Every AI project eventually has to survive the same meeting: the one where someone with a spreadsheet asks what this actually costs, and whether it's worth it. Most articles about AI in HR skip that meeting entirely and stay comfortably in the realm of "transformative" and "reimagined." This one won't. Here's the actual math behind MITS's AI onboarding system, and why the numbers are more interesting than the marketing.

The Old Cost: Human Time, Repeated Endlessly

The real cost of traditional onboarding isn't a line item, it's hidden inside HR headcount. A 20 to 30 minute orientation call, repeated for every single new joiner, run by a person who could otherwise be doing higher-value work, doesn't show up on a budget as "onboarding cost." It shows up as a permanently busy HR team and a first day that gets whatever attention is left over after everything else. At a company bringing on an estimated 20 to 25 new joiners a month, that's 8 to 12 hours of repetitive, scripted HR time monthly, every month, indefinitely, just for the orientation call itself.

The New Cost: Mostly Usage-Based, Mostly Small

MITS's system replaces the bulk of that repetition with a handful of metered cloud services, and the actual bill is smaller than most people expect.

D-ID's avatar streaming, the most expensive line item per minute, is billed only while the video stream is actively open. At roughly three minutes of avatar time per call and 20 to 25 calls a month, that's 60 to 75 streaming minutes monthly, comfortably inside D-ID's Launch plan at $35 a month billed annually, which includes 90 minutes with a commercial license. No overage expected at this volume, and HR can shorten avatar time further at will by toggling it off mid-call.

The knowledge base that powers Q&A runs on Azure AI Search during development and demo, on its free tier, $0, with 50 MB of index storage, enough for roughly 800 to 1,000 source documents, more than sufficient for a company knowledge base. In production, that same workload moves to Qdrant, running in Docker on MITS's own on-premise servers, where the marginal cost is server infrastructure MITS already operates rather than a new recurring SaaS fee.

GPT-4o and the embedding model are billed per API call, priced by usage rather than seats, which means the system costs roughly the same whether it's used for 5 onboarding calls a month or 50, unlike a headcount-based cost that scales in fixed increments.

The Part That Doesn't Fit on a Spreadsheet

The harder number to quantify, but arguably the more important one, is consistency. A human HR rep on their fifteenth call of the month is, understandably, not delivering the same energy as their first. Priya's twentieth call of the month sounds identical in warmth and pacing to her first. That's not a cost saving in the traditional sense. It's a quality floor that doesn't degrade with volume, and it's the kind of thing that's genuinely difficult to buy with headcount, no matter the budget.

Where the Human Time Actually Goes Now

The point was never to remove HR from onboarding, it was to redirect their time toward the parts of the job that actually need a person: the confidence-flagged questions that need judgment, the moments a new hire seems uncertain or uncomfortable, the exceptions that don't fit a script. HR still sits on every single call, live, the whole time. What's changed is the ratio, less time narrating the same script for the fortieth time, more time available for the two or three minutes per call where a human genuinely matters.

That's the actual return on investment here, not headcount eliminated, but attention reallocated to where it does the most good.

---

MITS Global Consulting has spent 24 years building efficient, scalable IT operations for Fortune 500 clients across banking, insurance and fintech. See how that same discipline applies internally at https://www.mitsit.in/