HR GPT: A Pragmatic Look Into the Hype (for HR Professionals)
Kurt Anderson
Dynamic human resources leader who loves growing small companies to their next inflection point.
I’m a technology enthusiast so, naturally, I’ve been following the slow-rolling explosion of AI tools into the general marketplace. It’s definitely an exciting (and scary) time to be alive, and I’m here for it, either way.
I’m also a human resources professional and my industry isn’t particularly known for early adoption of technologies. (If I’m being brutally honest, we’re not even known for being especially tech-savvy with well-established technologies.) I was reminded of this trend when researching ways that HR practitioners can leverage Chat GPT.?
The listicles that I found were one-third “Chat GPT will be able to do this amazing thing (eventually… and with extra tools in addition to Chat GPT),” one-third “Chat GPT has solved this problem (that’s already been solved for some time now),” and one-third “this is a real and interesting Chat GPT use case.” Maybe the authors needed to hit a particular word count (I get it), maybe they themselves weren’t HR people (so it was hard for them to distinguish between useful applications and fluff), or maybe it’s something else entirely. Either way, I’ve decided to write the article that I wish they’d written and hopefully save you, dear reader, some time.
I don’t want hype, doomerism, or protectionism; I want to affect change. So, I tend to look at Chat GPT use cases through two lenses:
Hey! These two lenses can serve as axes, can’t they? Add in some relevant thresholds and you have a lovely blank matrix:
Looking at the matrix while asking “so what should I focus on next?” yields some distinct categories:?
Let’s explore each category. If you’re short on time, just read the sections on categories 1 and 2.
You know what? I’ll give you the master list upfront. No need to keep you waiting:
A brief aside: the "intern frame"
Before we get to the categories themselves, let’s talk mental frameworks.
In general, the best way to think of Chat GPT (at least for now) is like your average college intern: little life experience, no work experience, but generally able to follow instructions (particularly when given examples). So, when asking yourself “how can I best leverage AI today” just imagine that you’ve been given a free intern for the summer. What projects would you give her? How much explanation (and oversight) would be needed for each initiative? The answers you come up with won’t be off by much when applied to the Chat GPT question.
If you’ve ever worked with this type of intern you’ll know that, just because you’ve thought of a project that they hypothetically could do, it doesn’t mean it’s a project that you should give them. Whatever output they generate needs to be considered a first draft (at best) which you should review for quality and then edit to better suit your needs. If the time and effort to QA this output exceeds (or even approaches) the time gained from outsourcing the task to your ‘intern,’ then it’s probably not a great project to delegate to this particular ‘individual.’
OK. Now we can move on to the tactical.
Category 1: Start Soon or Get Left Behind
These are things you can do with basically no more training than watching a few tutorials on YouTube and playing around with Chat GPT for a half hour. The gains in productivity are so obvious and manifold that, if you aren’t doing them, you will soon be shown up by someone who is.
1. First Drafts & Brainstorms?
I find that the majority of us are afraid of a blank page, and are much more comfortable editing than drafting. So using Chat GPT to immediately jump a project from 0% to 20% completion is a no-brainer. It’s no surprise that this category leads the pack; generating passable chunks of written language instantly and at scale is literally the point of the LLM (“large language model”) chat interface.??
You’ll see a lot of variations on “Chat GPT can draft emails for you” which, while true, is thinking small. GPT-4 is able to generate outputs of 25,000 words (that’s a 100-page novella); now we’re talking real gains! And the output isn’t just limited to emails, paragraphs, and essays. Chat GPT excels at lists, which means that you can jump to 20% completion for brainstorming and project outlining.
Here are just a few items that you should start using Chat GPT immediately to generate first drafts when you have no template and are starting from scratch:
Just remember: this is a first-draft-producing 'intern' with no work experience who moves your starting line from 0% to 20%. This is not a wish-granting genie who moves the starting line from 0% to 85%. Set your expectations accordingly.
2. Data Hygiene
One of my favorite intern projects is downloading my HRIS data and saying “go through this, see where the gaps are, and fill them.” You know that every employee has some data that isn’t present: personal emails, demographics, emergency contacts, you name it. And business analysis is only ever as good as business data, so this type of data hygiene is something to stay on top of.?
Depending on your company size, a project like this can take weeks (and, let’s be honest, you might only find out after he returns to school that your intern doesn’t have the “attention to detail” core competency). However, if you’re able to download your HRIS data in a spreadsheet, you can feed Chat GPT a table and have it spit out a list saying “John Smith is missing an emergency contact and home address. Jane Doe is missing a personal email address…” and so on.
If you’re very clever, you’ll combine the use case from item #1 and have Chat GPT draft an email for each person telling them what they, specifically, are missing and to please send it to you (or add it directly to the HRIS themselves, if they’re security-minded). Boom: you can audit gaps in employee records for a hundred, a thousand, or ten thousand employees and have custom comms drafted for each individual with an hour or two of work.
Maybe it’s just me, but I get giddy just typing that last sentence.
Category 2: Stretch Yourself and Find Hidden Value
These are the areas that will take some time and additional skills beyond just learning how to talk with Chat GPT. You won’t be able to show up to work tomorrow and immediately change the status quo. But, with a little time (weeks and months, not years and decades), you’ll see the gains. You’ll be the ant who gathered in the summer while the grasshopper was dancing. (And, to completely left-turn that metaphor: “winter is coming.”)
So, while you should absolutely start using Chat GPT for the Category 1 use cases, maybe pick just one or two items in this section to explore further.
3. Data Analysis
This is in the “extra skills” row because, let’s face it, your ability to extract data insights using Chat GPT will correlate closely with your ability to extract data insights without Chat GPT. Data analytics is often about knowing what questions to ask and what data sets to assess. If you really want to supercharge things (and overcome security concerns), you’ll also want to pick up a fluency with APIs so that you can access your data directly and in real time. After all, we know how temperamental most HRIS report generators can be.
That being said, if you work on improving your data analysis skills along with getting comfortable with Chat GPT, a whole lot of possibilities are unlocked. Evaluations that might have taken a couple of days in Excel - flagging out-of-bounds compa-ratios, checking correlations between tenure and demographic groups, normalizing performance evaluation metrics across managers - can take a half hour or less with a skilled operator (or "prompt engineer," as the kids are calling them these days).
4. Upskilling Yourself
Speaking of learning additional skills, Chat GPT is already here to help you! (A bit meta, I know.) There’s never been a better time to learn to code using AI assistance. Or just skip the coding altogether and build sites, plugins, and web tools using Chat GPT as your coder. And, if you’re a company that uses Microsoft Office, their new in-platform integrations with Chat GPT will enable you to do things in Excel that you once thought were beyond your skills!?
5. Custom Chat Bots
Half of the ideas in the “here’s what HR people can do with Chat GPT” articles that I’ve read boil down to chatbot functionality. And it’s a good idea: if employees have questions, they can type the question in and they’ll receive an answer, whether about internal policies (bereavement leave, change of address workflow, locating a needed form) or external laws and processes (“what are the steps in the H-1b immigration process?”). This is an obvious use case and one that naturally appeals to the HR professional who spends much of her day answering employee questions (ones that likely could have been answered by the employee going “CTRL+F” in the handbook or using the search field on your intranet).
However, there are two things to consider here.
First: this is hardly an out-of-the-box use case of Chat GPT. Because every business is different, you’ll have to set up a custom chatbot that knows your policies and procedures (because wrong answers are worse than no answers). You can use Chat GPT to generate training data for that chatbot (which is immensely helpful), and your proprietary chatbot might ultimately be powered by Chat GPT’s large language model. But it’s still a separate technical entity that you (or, more likely, your IT organization) will have to set up, train, and maintain.
领英推荐
Second: your mileage may vary with the payoff here. Some organizations are more culturally “trainable” than others; I’ve been in places wherein employees quickly acclimated to looking for their own answers on the intranet, and I’ve been in places that never quite got the hang of the self-service model. If you’re in a place where employees are good about looking for answers themselves before consulting HR, then the payoff of implementing a custom chatbot diminishes.?
There’s a similar tradeoff between the size of the organization and payoff: a company of 50 people likely doesn’t generate enough random HR questions to justify the implementation and maintenance of a chatbot.?
However, the larger your company is and/or the greater its cultural aversion to employees looking things up for themselves, the more this Chat GPT use case moves toward “no-brainer” territory.
6. Workflow Intervention
Here’s a neat use case that, as with #5, is a bit situational and might require some additional tools. (I’m thinking of Microsoft Power Automate in particular here; it’s already amazing and, when paired with Chat GPT-like features, could be unstoppable.)
Here’s an example: as part of your onboarding process, you send out automated 30-day and 90-day pulse surveys to new hires and to their managers. However, “review survey responses” is the kind of non-urgent task that constantly gets pushed down in your priority hierarchy. So, you decide to set up Chat GPT to scan the survey responses (you’ll likely need to set up some custom Python or an API to make this automatic, even if you’re using Power Automate). Chat GPT then sends you a message if a survey response indicates that something is off or could use your attention and (just as importantly) leaves you alone if everything is nominal.
Or (slightly more advanced but still entirely doable), in addition to everything above, maybe Chat GPT proactively messages the new hire and says (on your behalf), “I saw your survey response. Let’s meet to figure out how to solve your problem. Click this link to pick a 10-15 block of time that works for you.”
This type of use case can quickly edge close to Category 3 as you go down the rabbit hole of possibility. But there are a bunch of implementations that are completely doable today.
Category 3: Keep an Eye Out for Developments
These items are theoretically possible, and may even be on the horizon. But the tech isn’t quite there yet, either due to AI not being sophisticated enough or due to a current lack of integration between disparate systems.
7. Sentiment Analysis
This functionality is creepy and Big Brotherly, but it’s so insanely useful that I have trouble believing that companies won’t implement some version of it.
The fact is that most work activities in many companies are currently conducted in ways that are ‘readable’ to an AI: Slack/Teams chats, email, cloud documents… heck, even Zoom meetings can now automatically be transcribed into text (and, thus, are able to be scanned by a large language model AI). Imagine an entity that can monitor all of these communications and (leaving aside privacy and security concerns) generates a weekly report for leadership that says “Engineering Team Alpha seems to be unusually frustrated lately” or “people who report to Manager Beta and Manager Delta exhibit above-average levels of disengagement.”
Like I said: this kind of leverage is so insanely valuable that it’s hard to imagine a future (however unsettling) where something like this isn’t present at top companies.
8. Onboarding Logistics
Onboarding, from signed offer through the end of probation, can be seen as a long series of simple-but-time-sensitive-and-workflow-contingent tasks that involve activities ranging from mailing welcome packages full of swag to having live check-ins with new hires and their managers. The “simplicity” and “predictable workflow” aspects of onboarding are ripe for automation but, for the solution to do anything more than send out text reminders, it would have to be integrated with swag vendors, FedEx, survey software, your HRIS, etc. Automating this through Chat GPT-like AI falls squarely in the “theoretically possible, but not many yet practicable” camp.
9. Advice Bots
This solution relies on the same logic as the ‘Custom Chat Bots’ use case, but I’ve categorized it as something distinct because it requires a level of judgment that goes well beyond current AI capabilities. If a manager asks for advice regarding an employee relations issue or wants tips for delivering tough performance management advice, there are dozens of factors to consider that an AI would not only have to be capable of grokking but also be wise enough to ask about in the first place. It’d be amazing functionality but, frankly, I think this will likely be the last thing to be automated away from HR. (And, when it is, we’ll be so close to AGI that we’ll have a lot more pressing concerns than losing this type of task.)
10. Data Completion
This is like the ‘Data Hygiene’ use case, but on steroids. Let’s say you are interested to learn the correlation between your employees' college majors and performance outcomes. But, though most HRISes have fields for things like education, I’ve yet to see those fields consistently filled. However, with a large language model that is connected to?LinkedIn, you could have the AI do the work for you by looking at resumes and work histories, extracting the relevant data, and creating a mass upload file for your HRIS. The technology isn’t there yet, but I’d be surprised if it’s not here relatively soon.
Category 4: Ignore the Hype
This is the “so what” category. (Put more bluntly, it's the "stop including these in your AI hype articles" category.) Chat GPT can do these things, maybe even do them well, but the impact is so minimal (and possibly counterproductive) that an appropriately lean organization will never get around to implementing them.
11. Glorified Search Engine
“Chat GPT can alert you to changes in legislation or industry practices.”
If you’ve been in HR for more than a few years, you already have several different (human) institutions - your local SHRM chapter, your national SHRM subscription, your benefits broker, your retained attorney, etc. - that curate and email you this information once a week. Adding another news courier to my feed, at this point, is more of a burden than anything else.
12. Evaluate Resumes
“Chat GPT can screen applications.”
Aside from the ethics- and inclusion-related minefield that is “automated candidate elimination,” I’ve found that dedicated resume-evaluation solutions simply aren’t very good (sorry, Jobscore). And, even if such dedicated solutions were consistently effective and predictive, that very fact would make them easily gamed by savvy applicants. Because of that (and the other minefields), I don’t think that a non-dedicated, non-optimized resume evaluation solution like Chat GPT is something worth leveraging.
13. Paperwork Delivery
“Automate sending onboarding paperwork and tasks to new hires.”
Nearly every notable HRIS already has this automated reminder functionality. Moreover, this feature is already integrated with the forms/fields/systems that need to be completed. Tacking on a 3rd-party solution in the form of Chat GPT simply for the sizzle of saying “you’re being reminded to do this by an AI!” seems a bit silly to me.
14. Automated Scheduling
“Now an AI can schedule interviews so you don’t have to!”
If part of your recruiting process involves a coordinator emailing/texting and saying “send me some dates and times that you’re available,” waiting for the applicant to respond, then mapping their response to a schedule, then writing back and saying “let’s do this date,” then waiting for the applicant to confirm… I hate to say this, but you’re already way behind the times. Changing things so that Chat GPT acts as the coordinator just means that you’re behind the times in the most efficient way possible (which is an excellent way to get stuck behind the times).
This is already a solved problem, and one that doesn’t include inconvenient back-and-forth (that no candidate enjoys). Check out Calendly . If your company uses Google Suites, this is a built-in functionality .
OK, posts like this need conclusions don’t they? Ahem…
Like most HR technologies, Chat GPT has the potential to be a force multiplier that, at the end of the day, raises the floor of your company’s ideal employee-to-HR ratio. I generally live in the startup space where resources can be tight, so I can see this allowing a lean organization to delay increasing HR headcount for an extra two-to-four dozen hires. Not bad!
But make sure you’re spending your time implementing Chat GPT in high-value areas or you'll quickly cannibalize your productivity dividend. Ignore the hype, focus on impact, and you’ll be able to make some impressive moves.
Head of People & Commercial Ops @Precisely - Automating Contracts
1 年This was a good read, Kurt. Thanks for taking the time. I agree on seeing LLMs as an intern, that’s the way I view it as well. An additional use case is using LLMs to analys survey data to find themes, topics and sentiments that you should focus on.
Sr. Leadership Development Consultant @ Mass General Brigham | Organizational Psychologist | Adjunct Professor @ William James College
1 年I gave chat GPT a crack at writing a TD program design overview after putting in some inputs. It wasn't bad! Gave me a head start.