Welcome to FullStack HR, and an extra welcome to the 107 people who have signed up since last week. If you haven’t yet subscribed, join the 10 000+ smart, curious HR folks by subscribing here:
Happy Wednesday,
Ok, we did it. We crossed the 10k mark.
I can not believe that there are over 10,000 of you reading this newsletter.
It’s insane.
I do not take this for granted, and I’m forever grateful for this.
Thank you!
But that said, let’s dive right into today’s article.
This is more food for thought than anything. Please remember that when reading this!
TL;DR: If you price only compute and voice, the gap is not small; it’s insane.
And that explains a thing or two.
(And yes, I know work is more than tokens and minutes.)
When discussing AI and its potential to replace us, we often assume that humans are unique. I’ve discussed this before, and I think we often overestimate our uniqueness.
But what if we, for a moment, set aside that and look at the pure economics of AI vs. a human?
Simply, what happens if we compare the monthly cost of these two:
A human HR Business Partner or a human Recruiter
A narrow AI version of the two
What would that look like? Together with GPT-5 Pro, I tried to figure that out.
I loaded GPT-5 Pro with data on what an HRBP and a recruiter typically do, as well as generalized employment data and pay data from open sources.
I know that this is a simplification, just as with tests such as GDPval, it’s hard to break down tasks and call that a job.
Sidnote here, but GDPval is OpenAI’s new way to measure what AI can actually do in the real economy.
Instead of testing models on trivia or coding puzzles, it looks at real jobs, 44 of them, across different industries.
The models get real tasks, like writing reports or building plans, and human experts judge if the AI’s work is better, equal, or worse than what professionals deliver.
It’s basically a new benchmark for “AI at work,” tied to GDP, not theory.
And early results show that overall, these models are slowly reaching human-level answers on these tasks.
Back to the piece here, this is not a blueprint. This is more food for thought than anything.
But I kind of like that as well, to investigate and ponder about things like this, and I’ll get back to this shortly, but let’s look at the math first.
The calculations for the AI versions are based on a task-based assumption, e.g., how many tokens would the AI HRBP need to fulfill the same tasks as a human HRBP would? You can see how GPT-5 Pro and I made these assumptions at the end of this post.
The cost
AI HRBP (tokens only): ~$226 per month.
AI Recruiter (tokens only): ~$308 per month.
These include speech in and speech out with OpenAI’s Realtime model, plus text in and text out with GPT‑5. Voice cost dominates. Text is pennies (!).
Human HRBP (total employer cost): ~$14,317 per month.
Human Recruiter (total employer cost): ~$13,476 per month.
That is the base salary, as reported in open sources such as Glassdoor, LinkedIn, Indeed, and similar platforms, and then multiplied by a standard 1.3 to account for benefits and taxes.
To summarize this in another way:
1 human HRBP ≈ 63 AI HRBPs at token‑only cost.
1 human Recruiter ≈ 44 AI Recruiters at token‑only cost.
Even if I am off by a factor of 10, the result still bites.
1 human HRBP ≈ 6 AI HRBPs
1 human Recruiter ≈ 4 AI Recruiters
“But the system…” Correct. The system around the model is not free. Hosting, database, observability, and retrieval add up. Plan for a monthly budget of around $100 to $200 for a small, production-ready setup. And you can share that across a team. Spread $150 over 20 seats, and you add about $8 per seat. The ratios barely move.
I want to emphasize again that it is not as simple as having the model do tasks and voila we can replace a person. It shows that the raw cost of compute and voice is tiny compared to a full salary.
But it also shows the incentives for rapidly creating models that can replace people - saving money.
Why am I writing this?
As said, we in HR like to say, “This job cannot be replaced.” I get it, and as said above, tokens and tasks do not solely make up a job. There are human parts here that matter. Context. Judgment. Trust. Legal risk. The list is long.
But yet, the price signal is clear.
An AI HRBP or AI Recruiter is way cheaper.
And if compute keeps getting cheaper, the allure of cheap “labour” for vendors and organisations is even more obvious. And since we operate in a world that tends to favor cheaper options, we can potentially see where the push for using AI is coming from.
Where’s the trade-off between a slightly less capable AI HRBP for 63 times less the cost vs. a decent human HRBP? Let’s say you currently have 10 HRBPs, would your CEO be ok with 50% of the HRBPs being AIs to handle the bulk part of the work, the grunt work, for 1/10 of what ONE human HRBP costs? Yes? No?
And since GDPval shows that these models are becoming more and more capable of handling individual tasks, and when the price is ridiculously low, I think that will lead to us being challenged in not only our ways of working but also in our professions.
The small print
Assumptions for the “AI HRBP”
22 hours of meetings per week. Agent speaks 35 percent of the time.
OpenAI GPT-realtime pricing is used for audio tokens. $32 per 1M audio input tokens. $64 per 1M audio output tokens. Per‑minute figures are estimated from OpenAI’s examples and scale with these token prices. OpenAI
GPT‑5 text pricing: $1.25 per 1M input tokens and $10 per 1M output tokens. 1 token is about 0.75 words in English. OpenAI
That yields about $225 for voice and ~$1 for text per month. Total rounded to $226.
Assumptions for the “AI Recruiter”
28 hours of calls per week. The agent speaks 40 percent of the time.
Same models and pricing as above. Voice dominates. Text is about a dollar.
Total rounds to $308 per month.
Assumptions for the human roles
HRBP average base salary in the US: $132,158/year on Glassdoor. Multiply by 1.3 to include benefits and taxes. Divide by 12. ≈ $14,317/month. Glassdoor
Corporate Recruiter average base salary in the US: $124,394/year on Glassdoor. Same 1.3 multiplier. ≈ $13,476/month. Glassdoor
Note: If you prefer ElevenLabs for TTS or Scribe for STT, the voice math changes a bit but the story does not. Scribe lists $0.22–$0.40 per hour depending on plan. The effect on the totals is small compared to salary. ElevenLabs
Very interesting and challenging
Great thought exercise. The most interesting part isn't if you could replace all HRBPs, but some and have them work with AI. I feel like I leverage this a lot as an independent HR consultant. AI helps me get to solutions much quicker for my clients and allows me to do a lot of work in a short amount of time - which is the point, right?