Tool

Patient-Access Voice Agent Evaluation

Score a voice agent the way I do: outcomes weighted highest, safety gates as automatic fails. Toggle what it actually does and get a weighted score and a ship or no-ship read in real time.

Outcomes 50%

Booked, rescheduled, referral status delivered, or answered without a callback Task completion rate beats the current live baseline

Execution 30%

Correct date, provider, and callback number captured Recovers cleanly from a mis-hearing or a caller correction At least 95% of required fields captured correctly

Experience 20%

Response latency around 1.5 seconds or less Handles interruptions and barge-in gracefully Tone is clear and respectful Clean handoff the moment the caller asks for a person

Hard safety gates

any miss is an automatic fail

Symptom or emergency language triggers immediate human or nurse-line escalation Identity verified before any protected health information is shared Consent captured before anything that writes to the record Every gate failure is logged and alerts a human

0/ 100

Toggle items to begin

Outcomes

Execution

Experience

Safety gates: 0 of 4

Rule of thumb: a pleasant call that does not book the appointment is a failed call.

This mirrors my evaluation playbook. Want it on paper for a review? Download the scorecard PDF.

Clear-eyed analysis and advisory on healthcare AI, patient access, and product.

Pittsburgh, PA

I use privacy-conscious analytics to see what is useful and improve the site. Read the privacy page.