The architecture is reusable. The installed configuration is not.
Two workflows contain 84 real nodes and 113 connected edges. The core pattern is strong, but four booking nodes have no calendar, four appointment triggers have no source, copied account references remain, installer placeholders are visible, and no run history proves an end-to-end route.
Patient lifecycle
Multilingual intake, booking, confirmation, cancellation, messaging and staff handoff are represented.
Scheduling
Four booking nodes and four appointment triggers require installer-owned sources.
Execution
The agent is inactive, has zero active channels and has no recorded runs.
One agent, two workflows and a six-stage dental Lead pipeline.
GPT-5.4 Mini, Nova voice, en-US, 207-word prompt and zero active channels.
61 edges plus three placeholder terminals; multilingual intake and five English intent branches.
52 edges plus sixteen placeholder terminals; four repeated appointment-to-email fanouts.
Appointment Scheduled, Confirmation Request, Cancellation Request, Call Forwarded, Left a Message and Not interested.
Four booking nodes show “No calendar”; four appointment triggers show “Source: Not configured.”
Four Incoming Call emails plus 32 patient/internal notification emails; one patient confirmation is disabled.
No selected knowledge base, no custom tools, 0/17 enabled built-in tools and no widgets.
No simulator or live run appears in the agent’s execution history.
Text version of this diagram
The snapshot installs one inactive agent, two workflows with 84 real nodes and 113 edges, a six-stage Lead pipeline and 36 email nodes. The workflows contain five triggers, four unbound booking nodes, three Ask and Save nodes and three extracted variables. No knowledge base, tool, widget or run is configured.
Text version of this diagram
The Dental AI Receptionist runs Incoming Call and Appointment Notification. Incoming Call writes contact data and Lead pipeline stages and depends on calendar, staff and email configuration. Appointment Notification depends on appointment sources and email. The prompt reads company and support values and expects dental knowledge that is not selected.
The front-desk workflow covers the right jobs but carries copied dependencies.
The conversation-started trigger accepts chat, voice or SMS, branches by language, then sends English callers into booking, cancellation, message, staff or confirmation routes. The Spanish route transfers to a named AI agent from the source account.
Language, intent, appointment-type, reschedule, success and fallback copy.
Five questions each for confirmation, message and cancellation; maps first name, last name and phone.
Eight questions each; all are set to America/New_York and show “No calendar.”
Four Lead stage updates plus one TOIP / Cancellation Request mismatch.
One account-specific AI-agent transfer and one call-forwarding rule set.
Produces appointment date/time and appointment location.
Text version of this diagram
Conversation Started routes by English, Spanish or Chinese. English exposes five intents. Spanish transfers to an account-specific AI agent, and both transfer outcomes end at placeholders. Chinese continues through conversation handling.
Text version of this diagram
Booking chooses a dental need, enters one of four unbound calendars, updates Lead / Appointment Scheduled, extracts date/time and location and sends email. Cancellation collects identity, points to TOIP / Cancellation Request, sends email and offers rescheduling. Message collects identity, updates Lead / Left a Message and emails staff. Staff updates Lead / Call Forwarded and uses answered, no-answer and outside-hours outcomes.
Four repeated notification trees generate 32 emails.
Each appointment-booked trigger feeds a four-path decision: Consult, Followup, Cleaning and Treatment. Each path sends a patient confirmation and then an internal/support notification. All four triggers lack a source and all sixteen final internal emails terminate at placeholders.
Appointment triggers
Every trigger reports “Source: Not configured.”
Decision paths
Consult, Followup, Cleaning and Treatment repeated for four sources.
Terminal placeholders
Each internal notification ends without a connected next action.
Text version of this diagram
Four unconfigured appointment triggers each lead to a decision with four appointment-type paths. Every decision leads to four patient emails, then four internal support emails. The sixteen internal emails end at placeholder terminals.
The contact contract is clear; the account contract is incomplete.
Contact fields, pipeline context and message copy.
Appointment summary and notification content.
Booking and confirmation messaging.
Prompt and email subjects; installer must supply.
Internal appointment, cancellation, message and booking notifications.
Patient-facing confirmations.
Text version of this diagram
Ask and Save writes identity and phone information to the contact record and pipeline. Extraction produces appointment date, time and location for patient and internal email. The account must supply company name, support email and calendar or appointment sources.
Text version of this diagram
The installed core includes the agent, two workflows and Lead pipeline. The owner supplies company copy, four scheduling sources, email delivery, human handoff and healthcare rules, then runs synthetic validation. Remediation must separately repair the copied transfer, TOIP pipeline mismatch and unresolved email terminals.
Nine actionable findings, separated from normal owner setup.
Map four installer-owned dental bookables and verify success and error paths.
Bind every trigger to the intended booking source and appointment type.
Replace the copied AI-agent destination and connect both Forwarded and Failed outcomes.
Replace TOIP / Cancellation Request with the installed Lead stage.
Replace company, tagline, doctor and phone placeholders before activation.
Correct the FQA placeholder and attach approved dental knowledge.
Confirm intentional terminal behavior or connect explicit completion steps.
Remove the duplicate or intentionally enable and connect it.
Run the complete synthetic test matrix and preserve successful runs.
The next document is remediation, not activation.
- 1Freeze the source baselineComplete
Preserve this exact v33 inventory, node/edge counts, mappings and finding IDs.
- 2Build a private repair accountNext
Use a clean `[Votel Snapshot] Dental AI Receptionist` sub-account and never modify this source record.
- 3Repair snapshot defectsRequired
Fix the copied transfer, pipeline mismatch, placeholders, disabled email and terminal outcomes.
- 4Add the improvement layerPlan
Standardize calendar naming, setup copy, dental knowledge, safety escalation and review collection.
- 5Validate every routeRequired
Use synthetic patient data only; test all languages, intents, appointment types, handoff outcomes and failures.
Dental source audit complete.
The installed snapshot now has an exact inventory, workflow maps, data contract, dependency map and defect register. The private repair account is now tracked separately.
Open the Dental remediation report →