How Carolinas Healthcare saved $203,000/year — and tripled their interpretation volume

The headline numbers

Between 2015 and 2016, Carolinas Healthcare System — one of the largest non-profit health systems in the Southeast US — deployed 400+ video remote interpretation (VRI) devices across 100+ locations.

The results, published in an HFMA case study:

• Interpretation minutes served: 370,000 → 915,000 (more than 2.5× growth)
• Net annual expense reduction: $203,000

Read that twice. They served roughly 2.5× the volume of patient-interpreter sessions, and paid $203k less per year doing it. The conventional wisdom in healthcare ops is that more service = more cost. Carolinas inverted it.

How they did it

The hospital had been buying interpretation per-minute from human interpreter services — the industry standard model for decades. Major VRI vendors at the time (LanguageLine, Stratus, Boostlingo) charged in the $1.95–$3.49/minute range. For a hospital system seeing ~370k minutes/year of demand at, say, $2.50/min, that's $925,000 annually in interpretation costs alone — and that's only the minutes they could actually fund.

LEP (Limited English Proficiency) patients beyond that budget got served less well: longer wait times for an interpreter, fewer interpreter-attended encounters, more reliance on "ad hoc" translation by family members (which the Joint Commission has been telling hospitals to stop doing for over a decade).

Carolinas' move: deploy a fixed-cost VRI platform with self-managed devices. Instead of $2.50/minute, they paid roughly $0.30/minute equivalent on a capitalized + leased fleet basis. That margin difference is what unlocked the math: the budget that previously bought 370k minutes could now buy 915k.

What the same math looks like in 2026

VRI vendor pricing hasn't dropped much in the decade since. Industry per-minute rates are still in the $1.95–$3.49 range for human-interpreted VRI. The labor cost of human interpreters is the floor.

What HAS changed: real-time AI translation between English, Spanish, and Portuguese is now accurate enough for clinical conversations — including medical terminology — at a fraction of the cost. The compute cost (GPT-4-class models for translation, ElevenLabs-tier voices for output) for a five-minute exam-room conversation runs to under five cents in raw infrastructure.

That doesn't mean you should pay 5¢/minute — there's product, support, compliance overhead — but it does mean a vendor charging $0.30/minute can have 6-10× margins while still being 10× cheaper than human VRI for the same conversation.

The healthcare CFO calculation now looks like:

• A 50-bed hospital seeing 50,000 LEP minutes/year
• At VRI rates ($2.50/min avg): $125,000/year
• At AI translation rates ($0.30/min): $15,000/year
• Annual savings: $110,000

Multiply that by 100 locations and you're at Carolinas' scale — and you're saving an order of magnitude more than they did with 2015 tech.

The piece nobody mentions: 20% readmission reduction

There's a sub-finding in the Health Affairs literature on this topic that doesn't get cited as often as the cost-savings number. A multi-hospital study found that LEP diabetes patients who received professional interpretation services had 20% fewer 30-day readmissions than those who did not.

For a Medicare-billed hospital, every 30-day readmission triggers a penalty (the CMS Hospital Readmissions Reduction Program). The financial value of a single avoided readmission is on the order of $11,000–$15,000. So the math on professional interpretation isn't just the cost savings vs. VRI — it's also the avoided readmission penalties.

If you're a CFO at a hospital serving a high-LEP patient mix, this is the equation:

(Cost of interpretation) < (Cost of NOT interpreting, including readmission penalties + dissatisfaction + risk)

Almost always. Sometimes by an order of magnitude.

What modern AI translation actually requires for healthcare

To pencil out for a hospital, an AI translation product needs to clear a few bars that don't apply in construction or hospitality:

HIPAA-ready architecture. Encrypted in transit + at rest. No PHI leaving AWS unencrypted. The vendor signs a BAA. (This is table stakes — any vendor unwilling to sign a BAA shouldn't be at the table.)

Configurable retention. Some healthcare conversations need to be retained for 7 years (HIPAA-friendly default). Others, the patient may want deleted after the encounter. The platform needs to let admin set retention per tier or per encounter.

Audit logs. Who joined the session, when, in what language, for how long. Compliance and risk management will ask for these and they need to be tamper-evident.

Integration on-ramp. Eventually, the platform needs to flow conversation summaries back into the EHR (Epic, Cerner, Athena). Not necessarily on day one — but the architecture should support webhook delivery on session-end so the customer's downstream pipeline can pull it.

These aren't expensive to build into a platform that was designed for healthcare from day one. They're prohibitively expensive to retrofit into one that wasn't.

What Carolinas would buy in 2026

If Carolinas were doing this procurement today rather than in 2015, the math suggests they'd:

1. Replace the 400 VRI devices with a browser-only AI translation platform (no hardware to maintain — each provider opens it on their phone or workstation)
2. Pay a fraction of the per-minute rate (~$0.30 vs $2.50)
3. Serve probably 3-4× the minutes (because cost ceases to be the rate-limiter)
4. Capture transcripts + AI summaries for every encounter (audit + readmission risk reduction)
5. Spend the saved ~$700k/year on either staffing, technology elsewhere, or LEP outreach programs that drive even more patients to the network

That's the trajectory of this tech. Healthcare went through it with VRI in the 2010s. AI translation is the next compression, and the savings are 5-10× larger.

---

VoiceBridge is a browser-only real-time translation platform with HIPAA-ready architecture, configurable retention from 90 days to 7 years, and audit logs on every session. Talk to us about HIPAA → or try a free conversation →.