Why Radiology Software Locks You Into Long Contracts (and How to Add AI Without One)

Almost every piece of software a radiologist touches was bought on a multi-year contract by someone else. The PACS, the dictation system, the worklist, the AI triage module: each one arrived through procurement, each one came with a term measured in years, and each one is priced so that leaving costs more than staying. That model exists for reasons that made sense for hospitals. It makes no sense for the individual radiologist who just wants help with tonight's forty-finding restaging. This article explains why radiology technology is sold the way it is, what the lock-in actually costs you, and how to add capability to your day without signing anything longer than a month.

Why radiology software comes with a long contract

Enterprise radiology vendors are not being cynical when they ask for a five-year term. Their economics push them there. A PACS or dictation deployment involves on-site servers or a dedicated cloud tenant, HL7 and DICOM interfaces, a migration of prior studies, training for every user, and a support desk that has to know your site. The vendor spends most of that money up front and recovers it over the life of the contract. A short term would leave them holding the implementation cost, so the term is long, the renewal is automatic, and the early-exit clause is expensive.

The hospital accepts those terms because it is buying infrastructure. Nobody wants to re-platform the PACS every eighteen months, and a long contract with a known vendor is a reasonable way to get stability. The trouble starts when everything else in radiology gets sold on the same template, including tools that are not infrastructure at all.

The add-on problem

Over the past few years the enterprise vendors have bolted AI features onto their platforms: report drafting inside the dictation system, impression generation, prior comparison, worklist prioritization. The feature is new, but the commercial wrapper is old. To turn on AI drafting you typically need to be on the vendor's current platform version, commit to an add-on license for the whole department, and fold it into the existing term. A group that signed a five-year dictation contract in 2023 may find that the AI module it wants in 2026 is available only as a contract extension, priced per seat, for every seat, whether or not each radiologist wants it.

That is how a radiologist ends up in the strange position of reading about AI report drafting every week while having no way to try it. The capability exists. It is simply sold through a door that only a department chair can open.

What lock-in actually costs a radiologist

Lock-in is usually discussed as a budget line, but the costs that land on the individual are different from the ones that land on the finance office.

• You use the tool the contract picked, not the best one. The moment a department commits to one vendor's AI module, every other option is off the table for the length of the term. If a better tool appears next year, and in this field one will, you wait.
• Upgrades arrive on the vendor's schedule. A feature that needs a platform version bump waits for the IT change window, the validation cycle, and sometimes the next renewal. Months are normal. A year is not unusual.
• You pay in slow months as well as busy ones. Per-seat annual licensing costs the same during a week of chest radiographs as during a run of trauma pan-scans. The tool is priced for average use, so it is overpriced for everyone whose case mix varies, which is most of us.
• Your workflow is designed around the integration. Deep integration is the selling point, and it is genuinely useful when it works. It also means the tool's habits become your habits. Leaving is not just a licence question; it is retraining.
• Locums, contractors, and teleradiologists get nothing. If you read for three groups, you inherit three contracts, three sets of tools, and no say in any of them. The people with the most varied case mix have the least control over their software.

Infrastructure versus capability

The way out of this is to separate two things that the enterprise sales model bundles together. Some radiology software is infrastructure: it stores images, routes studies, captures dictation, and signs reports into the record. It has to be integrated, validated, and supported by the institution, and a long contract for it is defensible. Other software is capability: it helps you write a better impression, organize a long findings list, check your report for internal contradictions, or think through a differential. Capability does not need to live inside the PACS. It needs your dictated text and a way to give the result back.

Once you see the split, the question changes. It is no longer "which vendor's platform should the department commit to for AI." It is "keep the infrastructure that works, and add capability from wherever the best capability is, on terms you can walk away from." That is an ordinary software-as-a-service arrangement, the kind you already use for email or reference management, and there is no reason a reporting assistant cannot be bought the same way. Ordinary terms do not mean an ordinary tool. The assistant itself should be tailor-made for radiology: it should know what a findings section is, how an impression is prioritized, and where a report contradicts itself.

What a use-when-you-need-it tool looks like

RadPro AI is built on the capability side of that line. It is tailor-made for radiologists: designed around the shape of a radiology report rather than documents in general, and built for the sections, priorities, and consistency checks that a report needs. Its commercial terms are deliberately the opposite of an enterprise contract:

• Nothing to install and nothing to integrate. You dictate in whatever system you have, paste the text into a web app, and paste the polished result back. The optional Windows helper, RadPro Hub, turns the paste into a single keystroke without touching the PACS or dictation system.
• Month to month, on your own card. Ten free reports a month with no card on file. Premium is $99 a month, or $84 a month billed annually, with a 14-day free trial for first-time subscribers. No procurement, no minimum seats, no term.
• Pause when the case mix slows. A monthly subscription can be paused from inside the app and resumed when the complex cases come back. You should not be paying for reporting software during a week of routine work.
• Cancel from the account page. No retention call, no notice period, no exit fee. The exit cost is the thing to price first, and here it is zero.
• Built for the cases that need it. Trauma pan-scans, oncology restagings with multiple priors, multi-system inpatient studies. Use it on those and skip it on the two-view chest.

How to think about the next contract you are asked to sign

You may not control the PACS contract, but radiologists are increasingly consulted on the AI add-ons, and locums and small groups sign their own agreements. A few questions worth asking before any term longer than a month:

1. Is this infrastructure or capability? If the tool only needs text in and text out, ask why it needs a multi-year term at all.
2. Can each radiologist opt in or out? Per-seat licensing for the whole department means the interventionalist pays for a tool the body imager uses.
3. What does leaving cost? Early termination fees, data export, and the retraining time to switch. If the answer is vague, assume it is high.
4. What happens when something better appears? AI reporting tools are improving month by month. A term that outlasts the technology is a bet against progress.
5. Is there a way to trial it on real work first? A demo on the vendor's sample cases tells you nothing. Thirty minutes on your own de-identified cases tells you everything. The buying guide for individual radiologists has a step-by-step evaluation.

For a longer checklist written for the person who actually has to sign, see what to check before signing a radiology AI contract.

FAQ

Is a tool without PACS integration really usable?

On simple cases, no, and it should not be used there. On complex cases the paste is a small price for a structured report and a synthesized impression, and the Windows helper reduces it to one key. The honest position is that a category of tool that never touches the PACS will never match the convenience of a deeply integrated one; in exchange, you can start tonight and stop whenever you like.

Does my department need to approve a month-to-month tool?

Not the purchase. What you must follow is your institution's data policy about which services may receive patient information. RadPro AI detects and redacts PHI before any text reaches a model, offers a BAA on request to covered entities, and logs access, but a BAA between you and a vendor does not override the rules of the hospital whose patients you read. The HIPAA article covers what to check.

Will a month-to-month tool still be around next year?

That is the right question to ask of any small vendor, and the right answer is that the exit is free. If a month-to-month tool disappears, you lose nothing but the habit. If a five-year contract's vendor stalls, you are still paying.