AI vs. VAs vs. In-House: How to Get the Most Out of Your Eligibility Department
Picture a vendor sending you a quote for AI eligibility checks. It might be priced per check, per visit, or per provider per month. However it's built, the number looks reasonable.
Now try to answer one question: what do those same checks cost you today?
If you can't, you're in good company. Plenty of practices know what the eligibility department costs in total, but not what one check costs. Without that unit cost, you can't tell whether a quote is a bargain or just a new way to pay for the same work.
That question shaped how we approached AI at the multi-location practice we help run. Over the past several years, our eligibility department has changed shape twice, and we're in the middle of the third change. Each one was a different answer to the same question: who should do this work, and what should it cost?
We'll walk through all three, because the third is where the math gets less straightforward, and where we think it's easiest to overpay.
Here's the path in short:
Stage one: an in-house team of stateside employees.
Stage two: the same team, replaced one for one with virtual assistants (VAs).*
Stage three: AI agents handling the standard checks, with a smaller, dedicated VA team acting as the human in the loop. That means the people who supervise the AI, check its work, and handle whatever it can't.
*Virtual assistants are remote workers based outside the U.S., often in countries like the Philippines. You can hire them directly or through a third-party agency that supplies the labor, at a much lower cost than U.S.-based employees. HIPAA doesn't prohibit offshore access to patient data, but it still requires a business associate agreement, safeguards, and minimum necessary access, and some payer contracts and state rules restrict where patient data can be handled. Check your contracts and consult legal counsel before you start.
Why we started with eligibility
On paper, eligibility is the easiest department to measure. Every visit gets the same task: confirm the patient's active coverage and benefits before they walk in. The work is repetitive, high volume, and easy to count, so the unit economics are easy to see.
A note on the numbers: they're rounded, and they come from a cost model built on hourly pay rates and measured task times, not from our actual payroll records.
Stage one: the in-house team
We started with stateside employees and full benefits. For a clean example, picture eight seats (full-time positions) at about $18 an hour plus a 20 percent benefits load. That's a benefits-loaded rate of roughly $21.60 an hour, or about $360K a year.
This setup works. It's also one of the most expensive ways to handle a task that's largely repetitive.
Stage two: virtual assistants, one for one
Next, we replaced those seats with virtual assistants, one for one, at $12.50 an hour with no benefits load. Eight seats now cost about $208K a year. That's roughly $150K less, a cut of about 40 percent.
That rate is on the high end. We went through a third-party agency, which adds its own premium for managing the relationship. Hiring directly in countries like the Philippines can bring the rate down to roughly $6 to $8 an hour, so we treat $12.50 as a conservative estimate.
We made the change over roughly a year and a half, mostly through natural attrition, with some layoffs along the way. We didn't just swap people, either. We fixed a handful of processes as we went, which let us run a leaner staff-to-provider ratio as the practice grew.
If your department still runs on stateside employees with benefits, which describes a lot of practices, this is one of the biggest moves available to you, and it doesn't involve AI at all. It's a proven first step toward running a department leaner with the same results. It's a good place to start even if you're not ready to trust AI yet.
Stage three: add AI, but measure first
Stage three is where vendors are focused right now: AI agents doing the checks. Before we talked to any of them, we built a short internal brief on our own department. The idea is simple. Vendors price their work in different ways, so you need to price your own team in each of those ways.
There are two that matter most:
Cost per visit. Total department cost divided by the appointments that actually need a check. Benchmark this against a vendor quoting per visit, as long as their scope covers everything the department does, not just the checks.
Cost per check. The unit cost of one check, built from how long it takes and what that time costs. Benchmark this against per-check pricing.
Other structures exist, like flat monthly fees. The principle holds: convert your costs into the vendor's pricing unit so you're comparing like with like. It matters for another reason, too. A vendor's per-check fee is a variable cost, because it rises and falls with volume. A team is closer to a fixed cost, because you pay for the seats whether volume is high or low.
For us, a standard check takes about 2.4 minutes of VA time. At $12.50 an hour, that's roughly 50 cents in direct labor (a simplified figure that leaves out management time, overhead, and tool costs). The same team also handles front desk requests, like "can someone check this patient," "this doesn't look right," or a patient with a question. Those run about 7 minutes each, closer to $1.50.
The surprise: eligibility isn't the same job at every practice
AI vendors often start from one assumption: that the eligibility department is the standard check, run over and over. That's the workflow their product automates, so it's the scope behind the pitch. When a vendor says they can automate 90 percent of eligibility, they're typically describing 90 percent of that workflow, not 90 percent of your department.
When we measured our own team, the standard check turned out to be only part of the job. Checks came out to roughly half of paid time. The other half was everything around them: helping the front desk, handling last-minute requests, resolving results that didn't look right, and answering patient questions. That's the work mix, and at our practice "eligibility" covers far more than checks.
That gap limits what AI can replace. In our department, even a vendor that completed every check perfectly would free up about half the hours. At an 80 to 90 percent automation rate, meaning the share of checks completed without a human touching them (sometimes called the touchless rate), it's closer to 40 percent. Whatever the AI doesn't finish lands back on your people, and so does every conversation.
It also changes how you read a quote. Say a vendor charges 45 cents a check. (That's a placeholder, not a real quote.) Our VAs cost roughly 50 cents a check in direct labor, a simplified figure, so on that basis the AI would be only about 10 percent cheaper per check. The savings come from needing fewer people, and the work mix caps how many.
Your department may look different from ours. If your team spends nearly all its time on standard checks, with little patient or front desk interaction, the case for AI is stronger for you. We'd be surprised, but it's possible. Either way, you can't know until you measure the work mix, the time each task takes, and what that time costs. Do that before the demo, not after it.
It matters even more for larger groups. A wrong assumption about the work mix gets multiplied across more seats, locations, and workflows, and a demo showing AI clearing 80 percent of a department's volume is persuasive only if you know what the department actually does.
What stage three looks like for us
Our plan is to take the team from eight down to about five VAs and partner with an AI company that has real depth in eligibility and payer rules, the same test we described in our post on AI wrapper vendors.
That team becomes the human in the loop. Instead of running the checks, they supervise the software: reviewing what it flags, spot-checking what it doesn't, and resolving the exceptions it escalates. AI handles routine lookups well but is weaker on edge cases like coordination of benefits or mismatched patient data, which are the ones that cost you money when they slip through. They also keep covering the front desk and patient conversations. Keeping people in the process is consistent with what other leaders report: McKinsey's 2025 survey of care delivery leaders found them pairing technology with selective outsourcing, and only 6 percent planned to reduce outsourcing because of technology.
Some vendors supply their own reviewers, but those answer to the vendor. Yours answer to you and know your payers.
We expect stage three to cut eligibility department costs by roughly another 25 percent, which would put the full journey at more than half off where we started. We'll share actual results once we have them.
It looks less dramatic than stage two because the biggest savings came first. A department still in stage one can go straight to stage three, with a small VA team as the human in the loop, and capture both steps at once.
How to run this at your practice
Depending on your size and comfort level, this can be a weekend project or a month of work. The steps are the same either way.
Price your team per check. Multiply the minutes one check takes by what those minutes cost you, benefits included if you have them, and add any per-check fees you already pay, such as clearinghouse or portal charges.
Map your work mix. Split standard checks from everything else your team does, and measure real task times for a couple of people over a week or two. Don't estimate.
Ask each vendor what their automation rate actually covers, meaning which tasks, and what share of them are completed without a human. Then ask how accuracy is measured and who handles the rest.
Convert your numbers to their pricing unit. If they quote per check, you need your cost per check. If they quote per visit, you need your cost per visit for the same scope of work.
Eligibility is one department. We're also working through prior authorization and referrals with the same approach, and we plan to share similar studies on both in future posts.
Want help working out what one eligibility check actually costs your practice?