Rad AI connects radiology reporting with follow-up work

Rad AI applies AI to the work surrounding a radiology report: composing it, checking it and helping teams act on recommendations. Its products address distinct steps in that process. A department considering the company should establish which step creates the bottleneck before treating an AI reporting purchase as a solution to the entire workflow.

01 / ProductThree products address different points in radiology work

The Rad AI portfolio comprises Reporting, Impressions and Continuity. The distinction matters because a group replacing its reporting environment has a different implementation problem from one adding impression generation to an existing system. Follow-up coordination introduces another team and another set of success measures. A clear purchase specification should name the relevant products and the workflow each will own.

Reporting is a cloud reporting environment with speech recognition, templates, dictation and AI-assisted report composition. The product page describes features including real-time quality checks, calculation support and carrying forward unchanged findings. These capabilities concern the production of the report. The page is not evidence that an algorithm has independently interpreted every medical image or can replace the interpreting radiologist.

Impressions generates draft impression language from dictated findings while adapting to a radiologist’s style. It is presented as an addition to existing reporting workflows, including established voice-reporting systems. That provides a narrower adoption route: a department can investigate draft conclusions without assuming that it must simultaneously replace the environment in which all reports are created.

Continuity identifies follow-up recommendations in reports and supports worklists, communication and tracking. It addresses the gap between writing a recommendation and resolving the resulting task. A report can be clinically complete while the follow-up process remains incomplete; treating these as separate events makes both the product’s role and the organization’s responsibilities easier to assess.

02 / AudienceDifferent teams need different evidence of value

For a radiology group, the primary question may be whether report composition becomes less burdensome without reducing accuracy. For a health system, the priority may be consistency across subspecialties or integration with existing systems. A navigator team may care most about a reliable queue, clear ownership and visibility into patients who obtained care outside the organization. These are related goals, but one demonstration cannot establish all of them.

The FAQ says Impressions learns from de-identified reports rather than images and describes personalization using historical reporting material. It also makes radiologist review, editing and signing central to the process. Before choosing a product, assess whether the department can supply an appropriate corpus and whether its existing report language is something it actually wants to reproduce. Personalizing inconsistent habits could make inconsistency more efficient.

Abridge provides a useful adjacent comparison for clinical documentation from encounters. Aidoc is relevant when the evaluation concerns image analysis and care coordination. Rad AI’s report-language workflow should be compared on its own inputs and outputs. A claim about document generation does not demonstrate performance on diagnostic image detection, and vice versa.

03 / WorkflowA proposed pilot measures complete work, including corrections

Start by selecting one bounded reporting workflow and representative cases approved for evaluation. Include ordinary reports as well as examples with negation, multiple findings, measurements, laterality and follow-up language. Record the current process before introducing generated drafts. This is a proposed assessment, not a claim that Sequenced has used the software or measured its effect.

For each case, compare the dictated findings, generated impression and signed report. Have the responsible radiologist identify omissions, unsupported additions, changed certainty, inconsistent measurements and inappropriate recommendations. Preserve the difference between a harmless stylistic edit and a clinically meaningful correction. An average edit count can conceal a small number of consequential errors.

Measure time through the final approved report. Faster draft generation is useful only if review and correction do not consume the saving. Separate initial dictation, draft wait, editing and final sign-off where feasible. Compare like cases and account for the learning period. A complex examination and a routine follow-up study should not be combined into an apparently precise productivity figure without context.

Test integration at the handoffs. Confirm that the correct patient and examination remain associated with the report, that interrupted sessions recover appropriately and that signed content reaches the intended destination. Ask the implementation team to demonstrate normal work as well as missing information, disconnected systems and corrected reports. A polished composition screen cannot establish the reliability of the complete reporting pathway.

If Continuity is included, create a separate follow-up evaluation. Track a recommendation from extraction through review, assignment, outreach and resolution. Distinguish scheduled, completed, declined, clinically cancelled and outside-system follow-up. Those outcomes should not be collapsed into a single contacted status. Review ambiguous recommendations with the clinical owner instead of silently converting uncertain language into a definite instruction.

Agree on acceptance criteria before expanding. The reporting pilot might require acceptable error patterns and manageable review time; the coordination pilot might require clear queue ownership and complete status records. Keep the criteria tied to the problem the department wants to solve. Vendor adoption statistics and selected customer results are context, not substitutes for evidence about a new deployment.

04 / PricingCommercial access starts with a scoped discussion

The public pages provide a demo-request route, rather than a numerical subscription tariff. As of 13 October 2026, the reviewed materials do not establish a standard per-radiologist, per-report or enterprise price. A quote should specify which products and sites are included, implementation responsibilities and the assumptions behind any usage or support charges.

Request a written account of what the organization must provide. Historical reports, interface work, clinical review time and staff training can affect the cost and duration of implementation even if they do not appear as a software line item. Compare the full operational commitment across proposals, using the same products, sites and rollout assumptions.

05 / DistinctionsThe distinction is continuity across the report lifecycle

Rad AI’s most useful organizing idea is that reporting and follow-up are connected but independently assessable. The physician’s conclusion is a language task grounded in the examination and findings. Acting on a recommendation is a coordination task involving people, appointments and system boundaries. The portfolio offers products on both sides, without making either task disappear.

The security materials describe measures including de-identification and SOC 2 Type II reporting. Treat these as supplier representations to examine in the relevant documentation. Establish which data each selected product processes, how long it is retained, who can access it and which agreements apply. The appropriate review depends on the actual deployment and information flow, not solely on a compliance badge.

Style adaptation is another distinction worth evaluating carefully. A draft that resembles an individual radiologist’s writing may reduce editing friction, but familiarity can also make errors less conspicuous. Include deliberate review of the clinical meaning in the pilot. A convincing voice is a usability benefit only when the content remains faithful to the dictated findings.

06 / QuestionsResolve ownership and failure handling before rollout

What happens when a draft is wrong?

Specify how the user corrects the text, whether feedback affects future behavior and how a recurring error is escalated. Ask for a practical demonstration using the organization’s reporting conventions. The desired outcome is a clear correction path, not a promise that generated language can never require intervention.

Who owns an unresolved follow-up?

Continuity can expose tasks, but the health system still needs an accountable owner and an escalation process. Define how responsibilities transfer between radiology, referring clinicians and navigation teams. Include patient preferences and care delivered elsewhere in the resolution model, so the queue reflects the actual clinical process.

Which integration claims are included in the proposal?

Compatibility at a brand level does not establish the exact interface version, configuration or workflow at a particular site. Obtain an implementation plan naming systems, permissions, testing and support boundaries. Public product descriptions cannot settle these local details or the time required to complete them.

07 / DecisionChoose the product for the bottleneck you can measure

Rad AI is a substantial radiology-focused AI company with a differentiated portfolio for reporting and follow-up. Its suitability depends on whether the selected product improves a defined task while preserving accountable review. Begin with the workflow, measure the full effort and make the commercial scope match the operational problem.