Hire when the work is varied, judgment heavy, or patient facing. Automate when the work is repetitive, rules based, and high volume. Run the math on the fully loaded cost of the hire against the build plus maintenance, and check what share of that person’s day the automation could actually cover.
Last updated: July 2026
Most of the time this gets decided by whichever option feels less scary that week. It is a solvable arithmetic problem with a judgment layer on top, and both halves matter. Skip the arithmetic and you overpay. Skip the judgment and you automate something that needed a person.
What Is the Real Cost of the Hire?
Not the salary. The fully loaded cost, which is the number that leaves your account.
| Cost component | Typical treatment |
|---|---|
| Base wage | The advertised number |
| Payroll taxes | Employer FICA plus unemployment |
| Benefits | Health, retirement match, PTO accrual |
| Recruiting and onboarding | One time, but real |
| Training ramp | Weeks to months at reduced output |
| Management overhead | Someone’s time, every week |
| Turnover risk | Recurring if the role churns |
A common planning rule is that fully loaded cost lands somewhere around 1.25 to 1.4 times base wage. This is a widely cited rule of thumb in workforce planning, not a hard statistic; the BLS Occupational Employment and Wage Statistics provides current wage data for your role and metro, and your own payroll numbers will be more accurate than any multiplier. Use those.
The cost people forget is management. An additional admin is an additional person to direct, review, and cover for. That is real time from someone whose time is more expensive.
What Is the Real Cost of the Automation?
Also not the quote.
- Build cost. Fixed price, hourly, or share of savings. See the model comparison in our cost breakdown for back office automation.
- Software and transaction costs. API usage, clearinghouse transactions, licensing. Recurring, usually modest.
- Maintenance. Anything touching a payer portal or a third party interface needs upkeep. Portals change.
- Internal time. Your team validating outputs and handling exceptions. This never goes to zero.
- Coverage gap. The part of the job the automation does not do, which you still have to staff.
That last line is where the comparison usually goes wrong. Automation almost never replaces a whole role. It replaces a slice of several roles.
How Do You Run the Break Even?
Here is a worked example. Every number below is illustrative, chosen to show the method, and reflects a rough 2025 to 2026 US mid-market picture. It is not drawn from a client engagement and it will move with your specialty, payer mix, region, and how many systems and portals are involved.
Assumptions for this scenario:
- Specialty: general practice or primary care, US mid-market
- Claim volume: sufficient to make eligibility, claim status, and AR follow up the primary daily bottleneck
- Region: mid-cost metro area
- What the automation covers: eligibility lookups via 270/271, claim status monitoring via 276/277, and AR queue prioritization, reading from and writing back to the practice management system, athenaCollector in this example, through a clearinghouse connection such as Availity or Waystar
- The percentage coverage figure (60%) is an assumption for this example; your mix will vary by task type and volume
Scenario. A small practice is drowning in eligibility checks, recall calls, claim status lookups, and AR follow up. The obvious move is a second front office admin.
Option A, hire.
- Base wage: 45,000 dollars
- Fully loaded at 1.3x: 58,500 dollars per year
- Recruiting and onboarding, year one: 3,000 dollars
- Year one total: 61,500 dollars. Year two: 58,500 dollars
- Two year total: 120,000 dollars
Option B, automate.
- Fixed price build: 32,000 dollars
- Software and transaction costs: 400 dollars per month, so 4,800 dollars per year
- Maintenance: 3,000 dollars per year
- Year one total: 39,800 dollars. Year two: 7,800 dollars
- Two year total: 47,600 dollars
The correction that matters. Say the automation covers 60 percent of what that new admin would have done. The remaining 40 percent still needs hands. If you cover it with part time help at 24,000 dollars fully loaded per year, Option B becomes 63,800 dollars in year one and 31,800 dollars in year two, for 95,600 dollars over two years.
Still cheaper than 120,000 dollars, but the gap narrowed from 72,400 dollars to 24,400 dollars once you accounted honestly for the coverage gap. Any comparison that skips that step is selling something.
Break even point. In this example the automation passes the hire during year two, not in month three. Be suspicious of anyone promising payback inside a quarter on a build of this size.
When Hiring Wins: A Second Scenario
The math above does not always favor automation. Here is a scenario where hiring is the cleaner answer.
Scenario. A three provider specialty practice is growing fast. Referral volume doubled in the last six months and the front desk is behind on incoming referral coordination, prior auth follow up, and specialist scheduling. The work varies by referral type, requires payer-specific judgment, and about half of it involves real-time phone conversations with specialists’ offices.
Why automation loses here:
- Volume is growing but unpredictable, meaning the automation scope would need frequent rework
- The work is heavily judgment dependent and involves conversations that cannot be scripted reliably
- The practice needs capacity now, not after a ten-week build
- The growth spike means the role definition will change within six months anyway
In this case, hire first. Reassess in six months once volume stabilizes and the task boundaries are clearer. A defined, repetitive slice of that role may be automatable at that point.
The general pattern: automation favors stable, high-volume, rules-based work. Hiring favors variable, judgment-heavy, or rapidly changing work and situations where you need capacity immediately.
Which Signals Point Toward Hiring?
Hire when the answer to most of these is yes:
- The work varies day to day and cannot be described as a repeatable procedure
- The work is patient or client facing in a way that affects the relationship
- It requires judgment about exceptions more often than it follows a rule
- Volume is low enough that a build will not pay back
- You are growing fast enough that you need capacity now, not in ten weeks
- Nobody internally has the bandwidth to own an automation project
- The role also absorbs the unpredictable work nobody planned for, which is most front desk roles
That last point is underrated. A good admin is a general purpose absorber of chaos. Automation is not.
Which Signals Point Toward Automating?
Automate when the answer to most of these is yes:
- The same task runs dozens or hundreds of times a week with the same steps
- Following the rule correctly matters more than judging the situation
- The work is currently not getting done at all, which is common for recall and AR follow up
- The bottleneck is at nights and weekends when nobody is staffed
- You have already hired for this once and the person is still behind
- The task is a lookup, a status check, a form fill, or a queue build
- You can define the exception path and who owns it
The strongest signal is the third one. When the alternative is not a person doing it well, but nobody doing it at all, automation is not replacing labor. It is recovering revenue that is currently walking out the door.
What About Doing Both?
Usually the right answer, and rarely the one presented.
The realistic pattern is automating the repetitive volume so the person you already have stops spending sixty percent of their day on lookups, then hiring only if there is still a gap after that. You often discover you did not need the second hire, or you needed a different hire than the one you were about to make. That sequencing costs nothing to try first.
Where This Does Not Work
The math above assumes you can measure the current state. If you do not know how many hours go to eligibility checks or how large your unworked recall list is, you cannot compute either side of the comparison and you are guessing with a spreadsheet. Measure for two weeks first.
It also does not work when the real problem is process, not capacity. Some practices are not understaffed. They are running a workflow that creates rework, and both a hire and an automation would just process the rework faster. If your denial rate is high because intake captures bad data, neither option fixes it.
And automation is a poor answer when the role’s value is relational. Patients notice. Clients notice. There are seats where a competent human being is the product.
One more constraint that changes the math in healthcare specifically: any automation touching PHI has to run under a Business Associate Agreement, and the compliance review is real work with a real cost. A new hire signs an employment agreement and starts. Put the BAA and security review time on the automation side of the comparison, not off the books.
You can see how this gets scoped on our medical billing page and under solutions. No rip and replace, and no requirement to change the systems your team already uses, whether that is athenaCollector or Epic in a medical practice, Open Dental or Dentrix in a dental office, Clio in a law firm, or QuickBooks on the accounting side. Automation reads from those systems and writes results back into them.
FAQ
Is AI automation cheaper than hiring an admin? Often over a two year window, but not always, and rarely in the first few months. Compare fully loaded employment cost against build plus maintenance plus the staffing you still need for the part automation does not cover.
How long until an automation build pays for itself? For a build in the low tens of thousands against one fully loaded role, payback commonly lands somewhere in the first to second year, as a general pattern rather than a measured statistic. Payback promised inside a single quarter usually means the build is small or the savings claim is soft.
Will automation replace my front desk person? Typically no. It removes a slice of repetitive work from several roles rather than eliminating one. Practices that plan for full replacement usually end up rehiring.
What if I do both? That is frequently the best sequence. Automate the repetitive volume first, then reassess whether the additional hire is still needed. You often find the requirement changed.
How do I know if my volume is high enough? Track how many times per week the candidate task runs and how long each instance takes. If total time is under a few hours a week, a build is unlikely to pay back and better configuration of existing software is the cheaper fix.
If you want the break even run against your actual numbers, book the free 30 minute Waste Audit. If the math says hire, that is what you will hear.