What problem does it solve?
A first consultation for a serious condition carries more information than a patient can take in during the visit. ChristianaCare describes cancer patients arriving anxious and emotionally drained, facing complex information in a single conversation, which makes it harder to ask questions and understand treatment. At The Ottawa Hospital, patients scheduled for surgery had one Pre-Admission Unit appointment two to six weeks beforehand, which Deloitte calls an unrealized opportunity to give patients continuous access to information.
Clinicians feel the same gap from the other side. The urologic oncologist who built the Vanderbilt pilot describes giving "the same spiel over and over" about common conditions in visits limited to a few minutes by volume. An avatar that explains the basics at home, in the patient's own time, and answers follow up questions moves that repetition out of the visit and leaves the appointment for the patient's goals and preferences.
How does it work?
- The care team chooses the content. A department picks the condition or procedure and writes or approves the material: scripted education, or a curated question and answer document. At Vanderbilt the avatar was trained only on a document chosen by the clinicians.
- An avatar represents the clinician, or a neutral guide. ChristianaCare built an avatar for each radiation oncology physician with a cloned voice, so patients met a face and voice they would see in the clinic. The Ottawa Hospital used a lifelike avatar named Digital Teammate.
- The patient receives it before the visit. The patient opens a link at home on a phone, tablet or computer, and can use it at any hour.
- The avatar teaches or answers. In ChristianaCare's design it delivers personalized educational videos and then runs teach back quizzes. In the Vanderbilt and Ottawa pilots it holds a conversation and answers from approved content; the Vanderbilt avatar was told to refuse prescribing.
- The clinician takes over. The patient arrives better prepared, and the visit covers decisions and values. Teams in these pilots plan to extend the avatar to history taking and to questions after the visit.
- Audience
- Customer facing
- Autonomy
- Assist
- Adoption
- Emerging
- Channels
- Digital human, Web chat
What is it worth?
Benchmarks are computed from the public deployments below: one data point per organization per KPI, with who made each claim.
| KPI | Median | Reported range | Data points | Claimed by |
|---|---|---|---|---|
| Users served | Not pooled | 10 to 958 | 2 | 2 organization |
Value drivers: Customer experience, Employee productivity, Inclusion and access.
Indicative value
A cancer center with 3,000 new patient consultations a year
USD 10,000 to USD 116,667
Annual clinician time released in consultations per year
How this is calculated
Formula: newPatients * useShare * minutesSaved / 60 * clinicianCost. The low scenario uses every low input, the high scenario every high input.
| Input | Low | High | Basis |
|---|---|---|---|
| New patient consultations per year newPatients, consultations | 2,000 | 4,000 | Reference size for a mid sized cancer center. Replace with your own volume. |
| Share of new patients who complete the avatar education before the visit useShare, fraction of patients | 0.4 | 0.7 | Editorial assumption. ChristianaCare's study had 958 patients view the avatar videos but does not publish a completion rate for the invited patients (HospiMedica quotes the chair saying all the patients completed the quiz, but not how many were invited). |
| Clinician minutes saved per patient who completes it minutesSaved, minutes | 5 | 10 | Editorial assumption. ChristianaCare plans to evaluate consultation efficiency and has not published a result, so measure this in your own pilot. |
| Fully loaded cost of one clinician hour clinicianCost, USD per hour | 150 | 250 | Editorial assumption, replace with your own payroll data. |
What it leaves out: Counts only clinician minutes released in the visit. It leaves out production and review time for the content, the cost of the platform, any effect on no shows, anxiety, understanding or outcomes, and the accessibility gain for patients who need another language or a slower pace. Most evidence so far concerns understanding and stress, not time, so the saving is an assumption to test.
Who already uses it?
3 public deployments, strongest evidence first. Grades: A regulator or audit, B the organization itself, C vendor case study, D anonymous or estimate.
ChristianaCare
United States · Healthcare · 2026
ChristianaCare, working with an unnamed digital technology company, created an AI avatar for each physician in its radiation oncology department. Using cloned voices, the avatars delivered personalized educational videos to patients at home before the first in person consultation, followed by teach back quizzes. In a study of 1,464 patients, one group watched a standard educational video and another watched the avatar videos with personalized scripts and illustrations. The health system reports better understanding, greater engagement in care decisions, lower stress and higher satisfaction for the avatar group, and plans to extend the model to surgical oncology.
- Users served: 958
"The other 958 patients (65.4%) viewed an avatar-based video tailored with personalized scripts and explanatory illustrations."
Claimed by: organization
Vanderbilt University Medical Center
United States · Healthcare · 2025
Urologic oncologists at Vanderbilt ran a pilot in which patients talked with an AI avatar of their clinician before a clinic visit. The avatar looks and sounds like the physician and answers questions about prostate cancer and its treatment options from a curated question and answer document chosen by the clinicians. It declines to prescribe, and it also answers local questions such as where to park. Ten patients took part, liked the video format and gave feedback on improvements. The team plans to add medical history taking and post visit questions.
- Users served: 10
"We ran a pilot study with 10 patients to test out this technology."
Claimed by: organization
The Ottawa Hospital
Canada · Healthcare · 2024
The Ottawa Hospital piloted a lifelike AI avatar called Digital Teammate, built with Deloitte Canada on its CortexAI platform and NVIDIA technology. Patients scheduled for surgery usually have a single Pre-Admission Unit appointment two to six weeks beforehand, so the avatar gives them a 24 hour source of approved answers in English and French, including questions about anesthesia and the procedure. A first test phase also placed it at hospital campuses to answer questions about the new campus. NVIDIA describes patient access by smartphone, tablet or home computer. The vendor reports no measured outcomes.
No outcome disclosed.
How do you implement it?
A model agnostic playbook: what to prepare, the order to build in, and what goes wrong.
Data you need
- Approved patient education content per condition or procedure, with a clinical owner and review date
- Written consent for voice and likeness from every clinician who is represented
- Practical local information such as location, parking, what to bring and how to reach the clinic
- A list of questions the avatar must refuse, such as prescribing, diagnosis and dosing
- Languages and reading levels the patient population needs
Systems to integrate
- Appointment scheduling, to send the link and know the visit date
- Patient portal or secure messaging for delivery and follow up
- Electronic health record, only if personalization or history taking is added
- Quiz and analytics store for teach back results and usage
- Clinic phone or secure message queue for handover
Complexity: Medium
The build is mostly clinical content and consent. Each condition needs approved scripts or a question and answer document with a named clinical owner and review date, voice and likeness consent from every physician who is cloned, and a secure way to reach the patient before the visit. Linking the avatar to the electronic record for personalization adds identity checks and privacy review.
- 1
Start with one condition and one clinic
Choose a visit where the same explanation is repeated every day, such as first radiation oncology consultations or a pre admission appointment. A narrow scope makes the content reviewable and gives a clean comparison group.
- 2
Get the content and consent signed off first
Clinical leads approve every script or question and answer document, and every represented clinician signs a consent for voice and likeness. Write down what the avatar must refuse, and put the refusals in the test set.
- 3
Compare against the current material
ChristianaCare compared avatar videos with a standard educational video and measured understanding with a quiz, engagement in decisions, stress and satisfaction. Use the same design so the avatar has to beat what you send today.
- 4
Test the conversation with patients, not staff
Run a small pilot, as Vanderbilt did with ten patients, and ask about format, tone and gaps. Patients there liked video, which argues for an avatar over text.
- 5
Add the follow up questions and history taking last
Questions after the visit and structured history are the next steps teams name. Add them once the pre visit use has a measured effect and a clinician owns the review of what the avatar collects.
Guardrails
- Answers only from clinically approved content, with a plain statement when the content is silent
- The avatar does not diagnose, prescribe, change doses or triage symptoms, and refers urgent symptoms to the clinic or emergency services
- Disclosure that the avatar is AI generated, including when it copies a real clinician's face and voice
- Consent and a stop rule for every cloned voice or likeness, including what happens when the clinician leaves
- Minimal personal data in the conversation, masked in logs, with retention limits
KPIs to instrument
- Share of invited patients who complete the education before the visit
- Teach back quiz scores against the standard material
- Stress and confidence scores before the visit, using the same survey as the comparison group
- Clinician time per new consultation and number of repeat questions asked in the visit
- Questions the avatar declined or could not answer, grouped by topic
Human in the loop
Clinical leads own the content and review a sample of conversations and quiz results every month. Anything the patient says that suggests a symptom, a safety concern or distress goes to a nurse or the clinic line, and the clinician sees what the patient learned before the visit starts.
Common failure modes
- A confident wrong answer about treatment
- The avatar states a risk or option that the clinician would not. Restrict it to approved text, test refusals, and have clinical leads read samples every month.
- Patients assume the avatar knows their case
- A patient may expect an avatar that looks like their clinician to know their chart. Say plainly that it teaches general content and does not know the patient's case, unless it is linked to the record.
- A clinician's likeness is used beyond what was agreed
- Voice and image clones outlive a job or a project. Record scope and end date in the consent, and delete the model when it ends.
- Education replaces the conversation
- Patients arrive informed but the visit still runs short. Track whether visits spend more time on values and decisions, and cut repeated explanation from the visit script.
What are the risks and rules?
EU AI Act
Depends on design
Education from approved content is a limited risk system with transparency duties (Article 50), and an avatar that copies a real clinician's face and voice is synthetic content that must be disclosed as artificially generated. If the avatar assessed symptoms, triaged patients or advised on treatment it could be medical device software under the Medical Device Regulation. That makes it high risk under Article 6(1) and Annex I of the AI Act when the device needs a third party conformity assessment, and brings medical device conformity rules.
Guidance
- Article 50, transparency obligations for providers and deployers of certain AI systems (European Union, Europe). People must be told when they interact with an AI system, and deployers must disclose AI generated image or audio that resembles real people.
- HIPAA Privacy Rule (US Department of Health and Human Services, North America). Sets the rules for protected health information that an avatar may collect or show, including minimum necessary use and patient access.
Controls to put in place
- AI disclosure at the start of every session and on every video
- Clinical owner and review date for every script or question and answer document
- Written voice and likeness consent per clinician, with scope and end date
- Refusal tests for prescribing, diagnosis and urgent symptoms, rerun on every content or model change
- Privacy impact assessment and masked, time limited logs
Frequently asked questions
- Does a clinician's AI avatar replace the consultation?
No. ChristianaCare's radiation oncology chair says of the avatar, "This is not about replacing the physician in any way." The avatar prepares patients so they arrive calmer and better informed, and the visit still holds the conversation about the patient's own case and decisions.
- What results have hospitals reported?
ChristianaCare compared avatar videos with a standard video in a study of 1,464 patients and reports better understanding, more engagement in care decisions, lower stress and higher satisfaction, without publishing effect sizes. Vanderbilt ran a pilot with 10 patients, who liked the video format, according to a clinician who built it. For The Ottawa Hospital, NVIDIA reports that a majority of user acceptance testers found the answers clear and relevant, with no percentage given. That is a vendor claim.
- Can the avatar answer open questions or only play scripted videos?
Both designs exist. ChristianaCare's avatars deliver personalized scripted videos followed by teach back quizzes. Vanderbilt and The Ottawa Hospital use a conversational avatar that answers from approved content. The Vanderbilt avatar was told to refuse prescribing, and the Ottawa sources do not describe what it declines.
- Is a patient education avatar high risk under the EU AI Act?
Not when it only explains approved content: it carries transparency duties, including disclosure of the synthetic clinician. It can become high risk if it triages symptoms or advises on treatment, because it may then count as a medical device.
How to cite this page
Blits.ai AI Use Case Library, "AI avatar for pre visit patient education", last verified 14 October 2026, https://www.blits.ai/ai-use-cases/patient-previsit-education-avatar. Licensed under CC BY 4.0. Method: how we verify use cases.
Changelog
- 14 October 2026: First published