What structured-report automation is actually worth
Every ultrasound and bone-density exam carries a quiet data-entry tax. The way to price automating it is not a vendor list price — it is the labor you stop paying.
2–3 minutes of skilled time, per exam — plus the occasional transcribed typo.
Extracted, normalized, and delivered in seconds — a fraction of the labor it replaces.
Ask a practice owner what their reporting workflow costs and they will name the software line items. Ask what it costs to get the measurements into that software, and you will usually get a shrug. That gap is the whole conversation. The measurements from an OB ultrasound, a biophysical profile, or a bone-density study rarely arrive in the report on their own. Somebody keys them in — and that somebody is often the most expensive person in the room.
Automating that step is easy to justify once you price it correctly. The mistake most buyers make is anchoring on what a vendor charges instead of on the work the vendor removes. So let us do the arithmetic the industry tends to skip.
The hidden line item
The measurement tax nobody budgets for
On a fetal ultrasound, the numbers that matter — biparietal diameter, head and abdominal circumference, femur length, estimated fetal weight, amniotic fluid index — are produced by the machine and then transcribed by a person. On a DEXA study it is BMD, T-scores, and Z-scores. On vascular and cardiac studies it is a different list, but the same story: a skilled technologist or the radiologist reads the values off a screen or a printed worksheet and types them into the report or the RIS.
It feels like nothing. It is two or three minutes. But it is two or three minutes of a person you pay $40 to $80 an hour, repeated on every applicable exam, every day, forever. That is the tax — invisible because it is bundled into a salary you are already paying, and precisely because it is invisible, it never gets optimized.
Annualize the minutes
Do the math you are skipping
“A couple of minutes” hides the number that matters. The number that matters is the annual one. Take a mid-sized imaging center whose sonographers do roughly 400 structured-measurement exams a week. At three minutes of hand-keying each, that is 20 hours a week of skilled time spent transcribing numbers that a machine already produced.
The exact figures will differ for your practice — plug in your own volume and pay rate. But the shape does not change: the labor cost of hand-keying measurements is a five-figure annual line for a busy center and a six-figure one for a hospital. Automation that costs well under a dollar per report is not competing with that number. It is a rounding error against it.
The second and third costs
Labor is only the part you can see
Two more costs ride along with manual transcription, and both are larger than they look. The first is error. A person copying a two-decimal measurement under time pressure will occasionally transpose a digit — an estimated fetal weight off by a factor, a T-score with the wrong sign. Most are caught. The ones that are not become amended reports, follow-up phone calls, and, rarely, something worse. Every avoided transcription error has a value that is hard to price and easy to underweight.
The second is turnaround. Transcription is a queue: measurements wait for a person to have a free moment. Automation removes the queue — the structured values are ready the instant the study lands, which pulls report turnaround forward and frees the technologist to scan the next patient instead of typing. In a market where referrers choose on turnaround, that has revenue value on top of the labor it saves.
RadSR exists to collapse all three costs at once: it reads native DICOM Structured Reports or OCRs the measurements straight off image-only devices, normalizes them, and delivers them into the reporting fields your team already uses — in seconds, without a typo. For the mechanics of how that works, see how RadSR normalizes structured reporting.
The anchoring mistake
Value it by the work it removes, not a list price
Here is the reframe that makes the buying decision obvious. Do not ask, “What does the vendor want per report?” Ask, “What was I paying to do this by hand, and what fraction of that am I being asked for now?” When automation costs a fraction of the labor it replaces — before you even count the errors avoided and the turnaround gained — the per-report price is almost beside the point. The only wrong move is continuing to pay the full labor cost out of habit.
This is also why usage-based pricing is the honest model for this category. You pay per report the system actually produces — not per exam received, not a flat license whether you use it or not. The cost tracks the value one-for-one: more reports means more labor removed means proportionally more that you would have paid anyway. A quiet month costs you less; it does not strand a fixed license.
The worry everyone has
The AI-cost question — and why it should not be the blocker
The one hesitation we hear most is about the AI itself: if this runs on a language model, does a runaway token bill land on me? It is a fair worry, and the answer is that you get to decide how the AI runs — three ways, each with a different cost posture:
The AI cost is baked into the per-report price. You never see a token bill or manage a key — one number, no surprises.
Run the model on your own hardware. No per-token cost to anyone, fully air-gapped, and no PHI leaves the building. The answer when API cost or data residency is the real concern.
Prefer your own AI contract and BAA? Use your key for a lower per-report rate and pay your provider directly.
With the managed option the AI cost is our problem, not yours — it lives inside the rate. With self-hosting it disappears entirely. Either way, the model never receives a full patient record: identifiers are removed, redacted, or tokenized before anything reaches it, hosted or local. The privacy design and the cost design are the same design — send the model only what the task needs. More on that in our note on AI architecture and privacy.
What we charge, and why
What a fair price looks like
We price this the way we would want it priced if we were the ones signing. The core interoperability — routing, worklist, query/retrieve, the viewer, disc burning — is the plumbing every site needs, and we keep it deliberately cheap; the incumbents overcharge for it, and it should not be where the money is. The value engine is the automation, and we bill it per report at a rate that stays a fraction of the labor it removes, dropping further as your volume grows.
And because we are new, today’s numbers are founding-customer rates. Founding customers keep their contracted rate while the subscription remains active and the organization stays within the same volume tier; added scope, infrastructure, third-party costs, and tier changes are priced separately. We would rather earn a long relationship at a fair price than a short one at the maximum. The full breakdown, with no “request a quote” wall, is on the pricing page.
The honest summary is this: the question was never whether structured-report automation is worth paying for. Your team is already paying for it — in salaried minutes, amended reports, and slower turnaround. The only real decision is whether to keep paying full price by hand, or a fraction of it to a system that does not get tired and does not fat-finger a number.
See the number for your practice
Bring us your volume — we will show you the math.
Tell us how many structured-measurement exams you run, and we will put the labor you are spending today next to what automating it would cost. Clear line items, no runaround, no obligation.