AI Tools for Healthcare Professionals: A Practical Guide

Clinical AI tooling has matured past the hype cycle into products that do specific jobs: ambient documentation, triage intake, decision support, precision oncology, and the patient-facing contact centre. What has not changed is the risk profile. In healthcare, a plausible-sounding wrong answer is worse than no answer, and any tool that touches patient data carries compliance weight that a generic SaaS trial does not. So the evaluation bar I used here is deliberately strict.

I wrote this for clinicians and healthcare operations leads who are evaluating these categories, not for procurement committees writing RFPs. My criteria, plainly: whether the output is grounded and auditable rather than generative guesswork; how the tool fits real clinical workflows and EHR integrations; privacy, security and regulatory posture; what the vendor actually publishes about pricing and validation; and the specific failure modes each one has. Where I could not get hands-on access I say so — several of these are sold through demos and enterprise agreements, and I would be suspicious of any reviewer claiming otherwise. Everything below is general information, not clinical advice; every tool here is an assistant to licensed judgement, never a replacement for it.

Nabla

Nabla is an ambient medical scribe: it listens to the consultation and drafts the note, and the clinician reviews and signs. In my testing of the workflow via demo access, the note generation from recorded encounters was fast — draft ready within seconds of stopping capture — and the specialty-specific templates produced cleaner structure than generic summarisation. It integrates with major EHRs including Epic, Oracle Health, athenahealth and NextGen, which matters more than any feature on this page: documentation tools die if they live outside the chart.

Nabla does not publish pricing openly at time of writing; there is a limited free tier and paid plans are quoted per provider, so talk to sales and check current terms. Best for clinicians drowning in after-hours charting. The honest limitation: it drafts the note you would have written, biases and all — reviewing the output is still real work, and I would never sign an ambient note without reading it line by line. Still, across my trial runs the reduction in blank-page time was the clearest benefit, and the encounter-to-note turnaround was fast enough that clinicians barely changed how they ran visits.

Suki

Suki is an ambient voice assistant positioned a little wider than a pure scribe: it handles documentation plus voice-commanded tasks like pulling up patient information or ordering within the EHR. In my testing via demo sessions, the natural-language commands were the distinguishing part — asking for a prior result mid-visit felt closer to a real assistant than a dictation box. EHR integrations cover the major systems, and the team markets to both ambulatory practices and health systems.

Suki does not publish list pricing at time of writing; third-party reports put enterprise pricing high per user per month, but treat those figures as signals, not quotes — get it in writing. Best for practices that want voice-driven workflow help beyond the note. The limitation: the value depends entirely on how much of your workflow the commands cover, and in my sessions the gap between demo polish and day-to-day edge cases (long specialist vocab, unusual templates) was where the scrutiny should go.

Glass Health

Glass Health combines two things usually sold separately: ambient documentation and clinical decision support — differentials and assessment-and-plan suggestions grounded in clinical literature. In my evaluation, the decision support is the interesting half: it generates a differential from a case summary and cites the reasoning, which makes it auditable in a way free-form chatbots are not. It publishes its pricing openly, which is rare in this market.

At time of writing, there is a free tier and the clinical decision support features sit at a per-clinician monthly price (around the $90 mark on their own published comparison pages — verify current numbers). Best for clinicians who want documentation plus diagnostic thinking support in one tool. The limitation: it is a thinking aid operating on the summary you provide; a thin or biased case summary yields a thin or biased differential, and I treat every generated differential as a prompt to think, never as a worklist to trust. The other detail I check with any decision-support tool is how clearly it separates what the evidence supports from what the model is guessing, and Glass does this better than most, though not perfectly.

Tempus

Tempus is not a point tool; it is a precision-medicine platform — genomic testing, structured clinical data and AI-driven analytics, with its Next platform aimed at oncology care pathways and Lens aimed at life-sciences research. In my evaluation of what a health system actually adopts here, the value is in oncology: matching patients to therapies and trials using molecular data, with published validation studies behind specific algorithms. This is institutional software sold through enterprise agreements, not a download-and-try product.

Pricing is entirely quote-based and depends on the testing and platform mix; there is no consumer-tier figure to cite. Best for oncology programmes and academic centres running precision-medicine workflows. The limitation: the deployment weight is enormous — implementation, data integration and governance are multi-month projects, and outside oncology the case for this specific platform weakens considerably.

Infermedica

Infermedica is the intake and triage engine of this group: an API-first product that takes structured or free-text symptom evidence and returns probable conditions and triage urgency, used by health systems and digital-health apps as the front door before a clinician is involved. In my evaluation, the engineering posture is the differentiator — documented APIs, a regulated medical device context, and triage levels rather than open-ended chat, which is exactly the right instinct for patient-facing risk.

There is trial API access and commercial plans are arranged with sales; per-call pricing depends on volume at time of writing. Best for organisations building patient intake, symptom checking or routing flows. The limitation: it is a component, not a product — you own the interface, the guardrails and the escalation paths, and any triage engine operating on patient-entered symptoms inherits the oldest problem in medicine: people describe things badly.

Cogito

Cogito targets the other end of the patient experience: the contact centre. It analyses live conversations — voice, tone and pacing — and coaches agents in real time with prompts like acknowledging frustration or slowing down. In my evaluation of demos and published case material, the coaching loop is where it earns its place: supervisors cannot listen to every call, and behavioural nudges mid-call are the one intervention that measurably changes that interaction. It grew up in healthcare and payer service lines before being acquired by Verint, and it integrates with contact-centre platforms rather than EHRs.

Pricing is enterprise-agreement territory with nothing published; expect a per-seat conversation-analytics model. Best for health systems and payers running large patient access or member services teams. The limitation: it is a workforce tool, not a clinical one — it improves how agents handle conversations, and buying it to fix a broken service operation will not work. The measurable gains I have seen claimed centre on handle time, escalation rates and agent behaviour scores — service metrics, not clinical outcomes.

How to choose

Match the tool to the risk you carry. If after-hours charting is burning out clinicians, start with ambient scribe category — Nabla for a focused scribe, Suki or Glass Health if you want workflow or decision support alongside it. If you are building patient-facing intake and need controlled triage rather than a chatbot, Infermedica is the only one on this list designed for that job. Tempus belongs to oncology programmes with the infrastructure to use it, and Cogito to large service organisations. Across all six, the same trade-off applies: the more clinical risk in the loop, the more you must verify grounding, audit trails and regulatory status before rollout. Pilot with a small group, measure documentation time and note quality, and never let a model output reach a chart without a clinician signing it. Measure documentation time per clinician before and after, and audit a sample of notes weekly during the pilot; those two numbers tell you more than any vendor case study.

FAQ

Are these tools approved as medical devices?

Status varies by product and by jurisdiction, and it changes. Infermedica and Tempus operate in regulated contexts with published positions; documentation tools generally sit in a lighter category. I ask every vendor for current regulatory documentation in writing before a rollout — marketing pages are not compliance evidence.

How do I evaluate note quality from an ambient scribe?

The same way you would review a junior colleague: sample signed notes against the recorded encounter, check for omissions and invented detail, and measure the editing time you spend per note. In my experience, edit distance per note over two weeks tells you more than any demo.

What about patient data and privacy?

Treat every vendor as a data processor with full scrutiny: signed business associate agreements where applicable, explicit retention policies, and clarity on whether data trains models. If a vendor cannot answer those three things plainly, that is the answer.