HR Grease | Issue #003 | For HRIS practitioners who keep the machine running
In this issue:
The gap between what your vendor calls "AI" and what it actually does to your data
Hot Take: the AI adoption problem is a vendor problem, not an admin training problem
Quick Win: three questions to ask in any AI feature demo before you say yes
01 — The Grease
What "AI-powered" actually means in your HRIS — and what it doesn't
Your HRIS vendor's rep just demoed a new "AI-powered" feature. Maybe it's a natural language report builder. Maybe it's a predictive scheduling tool. Maybe it's an AI assistant that "interprets" your HR policies and answers employee questions directly.
The demo looked clean. The use case was real. You left the call thinking: that could actually save someone time.
Here's what didn't come up in the demo:
Where does your data go? When an employee asks a question and the AI "interprets" your policy documents to answer it — which system processed that request? Your vendor's model? A third-party LLM? A shared training environment? Most reps genuinely don't know the answer to this, and that's a problem on its own.
What's the failure mode? AI features in HRIS don't fail by crashing. They fail by confidently giving wrong answers — a policy misquoted, a calculation that looks right until an audit, a manager who made a decision based on an AI summary that missed a key exception. These failures are invisible until they're expensive.
What happens when it hallucinates on a compliance question? This isn't hypothetical. Employees will ask the AI about FMLA eligibility, overtime rules, PTO carryover. When the AI gets it wrong — and it will, eventually — who's responsible for the downstream decision? It won't be your vendor.
The problem isn't that AI features are bad. The problem is that "AI-powered" is a marketing claim, not a technical specification. Your job is to figure out which one you're actually buying.
The question isn't "should we use this AI feature?" It's "do I understand this feature well enough to know when it's wrong?" If you can't answer that, the feature isn't ready for production — regardless of what the vendor says in the demo.
02 — Hot Take
Low AI adoption isn't an admin training problem. It's a trust problem.
"Training fixes AI adoption" is the same thing we were told about every other HRIS feature that launched with a splash and quietly died in a help desk queue.
Vendors frame low AI adoption as an admin skills problem. It's not. It's a trust problem — and trust in an AI feature is earned by the feature doing what it claims to do, consistently, with visible failure modes. Not by running employees through a 30-minute onboarding video.
I've watched orgs roll out AI features with full change management plans and adoption dashboards while the underlying feature produced wrong answers at a rate nobody was measuring. The metric was "did the feature get used?" Not "did using the feature make things better?"
There's a version of AI in HRIS that's genuinely useful. I've seen it in workflow automation, anomaly flagging, and specific narrow use cases where the failure mode is visible and recoverable. That version doesn't need a change management plan. It just works, and people keep using it.
The real question: when your vendor says "AI adoption is low" — are they measuring usage, or outcomes?
03 — Quick Win
For any HRIS platform: three questions for your next AI feature demo
Before you approve rolling out any AI feature — or before your next vendor call — have these three questions ready. Write them down. Ask them out loud.
"Where does our employee data go when this feature runs?" Get a written answer. "The vendor's secure infrastructure" is not an answer. You need to know whether a third-party model is involved and what data leaves your environment.
"What does a wrong answer look like, and how would we know?" If they can't describe a specific failure scenario, they haven't stress-tested it. The vendor should be able to name the failure mode before you can.
"What's the escalation path when the AI contradicts our written policy?" This question tells you how seriously they've thought about the liability side. If they haven't, that's your answer.
You're not trying to kill the feature. You're establishing that you understand it. That's the same standard you'd apply to any configuration change. AI features don't get a lower bar.
📊 Quick Poll
Has your HRIS vendor demoed an "AI-powered" feature to you in the last 6 months?
→ Yes — and I had more questions than they had answers
→ Yes — and it actually seemed solid
→ Not yet, but I'm expecting it soon
If someone forwarded this to you — welcome. 250+ HRIS practitioners are already getting this every month; subscribe free at hr-grease.com to get the next one directly.
What's the AI feature your vendor is pushing hardest right now? Reply and tell me — I'm tracking which ones are showing up across orgs, and which ones HRIS admins are quietly ignoring.
Next issue: the RBAC configuration mistake that almost every implementation gets wrong — and why it usually doesn't surface until someone has access they definitely shouldn't have.
— JR Cecil, Lead HRIS Sysadmin & UKG Pro Specialist
linkedin.com/in/jr-cecil/