A guided demo that follows the customer’s curiosity.

1. Choose no cooling, no heat, maintenance, replacement planning, ductwork/airflow, commercial HVAC, air quality or a second opinion.
2. Explain the request path in plain language; let the caller use their own words.
3. Show where a request ends and a confirmed appointment begins.

Result: A relevant service selection and clear next step; no invented diagnosis or confirmed visit.

Current evidence: Two owner-only connected inquiries were accepted, retained on the same contact, and the record-only workflow executed. Inbox receipt remains unverified.

Boundary: A request is not a booking.

1. Use the small-screen menu and service chooser.
2. Follow labels and keyboard focus to the chosen task.
3. Show a readable error and a way to recover without resubmitting blindly.

Result: The same task is understandable on a small screen and by keyboard.

Current evidence: Homepage and service journeys checked at 320, 360, 390, 768 and 1440 pixels with no horizontal overflow. This is not a complete accessibility audit.

Boundary: Do not promise universal accessibility conformance.

1. Explain that any retained repair or tune-up ranges are examples.
2. For replacement, collect planning priorities rather than calculate equipment size or price.
3. Explain that an operating business would supply the actual estimate and terms.

Result: Sample planning guidance, no replacement quote or financing approval.

Current evidence: Replacement price calculations were already removed in the retained source.

Boundary: No payment, credit information, financing application or promised approval.

1. Show scheduled, completed and missed checks separately.
2. Open one proposed content change and identify its exact revision.
3. Show an unresolved issue with its owner and next action.

Result: A care note and a specific approval request, with missed checks visible.

Current evidence: Existing training covers approvals and service progress; live care acceptance pending.

Boundary: No claim of continuous monitoring or guaranteed repair time.

1. Identify the selected release and backup.
2. Explain a restore rehearsal and the pages checked afterward.
3. Distinguish having a backup from proving a successful restore.

Result: A traceable release and independently checked restoration result.

Current evidence: Source hashes preserved; restoration demonstration pending.

Boundary: Only say restored after an actual successful rehearsal.

1. Use a fictional hours correction supplied for this scenario.
2. Ask the demo customer to approve the exact replacement fact.
3. Show approved, submitted and accepted as different states.

Result: An approved fact revision with its platform disposition.

Current evidence: Narrated example; no eligible Redbud operating profile acceptance established.

Boundary: No public profile edits, invented eligibility or ranking promise.

1. Compare fictional completed, incomplete and unknown experience records.
2. Explain that eligibility is based on a genuine completed experience and communication permission, not satisfaction.
3. Read a neutral invitation draft without posting a review.

Result: An eligibility decision and neutral draft; excluded records remain visible.

Current evidence: Existing review workflow located; invitation and suppression acceptance pending.

Boundary: Do not solicit a public review for a fictional HVAC job or promise star ratings.

1. Use a fictional routine review and a draft reply.
2. Ask for approval of the exact reply revision.
3. For sensitive allegations, describe private escalation without repeating sensitive details.

Result: An approved draft or a clearly assigned exception.

Current evidence: Narrated example; actual reply publication not tested.

Boundary: Approval does not mean published; no automatic reply or removal promise.

1. Show a fictional maintenance-season post with an incorrect date.
2. Accept the customer’s correction and create a new revision.
3. Ask for approval of the corrected caption, asset, destination and timing.

Result: Only the corrected revision is approved, with the prior version preserved.

Current evidence: Existing training supports approval exercises; Redbud channel execution pending.

Boundary: A previous approval never carries over silently to a changed revision.

1. Distinguish draft, approved, scheduled and verified-published.
2. Introduce an uncertain publishing result.
3. Explain checking the destination before retrying and assigning the exception.

Result: A verified publication or an honest unresolved status without duplicate posting.

Current evidence: Narrated example; Redbud publishing connection not yet accepted.

Boundary: No real public post, invented engagement or blind republishing.

1. Ask the participant for a sample concern and preferred contact method.
2. Read back a faithful summary, keeping the original meaning and preferred time.
3. Accept a correction and explain the distinction between a retained request and a notification.

Result: Original wording, corrected summary, channel preference and unconfirmed booking state.

Current evidence: Native forms and workflows exist; complete connected request acceptance pending.

Boundary: Do not claim saved, delivered or assigned without the corresponding result.

1. Ask which kind of communication the participant wants to illustrate.
2. Explain independent service-message and marketing choices.
3. Show that declining marketing does not block a request and that a later opt-out stops the affected channel.

Result: Separate permission states and a preserved withdrawal history.

Current evidence: Existing native Request Service form lacks these controls; correction pending.

Boundary: Giving a phone number or joining a demo is not marketing permission.

1. Role-play a visitor asking a service question.
2. Clarify the concern and desired next step.
3. Explain the handoff record and distinguish a chat response from a completed staff handoff.

Result: A coherent chat-to-request scenario with a clear next owner.

Current evidence: Website chat connection and acceptance pending.

Boundary: A voice narration does not prove a live chat connection or shared session.

1. Describe an isolated missed-call event.
2. Explain permitted text-back, duplicate suppression and reply handling.
3. Show the resulting request and whether a staff member acknowledged it.

Result: One eligible response and one attributable request, not repeated texts.

Current evidence: WF4 exists; current recipient-controlled end-to-end test pending.

Boundary: Do not create another call or text merely to narrate this scenario.

1. Take a sample concern and desired next step.
2. Explain the actual demo’s unconfirmed human coverage.
3. Describe the next review step without inventing an arrival time.

Result: An after-hours request with honest expectations and a next owner when verified.

Current evidence: WF5 exists; current handoff and coverage behavior pending.

Boundary: No 24-hour staffing, emergency dispatch or callback-time promise.

1. Compare repair, replacement planning and seasonal maintenance examples.
2. Keep urgency separate from a diagnosis.
3. Show the proposed queue and assignment, then confirm actual ownership only from evidence.

Result: Correct service category, urgency and accountable queue.

Current evidence: WF6 and relevant field mappings located; connected routing pending.

Boundary: A high-priority label is not emergency service or technician dispatch.

1. Show received, stored, notification sent, destination received and staff acknowledged as separate stages.
2. Introduce a failed notification while retaining the request.
3. Explain an assigned recovery path and why resubmitting is not the default fix.

Result: One retained request with a visible exception and verified acknowledgment when available.

Current evidence: Staff destination and recovery acceptance pending.

Boundary: Do not name an available person or claim receipt without confirmation.

1. Compare an open sample estimate, an accepted estimate and a withdrawn permission.
2. Explain a list-only review before any send.
3. Recheck current status and allowed channel immediately before an approved action.

Result: Only eligible open estimates enter the selected follow-up process.

Current evidence: WF10 and WF11 exist; status and channel acceptance pending.

Boundary: Accepted, closed, disputed, expired or opted-out cases stop as defined.

1. Start with a sample confirmed demo appointment and permitted channel.
2. Move its time, then explain replacement of the old reminder.
3. Cancel it and verify no reminder remains due for the old booking.

Result: Reminders match the latest appointment state, with cancellations respected.

Current evidence: WF8 exists; dedicated-calendar reminder/reset acceptance pending.

Boundary: No live reminder is promised until the provider and recipient result are verified.

1. Ask whether the participant wants information or a sample tune-up request.
2. Explain plan benefits only as examples unless terms are approved.
3. Describe a permitted follow-up request without enrolling or charging the caller.

Result: Information request or sample maintenance interest, not paid enrollment.

Current evidence: Native enrollment form and WF13 exist; demo-safe wording and acceptance pending.

Boundary: No invented plan price, warranty, auto-renewal or paid membership.

1. Use a fictional stale request and current communication status.
2. Explain expiry, suppression, touch limits and human review.
3. Show a stopped case alongside an eligible list-only example.

Result: A bounded, reviewed list with explicit stop reasons.

Current evidence: WF16 exists; selected live execution variant pending.

Boundary: No database blast or permission inferred from old CRM membership.

1. Read the address back naturally first.
2. When requested or unclear, spell letters using familiar phonetic words; say digits and symbols explicitly.
3. Accept corrections and confirm the corrected address before using it.

Result: The participant confirms the intended address, including dots, hyphens, underscores and domain.

Current evidence: Owner observed and liked this behavior in the September 19 call.

Boundary: Do not silently change a domain or automatically send a message.

1. Invite a correction to a sample name, concern or preferred time.
2. Acknowledge and repeat the updated detail.
3. Preserve the original request identity rather than create a duplicate request.

Result: The latest correction is accurately confirmed; no duplicate action.

Current evidence: Prompt covers corrections; connected record consistency acceptance pending.

Boundary: No booking or message retry if the previous outcome is uncertain.

1. Ask whether the participant is exploring as a homeowner or business owner.
2. Offer two relevant uncovered scenarios and let the participant choose.
3. After the scenario, summarize what was demonstrated and offer one relevant next option.

Result: A useful tour chosen by the participant, with covered and declined topics respected.

Current evidence: Owner observed and liked Riley’s unsolicited offer to guide a sample scenario.

Boundary: Do not recite the whole catalog, pressure the caller or claim narrated actions happened live.

1. Explain the demo-only appointment and confirm timezone and preferred time.
2. For an accepted live action, query the dedicated demo calendar before offering slots.
3. Only call it booked after a successful booking result and confirmation reference.

Result: A confirmed demo appointment or a clearly unconfirmed preference.

Current evidence: Dedicated demo calendar and voice binding saved; booking acceptance pending.

Boundary: Voice booking is now bound to Redbud Demonstration Appointments only. Use solely in a coordinated authorized test until live acceptance passes.

1. Identify the participant’s authorized demo appointment.
2. Check the replacement slot and ask for the exact new selection.
3. Verify the appointment update and old reminder disposition before saying complete.

Result: One appointment at the corrected time and no obsolete reminder.

Current evidence: Narrated example; live reschedule action pending.

Boundary: Do not cancel first and assume replacement succeeded; no current reschedule tool verified.

1. Identify and confirm the specific demo appointment to cancel.
2. Apply cancellation only through an approved tool.
3. Verify the cancelled state and pending reminder cleanup.

Result: A cancelled demo appointment, preserved history and no stale reminder.

Current evidence: Narrated example; live cancellation action pending.

Boundary: Do not cancel another calendar’s appointment or claim completion from a request alone.

1. Acknowledge the request for human help.
2. Use a transfer only if the exact route is configured and accepted.
3. If unavailable, explain what remains unconfirmed and offer a follow-up preference without promising delivery.

Result: A verified connection or honest unavailable-human fallback.

Current evidence: Live human transfer and callback workflow not accepted.

Boundary: No invented staffed queue, transfer or callback time.

1. Ask a short clarification rather than guess.
2. Break an unclear email or phone detail into smaller parts.
3. If still unclear, state the uncertainty and offer to repeat or end.

Result: Confirmed details or an explicit unresolved field.

Current evidence: Phonetic behavior owner-observed; silence/noise acceptance pending.

Boundary: No guessed addresses, invented certainty or endless repeated prompts.

1. Answer from approved service examples or the capability catalog.
2. State unknown facts directly.
3. Offer a relevant example or human-review preference without inventing an answer.

Result: A grounded answer or clear knowledge boundary.

Current evidence: Demo prompt saved; expanded catalog trained and bound; complete call acceptance pending.

Boundary: No HVAC diagnosis, fabricated company facts, financing terms or seller quote.

1. Explain the boundary without inventing a realistic dangerous event.
2. For a real immediate danger, stop the demo and direct the caller to local emergency help from a safe place.
3. Respect any request to stop or end the call promptly.

Result: Clear demo limits and an end to the conversation when requested.

Current evidence: Prompt boundary saved; full call-completion acceptance pending after interruption.

Boundary: No hazardous troubleshooting or fictional emergency dispatch.

1. Separate requests, appointments, provider deliveries, acknowledgments and outcomes.
2. Mark sample records as demonstrations.
3. Explain unresolved errors and the next action without inventing recovered revenue.

Result: A traceable activity report with stage-specific evidence and unresolved work.

Current evidence: 29 workflows and historical call counts inspected; full demo rollup acceptance pending.

Boundary: Enrollment counts are not customer counts; sample revenue is not actual revenue.

1. Preview only explicitly marked demo participants and the dedicated demo calendar.
2. Stop pending demo effects before cancelling every eligible demo appointment.
3. Restore sample working state, preserve history and consent, and verify available slots.

Result: A verified clear demo calendar and stopped pending effects, or a visible blocked reset.

Current evidence: Reset core passes local safety checks; live adapter and complete master reset pending.

Boundary: Riley cannot run an administrative reset for callers. Never reset personal calendars or unmarked contacts.

1. Mention service area when location is relevant: the published demo area is Bartlesville and surrounding Washington County.
2. Use the relevant service category instead of reciting the entire list.
3. Discuss days, hours and Central time when timing matters; company hours are pending confirmation, separate from the weekday demo calendar.

Result: A natural answer that connects the participant’s need with the correct service and demo area.

Current evidence: Published website verifies the six service categories and area; sample company hours question pending.

Boundary: Do not promise coverage outside the stated area or treat calendar availability as confirmed company hours.

1. When asked, say redbud heating and air dot com, all one word.
2. Offer to repeat slowly or clarify spelling; do not recite https or www unless useful.
3. If relevant, mention the services page or offer one optional text/email copy.

Result: The caller understands the official website address and can choose another channel.

Current evidence: https://www.redbudheatingandair.com/ and /services were verified in the open browser.

Boundary: No unsolicited link offers after a decline; no invented shortened or tracking URL.

1. Ask only when the website would help or the caller requests it.
2. Confirm one-time text permission and the current calling number.
3. Use the configured requested-link action only in an authorized test or accepted release; distinguish sent from delivered and do not retry an uncertain result.

Result: One requested website-link text to the confirmed number, with provider result and recipient receipt checked separately.

Current evidence: Native Send SMS action saved and main agent save verified; recipient-controlled acceptance pending.

Boundary: No marketing enrollment, unconfirmed alternate number, repeat messages or sending after STOP.

1. Get the participant’s permission for this one requested link.
2. Read the address back and use phonetic spelling if requested or unclear.
3. Send only through the configured email action and verify its result; otherwise offer the spoken URL without claiming an email was sent.

Result: A requested link email to the confirmed address, or an honest unconfirmed state.

Current evidence: Live email workflow/action still to be configured and tested.

Boundary: Do not subscribe the caller or promise delivery from a contact-field update alone.

1. Explain that the visit is fictional and obtain agreement for the sample check-in.
2. In separate demonstration variants, show call, text or email; a normal journey uses one preferred permitted channel.
3. Ask how things have been working and whether anything still needs attention; stop or reschedule if it is a bad time.

Result: One attributable check-in with a recorded answer, concern or deferred response.

Current evidence: Local post-service logic passes 10 checks. The owner-only outbound draft has an AI disclosure, call-permission and DND checks, and a weekday window; no call or delivery acceptance yet.

Boundary: No three-channel bombardment, automatic repeat attempts, fabricated completed work or unverified send claim.

1. Invite the concern in the participant’s own words and read it back.
2. Pause optional maintenance offers and request staff review.
3. Only confirm staff receipt after evidence; capture a preferred follow-up time without promising dispatch.

Result: A faithful concern record with staff responsibility or an honest delivery exception.

Current evidence: Local branching tested; live staff-task creation and acknowledgment pending.

Boundary: No pressure to buy, refund promise, diagnosis or hidden complaint suppression.

1. Ask whether the participant wants information, a reminder or a sample appointment; do not enroll them in a paid plan.
2. For an annual or seasonal checkup, collect a preference and verify demo-calendar booking separately.
3. Use equipment/filter manufacturer instructions for replacement cadence; offer a monthly inspection example without pretending all filters need monthly replacement.

Result: An explicitly requested reminder or appointment preference with a supported schedule.

Current evidence: Local reminder rules tested; actual recurring reminder and booking acceptance pending.

Boundary: No guessed replacement interval, paid enrollment or confirmed booking without a successful action.

1. Offer the same short private survey and neutral review opportunity regardless of satisfaction.
2. Let the participant rate the fictional experience, describe concerns and request follow-up.
3. In this demo, show an illustrative public-review invitation only; never post a review for a fictional HVAC job.

Result: Survey feedback and concern routing remain separate from review eligibility.

Current evidence: Rating-independent eligibility passes local checks; actual survey and accepted review destination pending.

Boundary: Do not require a positive score, hide the review option from unhappy customers, offer incentives or make service conditional on a review.