Four areas decide the purchase. This page answers each one
Leadership wants capacity and metrics. The clinical staff wants less time on what is not medicine. IT and information security want to integrate without opening a new gap. Quality wants proof that the process happened. Each area asks something different, and a generic answer convinces none of them.
Median of 52 s from editor to signature and about 10 h returned per 100 reports, with the methodology open.
Dictation in radiological Portuguese with 3.06% WER, automatic structuring and authorship always with the physician.
API, HL7, DICOM, Agent and identity are paths subject to local validation; data flow documented per deployment.
Critical finding with deadline, confirmation and per-exam record — what the audit asks for as proof.
Platform purchases in imaging services rarely die for lack of clinical interest. They die because leadership did not see a number that justifies the investment, or because information security did not approve the integration surface, or because the clinical staff tested it and found it got in the way, or because quality asked how, in an audit, you prove the process happened — and nobody had the answer.
These are four distinct questions, and each one can reject the purchase on its own. That is why this page does not repeat a single pitch in four tones: it answers each area separately, with the measured number when it exists and with the phrase "we do not measure this" when it does not.
What Laudos.AI is, in one line: the layer between the reported image and the report that is signed, delivered and proven. It does not replace the PACS, does not replace the RIS and does not replace the radiologist — it occupies the gap between them, which today is where time is lost and where the trail disappears.
What you gain
For leadership: capacity without hiring, with a number that can be audited
The measure that supports the conversation is the median of 52 seconds between opening the editor and signing the report, with 54.6% of reports under one minute — production telemetry on 5,200 reports over a 30-day window. The mean falls between 3 and 5 minutes, pulled up by the tail of complex cases, and we publish both because the median alone would be the convenient number. Converted into operations using the mean, it comes to about 10 hours returned per 100 reports. What this is not: a promise that your service will reproduce these numbers. The event definition, the window and the limitations are open at /metodologia, and the calculation that matters is the one against your baseline.
For the clinical staff: less friction, without losing authorship
The radiologist dictates in clinical Portuguese and the report is structured into technique, analysis and impression as they speak. The word error rate measured on radiology dictation in Brazilian Portuguese is 3.06% after vocabulary. The classification arrives with the criterion cited and the locator, for checking rather than blind trust. The institution's template is versioned and governed, so what changes for everyone changes with approval and history. None of this transfers authorship: review, editing and signature remain a medical act, in compliance with CFM Resolution 2.454/2026.
For IT: integration on top of what is already running, without a complete replacement of what already exists
API, HL7, DICOM and Agent are distinct paths. The IT team validates system, version, events, fields, identity, return, security and contingency. Interoperability standards do not guarantee universal compatibility.
For information security: a small, documented surface
Regions, subprocessors, processing, persistence, logs, retention and backups are documented by function and deployment. Some subprocessors may process data outside Brazil. Identity, access, audit and other controls are validated within the contracted scope.
For quality and accreditation: the critical finding no longer depends on a phone call
The CRIT module turns the flag into an operational event linked to the exam: priority, contact, escalation by phone call when there is no confirmation, recorded confirmation and closure — with a trail at every step. The practical difference for audits is that the question 'what time was the requesting physician notified, and how do we know they received it' now has a queryable answer instead of depending on the memory of the shift. It is this record that supports the critical-result communication metric in accreditation processes, including JCI.
For everyone: what the number does not say, said before you ask
The 52 seconds measure time from editor to signature, not clinical quality or total TAT. The 3.06% WER comes from a controlled internal benchmark whose public sample is not yet documented. LaiBench v3.10 measures fidelity on 120 internal cases and is not external clinical validation.
- 01EVALUATION
A 30-day pilot in the real routine
Not in a demo environment: with your volume, your exam types, your template and your network. Before starting, the institution's current baseline is defined, along with which metrics decide — because a pilot without a baseline becomes an impression, and an impression does not approve an investment.
- 02INTEGRATION
IT designs the topology with us
Integration points, standards, a test environment and documentation for information security to evaluate. The pilot scope starts with one modality or one queue, not the entire service, so that any adjustment happens before scaling.
- 03OPERATIONS
The full flow, from dictation to proof
The exam comes in through the worklist, the radiologist dictates and reviews, the classification arrives with the criterion, the critical finding triggers the flow with SLA and the signed report goes back to the institution's system. Every step leaves a record.
- 04DECISION
Comparison against the baseline, not against the promise
At the end of the pilot, what is compared is the metric measured at your institution against your institution's baseline. If it does not improve what matters to you, the honest answer is not to proceed — and we prefer that to a deployment nobody uses six months later.
The institutional decision requires verifiable evidence on supervision, data, integration, incidents and critical-finding communication.
Honest questions
Will we need to replace the PACS, the RIS or the hospital ERP?
No. Integration is based on open standards — HL7 v2, FHIR and DICOM-SR — on top of the existing topology. The platform occupies the layer between the reported image and the delivered report; it does not replace the systems that support imaging and operations. IT takes part in defining the integration points from the start.
Where are our patients' data and the dictation audio stored?
It depends on the function and the configuration. The documentation identifies regions, subprocessors, purpose, persistence, logs, retention and backups. Some subprocessors may process data outside Brazil; the institution validates the flow before the decision.
How does this hold up under CFM 2.454/2026?
The AI is assistive: it structures, suggests and organizes, and the physician reviews, edits and signs. Authorship and clinical responsibility remain with the radiologist. The record that human supervision took place stays in the trail, which is precisely what the resolution makes it necessary to demonstrate.
What exactly backs the 52-second figure?
The time between opening the editor for that exam and signing, as a median, in a fixed historical window from April 9 to May 9, 2026, with 5,200 finalized reports. In the same measurement, 54.6% were under one minute. It does not measure total TAT, clinical quality, causality or future performance.
Do you compare voice accuracy with that of other systems?
No. The 3.06% figure comes from a controlled internal benchmark in Brazilian Portuguese. The sample and the normalization rule are not yet publicly documented, so the number is neither reproducible nor externally comparable.
And what if the pilot does not confirm the gain at our institution?
Then the number that counts is yours, and the recommendation is not to proceed. The measure that decides is your operation's against your baseline, not the one on this page. That is why the pilot starts by defining the baseline and metrics before any deployment.
How is the institution's template governed?
The institution's template is versioned, with approval and history: you can tell what changed, who approved it and when it took effect. This matters in services with many radiologists, where variation between reports usually comes from diverging templates rather than real clinical divergence.
How much does it cost for an institution?
For individual use there is a public price on the plans page. For institutions, the design depends on volume, number of sites, modalities and integration points, and the proposal is built on that scope. We do not publish an institutional range because it would vary more than it would help — and a number that does not hold up is the opposite of what this page stands for.
Leadership, clinical staff, IT and information security, and quality. We show the integration architecture, the critical-finding flow with deadlines, the data residency model and the methodology behind each published number — and we design the 30-day pilot on your baseline.
