HR's Guide to Buying AI

Every vendor sounds the same in a demo.

Here are the 8 questions that tell you which ones are actually agentic AI, and which ones are chatbots with a new label.

Get the guide

Agentwashed AI vs. real agentic AI

Agentwashed AI vs. real agentic AI
  • Scripted demos that avoid your real scenarios
  • A chatbot that answers questions but doesn't do the work
  • Vague answers on data retention and model training
  • No dashboard, or a custom one built just for you
What real agentic AI looks like
  • Runs a live, unscripted case in the room
  • Works the case end to end, not just questions about it
  • A clear "no" on data retention, stated in the contract
  • Compliant, policy-based responses every time
  • A live dashboard you can see today

Where the AI-HR buying process goes wrong

Adoption is outpacing readiness
61%

share of HR orgs piloting or implementing generative AI, up from 19% in 2023 

Source: Gartner

The investment isn't paying off yet
88%

of HR leaders say they haven't seen real business value from AI 

Source: Gartner

Most "agentic AI" isn't
~130

real agentic AI vendors, out of the thousands marketing themselves that way
Source: Gartner

The 8 questions to ask before you buy

Harper isn't a general-purpose model told to act like HR. It's powered by a domain-specific reasoning layer purpose-built for HR's complexity and stakes.

01

Run it live

Can you handle a real, multi-stage scenario from our team, not the scripted one?

02

Cover the whole job

Can you walk us through our 10 most time-consuming processes, one by one?

03

Prove the connection

Read-only or write-back, for every system we run.

04

Answer it consistently

Same question, two different employees. Same rigor, different answers.

05

Say no to retention

Is our data used to train a model that isn't ours?

06

Show the references

Two customers, including what delayed them.

07

Show the dashboard

Live, today. Not a custom report built after the sale.

08

Show the math

Priced at our headcount, 1.5x, and 2x.

What a strong answer sounds like

Whether you're evaluating an internal build or an AI add-on, the gaps are significant. Here's how they compare.

Question
Strong answer
Weak answer
Run a real scenario
Runs your unscripted case live in the demo
Pivots back to the scripted walkthrough
Data retention
"No," and it's in the contract
Hedges on where data lives or who sees it
The dashboard
Shown live, today
No eval framework. No systematic visibility into what the model is actually saying to employees.

How to use our scorecard

Rate each vendor

Score them against the same 8 questions, not whatever came up in the room.

Compare side by side

Get scores you can defend, not a gut call.

Adjust the weights

Match the scoring to what your organization actually prioritizes.

Bring it into the room

Use the scored sheet as your actual agenda when vendors get on the call.

See how Harper holds up against your own scorecard

Meet the AI teammate purpose-built for the complexity of HR.