Clinics · Call compliance
DashboardClinics & diagnosticsChecking every front-desk call at a clinic, not just the ones someone heard
A front desk takes hundreds of calls a week and nobody can listen to them all. This dashboard checks each one for identity verification, consent, clinical boundaries and complaints, and points the practice manager at the calls that need a person.
Valnora & Elric team5 min read
Evidence, latest failed checks
| Call | Staff | Check | Quote | Result |
|---|---|---|---|---|
| F-1182 | Desk 2 | Identity before details | Your scan is booked for the 14th | Fail |
| F-1190 | Desk 1 | No clinical advice | You can double the dose until then | Fail |
| F-1194 | Desk 3 | Complaint route offered | That is just how it works here | Fail |
| F-1201 | Desk 2 | Identity before details | Can I take your date of birth? | Pass |
100%
Imported calls checked
Illustrative. Every call against the same checklist, not a hand-picked sample.
< 1 day
From critical breach to review
Illustrative. The monitor and Relay put the call in front of the practice manager.
4 checks
Core front-desk checks
Identity, consent statements, clinical boundary and complaint handling.
Illustrative example. Figures show how the workflow is measured; they are not customer results.
A clinic front desk handles bookings, results queries, billing questions and anxious relatives, often in the same hour. Most calls go fine. The few that do not are the ones that matter: a test date read out to someone who never confirmed who they were, a well-meant tip about a medicine, a complaint that nobody wrote down.
Spot checks catch some of these. A practice manager listening to a few calls a week will miss most. This post walks through a Call Compliance dashboard, built with Agent Serve Dashboards, that checks every imported call and sends the practice manager to the ones that failed.
Getting calls onto the dashboard
Create the dashboard from the Call Compliance template and import your calls. It takes audio files such as mp3, m4a and wav, which are transcribed with speaker labels so the front desk and caller are told apart. It also takes transcripts as txt, vtt, srt, docx, pdf, json, md or CSV, pasted text, or a folder from the Library.
Turn on PII redaction in Settings → Data retention before the first import. It masks personal data in logs, workflow inputs and block outputs. If calls arrive through a workflow instead, add a Guardrails block with PII detection set to mask, so names and numbers are hidden before text reaches the next step.
The front-desk checklist
Each call is checked against a checklist written in plain English. For a clinic front desk, four checks carry most of the weight:
- Identity before details. Staff confirm who they are speaking to before sharing appointment, result or billing details. Critical.
- Recording and consent statements. The call recording notice and any consent wording your clinic uses are said, and said before the relevant part of the call.
- No clinical advice. Front-desk staff do not advise on symptoms, doses or results. They pass the caller to a clinician. Critical.
- Complaint signals. If the caller expresses dissatisfaction, staff acknowledge it and offer the complaint route.
Each call gets a pass, fail or not applicable per check, with a quote as evidence. The same analysis fills fields for QA score, sentiment, escalation risk, topics, intent, talk ratio, next steps and a summary.
Reading the dashboard
Read it top to bottom. The tiles give the compliant share, critical breaches and average QA score against the previous period. The chart shows which checks fail most, week by week. The evidence table is where the practice manager works: each failed check with the staff member, the quote and the result.
Break the same results down by staff member, site and channel. A pattern of identity failures at one desk is a training gap, not a one-off. All figures above are illustrative.
How to measure it
100%
Imported calls checked
Illustrative. Every call against the same checklist, not a hand-picked sample.
< 1 day
From critical breach to review
Illustrative. The monitor and Relay put the call in front of the practice manager.
4 checks
Core front-desk checks
Identity, consent statements, clinical boundary and complaint handling.
Illustrative example. Figures show how the workflow is measured; they are not customer results.
Rules, monitors and the review note
Two definitions keep every chart and alert counting the same way. A rule flags any call with a critical failure. A metric tracks the share of calls with at least one failed check.
Monitors watch both, re-checking as often as every five minutes, with alerts by email, in the app or in Slack. One fires on any critical breach. One fires when the failure rate passes a level you set.
What the solution produces
Review note after a clinical-advice breach (illustrative)
Record- Call
- F-1190 · inbound · repeat prescription query · Desk 1
- Failed check
- No clinical advice from front desk (critical)
- Evidence
- At 01:12 the caller was told to double a dose until their appointment. No transfer to a clinician.
- Pattern
- First failure of this check for this staff member.
- Suggested action
- Practice manager to listen back today. If confirmed, the clinical lead decides whether the patient needs a call.
- Not done
- No contact with the patient or staff member.
Connect the critical-breach monitor to a Relay, an AI coworker that starts when the monitor alerts. In Observe mode it reads the call result and drafts a review note: what failed, the quote, whether the same check failed recently for the same person, and a suggested next step. It cannot change a result or contact anyone. The practice manager listens back and decides. If clinical advice was given, the clinical lead decides whether the patient needs a call.
Coaching, not surveillance
QA scores by staff member are most useful in one-to-ones, with the quote on screen. Share the checklist with the front desk before the first import, and tell them how results will be used. Row-level access keeps each site manager to their own site.
Controls and governance
PII detection
A Guardrails block detects personal data and blocks or masks it before text reaches the next step.
Row-level access
Row-level policies narrow what each person sees in tables, dashboards, exports and scheduled reports.
Roles and permission groups
Admin, write and read roles per workspace, plus permission groups that limit which models, blocks and integrations a team can use.
Audit trail
Proposals, approvals, declines and executed actions are written to the workspace audit log, which admins can export.
SSO and SCIM
Sign in through your identity provider with SAML 2.0 or OIDC, and provision people with SCIM.
Self-hosting
Run Agent Serve in your own cloud or data centre, next to the systems it reads.
Zero data retention
Zero data retention with supported model providers, backed by your own provider keys and agreements or self-hosted models. Some provider features keep data by design, such as Gemini Deep Research, so they are outside zero data retention.
More on how Agent Serve handles access, data and audit on the security page.
Health-data and call-recording rules differ by country and state. Check with counsel what your recording notices must say and how long recordings may be kept. Agent Serve gives you the controls; it does not certify a clinic as compliant with any law.
For how calls feed the complaints process, see patient complaints and feedback. For the general pattern behind this dashboard, see call QA on every call.
Questions teams ask
Does the dashboard judge whether clinical advice was correct?
No. It checks one thing: whether front-desk staff gave clinical advice at all, which most clinics do not allow. Whether the patient needs follow-up is decided by your clinical lead.
Can we import audio, or only transcripts?
Both. Audio files such as mp3, m4a and wav are transcribed with speaker labels. Transcripts import as txt, vtt, srt, docx, pdf, json, md or CSV, by paste, or from a Library folder.
Who sees the call content?
Only people your roles and row-level policies allow. PII redaction masks personal data in logs, and teams that need call data kept in-house can self-host Agent Serve.
From a front-desk call to a reviewed result
Data · Calls in
- 01Call audio or transcriptsAudio files such as mp3, m4a or wav are transcribed with speaker labels; transcripts import as txt, vtt, docx, pdf, CSV and more.
- 02Front-desk checklistIdentity check before details, recording and consent statements, no clinical advice, complaint route offered.
- 03Call resultsOne row per call: result per check, QA score, sentiment, topics and evidence quotes.
Dashboard · Front-desk compliance
- 04Call Compliance dashboardCompliant %, critical breaches, failed checks per check and QA score by staff member and week.
- 05Breach monitorRe-checks as often as every 5 minutes and alerts by email, in-app or Slack.
- 06Review RelayStarts on the alert in Observe mode and drafts a review note. It contacts nobody.
- 07Practice managerListens back, confirms or overturns the result and agrees coaching.
What you need
- Call recordings or transcripts exported from your phone system, with a staff and site field per call
- Your front-desk script, rewritten as plain-English checks with a severity each
- The exact recording and consent wording staff are expected to say
- A named practice manager per site to review breaches, and a clinical lead for clinical ones
- PII redaction turned on in Settings → Data retention before the first import
- Row-level access so each site manager sees their own site only