AI Medical Scribe Cost: Buy vs Build

Short answer: reported pricing for ambient AI scribes clusters between roughly $300 and $600 per clinician per month, with setup fees and annual commitments on top. None of the vendors publish it. At around 40–60 clinicians the cumulative licence cost passes what a custom build costs to deliver and run, and after that the gap widens every year.

Nobody publishes their pricing, which is not an accident. Enterprise sales models depend on the quote being shaped by your headcount, your EHR and how badly you need it this quarter. That is fine — it is how most clinical software is sold — but it makes budgeting genuinely hard, and it makes the build-versus-buy conversation almost impossible to have honestly.

So here is what is actually reported, where the numbers come from, and how to think about the comparison.

Per-clinician pricing is what makes the build-versus-buy question arithmetic rather than ideology once a practice passes a certain size.

What the vendors reportedly charge

Everything in this table is from third-party reporting and reseller disclosure, not vendor price lists. Treat it as a range to open a negotiation with, not a quote.

Vendor	Reported per clinician / month	Notes
Nuance DAX Copilot (Microsoft)	~$450–600 at small volumes	Tiered down by headcount; setup fees reported around $650 first user, $250 each additional; 12-month minimum on some plans
Abridge	~$208 reported, with implementations quoted far higher	Not sold to individual clinicians or small practices
DeepScribe	~$350–750	Contract-data derived
Suki	~$299–399+	Quote-based, typically annual

Two things stand out. First, the spread within a single vendor is wider than the spread between vendors, which tells you the list price is close to meaningless and the negotiated price is everything. Second, DAX pricing has moved a lot recently — reported list figures were cut sharply in 2026 — so any number more than a few months old is stale.

If it is specifically Dragon pricing you are pinning down, the DAX Copilot breakdown goes into the tiering, setup fees and contract terms in more detail than belongs here.

Ask every vendor for the setup fee, the minimum term, and the price at your headcount rather than the headline. The headline is the least informative number in the quote.

The part that gets underestimated: it compounds

A per-clinician licence is a recurring cost that scales with the thing you are hoping to grow. That is the structural problem, and it is easier to see as arithmetic than as an argument.

At a reported $500 per clinician per month:

Clinicians	Per year	Over 3 years
10	$60,000	$180,000
25	$150,000	$450,000
50	$300,000	$900,000
100	$600,000	$1,800,000

Volume discounts flatten this somewhat — most vendors tier down meaningfully past 25 and again past 75 — but the shape does not change. You are renting, the rent tracks headcount, and it never stops.

A custom build has the opposite shape: a one-time engineering cost, then hosting and maintenance that are roughly flat regardless of how many clinicians use it. Inference cost per encounter is real but small, and falling.

Where the lines cross

For a scribe that produces genuinely useful output — ambient capture, speaker diarisation, structured extraction, form population, a clinician review interface, and EHR integration — a realistic build is a few months of a small senior team. Call it a mid-six-figure programme for something production-grade with the compliance work done properly, plus ongoing hosting, inference and maintenance.

Set that against the table above and the crossover lands somewhere around 40 to 60 clinicians on a three-year view. Below that, buy. Above it, the arithmetic starts favouring a build, and it improves every year you keep operating.

But the crossover point is not the whole decision, and treating it as one is how organisations end up building something they did not need.

The four situations where building is right regardless

Headcount arithmetic is the weakest reason to build. These are the strong ones.

Your documentation burden is forms, not a note. This is the big one. Products are optimised to produce a good clinical note, usually SOAP-shaped, because that is the common denominator across their customer base. If your actual burden is a pile of mandatory structured forms specific to your setting — and for a lot of specialist and public-sector providers it is — then a beautifully generated note leaves most of the work exactly where it was. We built a scribe for a healthcare client that auto-fills more than ten mandatory forms per consultation from a single encounter, and the de-duplication across those forms is where the hours actually came back. No product on the market does that for their form set, because their form set is theirs.

Your EHR is not on the vendor’s integration list. DAX integrates beautifully with Epic. If you are on Epic, that is a strong argument for buying. If you are on something regional, older, or homegrown, you will be quoted an integration project anyway — and at that point you are paying custom-build prices for a product you still do not control.

Data residency rules out the vendor’s infrastructure. If audio containing PHI cannot leave your jurisdiction or your own boundary, most of the market is eliminated before pricing enters the conversation. This is a hard constraint, not a preference, and it decides the question on its own.

Per-encounter pricing has become your dominant line item. Some contracts price per encounter rather than per clinician. That is worse for high-volume settings, because your cost now scales with the exact activity you are trying to make more efficient.

Where buying genuinely wins

We build custom scribes and we still tell most people to buy one. Be honest about which side you are on.

If your burden is a clinical note, buy. Abridge, DAX and the rest solve that well, you will not beat them on price, and you will not beat them on time-to-value — days rather than months. If you are on Epic, buy. If you have fewer than about 25 clinicians, buy; the arithmetic is not close. If you need it working this quarter, buy, because no custom build ships that fast.

And if you have no internal engineering capacity to maintain what gets built, buy. A custom scribe is a production clinical system. It needs someone to own it after launch, and an unmaintained one degrades quietly — the model provider deprecates a version, an EHR API changes, accuracy drifts on a new intake pattern and nobody notices for a month.

The hybrid nobody quotes you

There is a middle option worth asking about, and it rarely comes up because neither side of the market is incentivised to raise it.

Buy transcription as a commodity — it genuinely is one now — and build only the extraction and form-population layer on top. Transcription APIs cost cents per hour of audio. The expensive, differentiated engineering is downstream: deciding which utterance belongs in which field of which form, scoring each field’s confidence, and putting a review interface in front of a clinician that is faster to check than to re-read.

This is often the right shape when your forms are unusual but your audio is not. You avoid rebuilding a solved problem, and you spend your engineering budget on the part that is actually specific to you.

How to run the comparison properly

Ask each vendor for the price at your headcount, the setup fee, the minimum term, and what happens to the price on renewal. Get the renewal answer in writing — the first-year discount is frequently the whole negotiation.

Then run the same encounter through each system and count what a clinician still has to do afterwards. Not accuracy on a demo recording: your audio, your accents, your specialty vocabulary, your forms. A system that is 96% accurate on a transcript but leaves eleven fields to fill in by hand has saved less time than one that is 92% accurate and files nine of them.

Then price the build against your real form set, not against a generic scribe. Most of the cost sits in form coverage and EHR integration, and both are specific to you.

If the answer is buy, buy. That is the advice we give most of the time.

We build custom AI medical scribes and clinical AI systems, including one auto-filling 10+ mandatory forms per consultation. If you want an honest read on whether building is right for your situation, get in touch — we will tell you if a product solves it.

What AI medical scribes actually cost, and when building your own gets cheaper