Most service conversations begin with too little information to act: “I need three ceiling lights installed” is useful, but the business may still need to know whether the fixtures are already purchased, the installation height, the address, or another service-specific fact. Ven Qualify turns that conversation into structured readiness without asking an operator to copy every answer into a form. It does so within a template and policy boundary rather than giving the model unrestricted write access.
The service template defines what ready means
Qualification begins with the catalog. Every active service can define typed fields, required-for-action rules, validation, order, and the exact question Ven is allowed to ask. A light fixture service can require quantity and fixture availability; a switch replacement can require a different checklist. This makes readiness a business rule the workspace can inspect. It also stops the AI from deciding on its own which personal or technical details to collect.
For each inbound message, Ven produces a structured reading: classification, intent, urgency, service match and confidence, extracted field values, and a proposed next action. Application code then normalizes that reading against the active catalog and current request. Low-confidence service matches do not create work. Unknown fields are rejected. Typed values must pass validation, emergency intent follows the emergency policy, and an operator-corrected answer cannot be overwritten silently.
Structured AI proposals pass through deterministic policy
Autonomy grows in visible stages. Observe records the reading but changes nothing. Assist turns safe differences into a review where the operator can apply selected changes, accept all, or reject them. Qualify grants a narrow customer-facing permission: Ven may create or update a request for a clearly matched service, apply facts above the confidence threshold, and send one exact template question at a time. The stage is a policy decision, not a more persuasive prompt.
The conversation remains natural even though the state is structured. Ven can recognize details already present in the first sentence, ask only for the next missing required field, read a short answer such as “three” in the context of its previous question, and continue until the checklist is complete. At that point the request moves to Ready for scheduling and the customer receives a clear acknowledgement that the details are complete and a teammate will confirm the next step.
Observe, Assist, and Qualify increase one permission at a time
Ambiguity is treated as a routing problem, not an invitation to guess. If a customer says the work might concern a socket, switch, or ceiling light, Ven first asks what outcome or fault they mean. No random Service Request should appear before the service is clear. Unsupported, emergency, stale, blocked, assigned, or human-taken-over conversations stop the autonomous write path and preserve the case for a person.
Durable state and audit make short conversational answers safe to apply
Every important step is auditable and retry-safe. The decision records its model and policy version, request version, source message, accepted and rejected fields, confidence, reason, and customer-facing action. Request changes, questions, completion notices, and downstream events are committed through an outbox with idempotency keys. Replaying the same customer message therefore does not create a second request or send the same question twice.
Build a qualification test set from one real service
Choose one low-risk service and list the minimum facts required before scheduling. Configure each fact as a typed template field with validation and customer-ready wording. Then create ten messages: complete, incomplete, ambiguous, unsupported, emergency, contradictory, corrected by an operator, a one-word follow-up, a duplicate delivery, and a message that tries to instruct the AI to ignore policy.
Run the set through Observe and compare the structured reading with the expected service, fields, confidence, and next action. In Assist, apply only selected proposals and verify the audit. In Qualify, confirm that Ven asks no more than one approved question at a time and that a replay does not create a second request or duplicate the question.
Do not judge the test only by how many requests reach Ready for scheduling. Record unsafe writes, customer corrections, unnecessary questions, silent conversations, and incorrect service matches. The qualification mode is ready to expand when its exceptions are visible and recoverable as well as when its common path is fast.
Frequently asked question
Can Ven create a Service Request in Qualify mode?
Yes, when Front Office is live, Qualify is enabled, the message clearly expresses service intent, and the match points confidently to an active catalog service. Ven may then create the request, record allowed high-confidence facts, and ask one configured template question at a time. Ambiguous, unsupported, emergency, stale, assigned, or human-controlled cases do not receive the same autonomous write authority.