Evals for Heidi Health

Transcript-to-note fidelity

Note content stays faithful to what was actually said in the encounter, without invented findings, history, or plan items.

Template and structure handling

Standard and advanced templates produce the expected note sections and formatting for the selected visit type.

Coding output

Codes surfaced alongside the note reflect the documented encounter, a capability gated to paid tiers.

Follow-up task generation

Post-visit tasks derived from the encounter, including task management and session status handling.

Snippet trigger expansion

Short triggers such as 'cc' or 'ros' expand into the full heading or template pulled from existing Heidi snippets.

Spoken formatting and correction commands

Commands like 'new paragraph', 'full stop', and 'scratch that' are applied as edits rather than transcribed as text.

Command versus clinical term disambiguation

Distinguishing dictation commands from clinical vocabulary that sounds similar, such as 'comma' versus 'coma'.

App-aware output shaping

Output format adapts to the destination app: structured in an EHR, prose in a document editor, casual in chat.

Citation transparency

Answers carry the transparent citations Heidi advertises, attributable to the sources actually used.

Patient-context-aware answering

Answers reflect the active patient session context where the tier enables it.

Source preference and library scoping

Preferred journals, premium sources, personal libraries, and org-wide libraries constrain which material an answer draws on.

Live evidence suggestions in visit

Evidence surfaced during a patient visit rather than only on explicit query.

Coding and CDI

Clinical documentation improvement and coding workflows at organizational scale.

Care gap identification and outreach

Identifying care gaps and generating the corresponding patient outreach.

Referral triage and leakage capture

Triaging referrals and flagging revenue or referral leakage.

Pre-charting and visit preparation

Preparing chart context ahead of an encounter, including repeat prescription and refill handling.

Abstention on insufficient input

Declining to assert a finding, code, or answer when the encounter or source material does not support it.

Uncertainty signaling

Surfacing when output is low-confidence rather than presenting it with uniform assurance.

Clinician-in-the-loop framing

Output is presented for clinician review and does not assert final clinical decisions on its own.

Degraded capture recovery

Behavior when audio is poor, truncated, or interrupted, including remote-mic capture during rounds.

Tier feature gating

Features listed as limited, unavailable, or included for a given plan behave accordingly.

Data retention configuration

Flexible retention, including team-level settings on higher tiers.

Sharing and team configuration

Document and session sharing, team template sharing, and free assistant users where the tier allows.

Subscription lifecycle

Cancellation preserving current-plan access through the end of the billing cycle, as stated in Heidi's published FAQ.