The Next Level Of AI In HR: Large Language Models

Safely leveraging the best of your data

Much of the general public discovered how prominent artificial intelligence is this year with the mass adoption of generative AI (genAI) platforms DALL·E and ChatGPT. From creating new images based on text to having a major impact on the Hollywood screenwriters' and actors’ strikes, genAI is here to stay.
 

For many, however, AI is nothing new. You have likely been using AI in some capacity in your HR role already, such as through data analysis or as a tool for self-guided compliance training and onboarding. This iteration of AI is possible thanks to machine learning (ML). Machine learning teaches AI how to perform a specific task in a specific way based on a specific set of instructions: If X then Y.

GenAI is more sophisticated thanks to large language models or LLMs. Large language models have the potential to impact any process that uses language. The reason they are so appealing is that in a world with limitless applications for words, LLMs unlock new capabilities for organizations to structure, manage, and gain insight into any process that involves language. Properly implemented, LLMs can provide reliable insights across documents and databases that previously had to be done manually or through use-case-specific machine learning.

How can HR teams take advantage of this new iteration of AI? What impact can it make on their teams and companies as a whole? Let’s explore just a couple of examples:

Personnel Data

Your company’s personnel records are full of education, past work experience, performance reviews, and other insights into your employees’ backgrounds and successes. One way to use this data is for recruitment. Rather than your recruiting team spending hours digging into anonymized data to try and see what commonalities exist between high-performing and/or long-tenured employees, they could simply ask, “What are the common education experiences between long-tenured employees?”

This can also help refine job postings to attract the exact right talent. Instead of using generic phrases like “self-starter,” recruiters can specifically share with candidates that the employees who typically perform well in this role are those who would be described as self-starters or take initiative on new projects. LLMs confirm what you might know about your recruiting efforts. They can also bring to light secret weapons to get a leg up on your competitors hiring from the same talent pool.

Similarly, LLMs can examine personnel data to see what correlations may exist between current employees and new job opportunities within the company. This may be easy to do manually when identifying promotions or vertical mobility, but AI can review the entire company to see what untapped skills employees have and how they could be a good fit for a new role, no matter their current job title. If an HR role opens that will require intricate knowledge of insurance policies, AI could identify a sales employee who is well-versed in contracts and held an internship at an insurance agency before this current role. Without knowing to look at someone in sales, HR may never even consider that person for a new role. AI can find this gem in moments.

Internal Operations

How do you train new employees? Does your organization have a multi-day in-person training seminar? Self-guided online learning courses? An informal walk-through of the employee handbook? No matter how you train employees, LLMs can help employees learn and stay up to date. For example, LLMs give new employees an easy way to reference onboarding materials. It may be easy to skim a 20-page employee handbook looking for the “PTO” header, but it’s much easier to ask “What is our PTO policy?” and have the answer delivered to you in seconds.

For companies with lengthy and complex safety and security processes, employees can pose hypothetical scenarios to the LLM, which can then distill the steps individuals would need to take. Because a large language model doesn’t just learn but contextualizes and provides answers in humanlike responses, it knows how to reply to complex questions in easy-to-understand language while also citing its sources, rather than spitting out a clause from the safety manual with no nuance. If an employee asks, “What is the maintenance schedule for the facility’s generator and when might I need to use it?” the LLM can say exactly what they should do, instead of just saying, “See page 23 of the handbook.”

HR research becomes much smoother, too. HR teams can examine their company’s existing salary ranges and then compare those with industry averages and current inflation rates. Pay doesn’t live in a vacuum, so HR teams can understand historical trends, identify outliers, and highlight pay discrepancies that may exist in their company. Compensation is a key area in which human oversight is important. In cases such as salary negotiations and work compensation, HR directors and managers need to validate the recommendations from LLMs and remember that’s all they are, recommendations. Humans should always have the final word, but it’s easier to know the final word when you have all the facts and data to back it up.

Getting Started

Don’t pigeonhole HR. LLMs can securely parse data across an entire organization so the right solution must be a fit for the company overall, not a specific department. Teams can then customize their data-gathering and reporting processes in ways that best fit their needs.

Identify what your company is trying to accomplish as a whole. Perhaps it is growing revenue, improving customer service, and measuring the productivity of a four-day workweek. The macro objectives of your company will help guide the strategy your genAI investment strategy.

Security concerns are valid, but they shouldn’t handicap your research and implementation. Well-managed LLMs integrate safely with a company’s data, allowing companies to protect their proprietary data while leveraging the best of that data and the LLM.

It’s also okay to start small. HR teams don’t need to revamp hiring processes just because they are now using LLMs. Ask questions on what you want to see from the LLM and start with those results. The most important step: Start. Don’t miss an opportunity to learn more about the people who make your organization successful and why.

Author Bio

Morgan Llewellyn, Ph.D., is the Chief Data and Strategy officer at Stellar. Morgan has a wealth of experience helping envision and implement data and AI solutions across government, healthcare, SaaS, IoT, retail, and manufacturing. At Stellar, he helps businesses assess their AI readiness, identify and prioritize AI opportunities, and implement solutions to securely improve end-to-end operations and financial outcomes.