Most healthcare practices don't know what it actually costs them to collect a dollar of patient responsibility. They have a feeling. They know it's “too high.” They know the front-desk team “spends a lot of time on it.”
The number is usually worse than they think. And the industry benchmark published by HFMA — 2–3% best-practice — is generous if you count the full fully-loaded cost honestly.
This post shows you the math that nobody publishes, using real numbers from my own practice's 123-day post-launch measurement window. The short version: we went from something in the 8%+ range on a legacy platform down to 1.39%on an AI-first stack. Here's how the arithmetic actually works.
What “cost to collect” actually means
Per HFMA, cost to collect = total cost of running revenue cycle ÷ patient service cash collected. Simple ratio. The problem is what you include in the numerator.
Most practices count: the patient-billing platform subscription, the processor fees, and the direct payroll for people in billing roles.
Most practices don't count: front-desk time spent on phone collections, the owner time spent reviewing aging reports, the time writing off old balances, the cost of patient complaints that bleed into online reviews, the opportunity cost of staff not doing something more valuable, and the contingency fees paid to outside collection agencies on balances that age out.
If you include those — and you should, because they're real money — most SMB healthcare practices are running a true cost-to-collect between 6% and 12%, not the 2–3% that gets published in best-practice benchmarks.
The manual phone-and-mail model
Let's model a typical small practice collecting on ~$535K of patient balances over 123 days — the same gross I ended up at with the new platform, so we can compare apples-to-apples.
Manual model assumptions:
- 2 FTE billing clerks at $25/hr fully-loaded (BLS OEWS 2024 base + benefits burden)
- 20 minutes per account touched (one call attempt + follow-up note)
- $3.00 per paper statement (print + materials + handling)
- $0.73 per USPS postage
- ~2,900 accounts touched across 123 days
The straight labor cost is 2,900 × 20 min × $25/hr ÷ 60 = $24,167. Add statements and postage (2,900 × $3.73) = $10,817. Add ~15% overhead burden on the labor = $3,625.
Direct cost to run a manual 2-FTE operation on this volume: ~$38,609, or 7.2%of the $535K collected. And that's assuming those clerks are 100% utilized, which they aren't.
The AI-first model
Same gross collections ($535,623), same 123 days, same patient panel. Here's what it cost to run the AI-first platform:
- SMS outreach: 8,636 messages × $0.0075 = $65
- Email outreach: 4,318 × essentially free (AWS SES) ≈ $2
- Voice AI inference (Retell AI): ~2,900 calls × ~$0.70 avg = $2,030
- Human follow-up queue (only exceptions): 216 hrs × $25/hr = $5,400
- Total: $7,426
$7,426 ÷ $535,623 = 1.39%. That's it. No fine print.
Notice what I'm NOT including in the 1.39%: the platform subscription itself (this was homegrown, not SaaS — if you license a platform like this from someone, add 1.5–3% on top), nor the credit card processing fees (which are passed to the patient in my setup and net out revenue-neutral). If you lump in a fully-loaded platform license, honest cost-to-collect lands in the 3–4% range. Still half the manual number.
The contingency-fee alternative
Some practices skip collections infrastructure entirely and ship aging balances to an agency on contingency. Industry-standard contingency for patient-responsibility balances is 25–40% (per ACA International benchmarks), rising to 50% for balances over 2 years old.
Math: if you place $200K of aging balances with an agency and they collect 35% of it at a 35% contingency, you get ~$45,500. The agency keeps ~$24,500. Or: you paid $24,500 to collect $45,500, effective cost of 35%.
Compared to 1.4% on an AI platform that would have caught most of those balances before they aged out in the first place, the economics are brutal.
Why the 1.4% number surprises people
The reason most practices don't get anywhere near 1.4% isn't technology — it's that the true cost of the manual workflow was never calculated in the first place.
When a practice says “we can't justify the cost of an automation platform,” they're usually comparing a fully-loaded platform quote (3–4%) against a manual cost that they've systematically under-counted by half. The proper comparison is 3–4% against 6–12%, which has never been a hard decision.
The platforms that charge 3–4% of collections also tend to be the ones that don't publish their own cost structure honestly, which creates a market full of vendors pitching margin gains that are partially mirage.
What to do with this
If you're running a healthcare practice with meaningful patient-responsibility balances, do this exercise once:
- Pick a 90-day window of collections data.
- Count the total cash collected on patient responsibility.
- Honestly count the total cost — platform fees, processor fees, fully-loaded labor time, aged write-offs, agency contingency fees.
- Divide. That's your real cost-to-collect.
If the number is above 5%, you have leverage to apply. If it's above 8%, there's money you're leaving on the table that would fund the build or license of a better system in one quarter.
I published the full 123-day methodology and source data so you can verify the math yourself. If you want a 20-minute conversation about how the math looks in your specific practice, book a call. No pitch — just an operator's read on whether the economics work for your size.