectacular Increase In White Collar Productivity From ChatGPT

Microsoft Launches OpenAI CoPilots For Dynamics Apps And The Enterprise.

Fighting ‘Woke AI,’ Musk Recruits Team to Develop OpenAI Rival

Mark Zuckerberg announces Meta’s new large language model as A.I. race heats up

Understanding Chat-GPT, And Why It’s Even Bigger Than You Think (*updated)

What Is A Large Language Model (LLM) Anyway? (good overview)

Why Microsoft’s Investment in OpenAI Threatens Google (Fortune)

Listen to Satya Nadella Describe Microsoft’s View of OpenAI

Something Bothering You? Tell It to Woebot.



Source link


Leave a Reply

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