A customer challenges the price. ANDIP LIVE investigates.
One worked scenario, followed stage by stage: a guest questions the room rate, and a coordinated set of supporting tasks runs beside the conversation under a strict deadline.
A question that a scripted agent cannot answer
A guest calls a hotel directly. Mid-conversation, they mention that another booking website is advertising a similar room for less than the direct rate. The guest is not making an accusation — they are asking a fair question, and they expect an answer now.
A conventional voice agent may explain the hotel's pricing policy or offer a generic response. That response may not address the guest's actual concern, because answering it properly requires comparing conditions, checking availability, evaluating eligible benefits, and confirming what the business is actually permitted to offer.
With ANDIP LIVE, the voice agent can request assistance without interrupting the conversation. The call stays live while the supporting work runs beside it.
“Why is your room $130 when another site shows $100?”
- Guest claim
- $100 competitor rate — illustrative
- Direct rate
- $130 per night — illustrative
- Unresolved
- Dates, room type, taxes, cancellation terms, availability
These figures are invented examples used to describe a workflow. They are not real competitor prices, and no price, rate, or benefit shown here represents a verified market observation.
Walk through the five stages
Select a stage to see what the customer experiences, what ANDIP LIVE is doing, and what the voice agent is allowed to say at that moment.
The agent asks which site, dates, and room type
Nothing is compared until the comparison itself is defined. This is the cheapest, fastest, and most important step.
What the customer hears
A short, natural clarifying question. The agent acknowledges the concern rather than defending the rate. The guest never hears silence while research happens.
“I can look into that with you. Which website was it, and do you remember the dates and the room type it showed?”
What ANDIP LIVE is doing
- Structured request submittedThe agent sends the question, the current context, a deadline, and the actions it is permitted to take.
- Request classifiedThe Live Decision Broker identifies this as a commercial comparison question, not a simple lookup.
- Scope narrowedOnly the conditions the guest actually named are scheduled for verification.
Compare conditions and check availability
Two rate figures are only comparable if the underlying conditions match. This stage establishes whether they do.
What the customer hears
A brief acknowledgement that the agent is checking, plus one or two follow-up questions about conditions the guest can answer immediately — for example whether the other rate included breakfast or allowed cancellation.
The agent does not promise a discount, a match, or an outcome at this stage.
What ANDIP LIVE is doing
- Internal state readCurrent availability, room categories, and the rate conditions attached to the direct booking.
- Approved sources consultedOnly data sources the business has authorised for this purpose, within the permitted scope.
- Condition comparison startedDates, room type, occupancy, taxes, cancellation terms, and inclusions are matched item by item.
Separate confirmed facts from uncertainty
This is where the system refuses to guess. Whatever cannot be confirmed is labelled as unconfirmed rather than smoothed into a confident-sounding answer.
What the customer hears
The first genuinely useful information. It may be a partial answer — for example that the two rates appear to cover different conditions — delivered as soon as it is verified.
If the evidence is not yet sufficient, the agent says so plainly instead of inventing a comparison.
What ANDIP LIVE is doing
- Parallel verificationSpecialist tasks check the conditions independently and report back as each one finishes.
- Evidence rankedEach item carries its source, retrieval time, and a confidence assessment.
- Uncertainty preservedAnything that cannot be confirmed within the deadline stays flagged as unresolved.
- Partial result releasedVerified findings are returned to the agent immediately rather than held back for a complete picture.
Apply pricing and offer rules
A useful recommendation is not the same as a permitted commitment. This stage decides what the business actually allows.
What the customer hears
Still nothing that commits the hotel. The agent may continue to gather context or explain a general policy while the eligibility check completes.
If an eligible benefit cannot be confirmed under current rules, the recommendation is escalated for human review rather than presented as an offer.
What ANDIP LIVE is doing
- Policy evaluatedRate rules, offer eligibility, and any conditions that must hold are checked against the current situation.
- Eligibility confirmed or refusedEach candidate option is either marked permitted, refused, or escalated for approval.
- Action authority resolvedThe response is labelled as an explanation, a permitted offer, an escalation, or a qualified statement of uncertainty.
Explain differences or present a permitted option
The agent closes the loop with what it is actually allowed to say, and nothing more.
What the customer hears
A concise explanation of how the two rates differ, and — only where authority allows it — a specific permitted option such as an eligible inclusion.
“The rate you saw may not include the same conditions as ours. Based on your dates, I can confirm what is currently available and what I am able to offer you today.”
What ANDIP LIVE is doing
- Response assembledVerified explanation, permitted option, or an explicit statement of what remains unresolved.
- Result attributedThe conversation record keeps the sources, timestamps, and confidence behind what was said.
- Conversation continuesAny remaining background work either refines the answer later or is discarded as no longer relevant.
Why the call never has to wait
The workflow above is not a sequence of blocking steps. It is a conversation running in parallel with bounded work that reports back as it progresses.
Illustrative timeline of one live request
Conceptual — illustrative sequencing, not a measured latency profile
The relative positions of these markers illustrate sequencing and dependency, not measured durations. No latency figure is claimed anywhere on this page.
What the voice agent sends and receives
The shapes below illustrate the kind of information that would travel between a voice agent and ANDIP LIVE. They are conceptual examples, not a published API, and no such interface is currently available.
{
"conversation_id": "call-illustrative-0001",
"question": "Guest says another site shows a lower room rate",
"context": {
"industry": "hospitality",
"stated_dates": "illustrative",
"stated_room_type": "illustrative"
},
"deadline": "set by the broker, not the caller",
"permitted_actions": ["explain", "offer_eligible_benefit"],
"may_commit_business": false,
"budget": "constrained per assisted conversation"
}{
"conversation_id": "call-illustrative-0001",
"status": "partial",
"evidence": [
{
"claim": "Rate conditions differ between the two offers",
"source_type": "approved_source",
"confidence": "assessed, not assumed",
"retrieved_at": "timestamped"
}
],
"policy": {
"permitted": ["explain_conditions"],
"requires_approval": ["non_standard_discount"]
},
"uncertainty": [
"Final competitor inventory could not be confirmed in time"
],
"next_update": "may follow if background work completes"
}No SDK, authentication endpoint, production specification, or deployment command is published or implied by the examples above. They exist to explain the intended contract between the conversation and the intelligence layer, and the interface will change as the architecture is validated. See the documentation for a fuller description of planned interfaces and their current status.
Questions about this workflow
No. Many requests resolve at the first or second stage — a simple internal lookup or a direct authorised API call needs no parallel verification and no policy escalation. The five-stage sequence describes the most demanding case, where a commercial question must be compared, verified, and checked against authority before the agent speaks.
No. This page is a conceptual walkthrough of a planned workflow, and every figure on it is illustrative. The first reproducible demonstration is targeted for the community alpha in Q1 2027. See the community alpha page for the planned scope of that demonstration.
The conversation continues through a defined fallback rather than blocking. The agent receives whatever was verified, together with an explicit statement of what could not be established. A failed research task never becomes a fabricated comparison, and it never becomes an indefinite wait for the caller.
Not by itself. ANDIP LIVE separates recommendations from authorisation. Before a specific commitment is presented, applicable pricing rules, offer eligibility, inventory, and any required approval are validated. Where direct authorisation is unavailable, the system suggests an option for human review instead of acting independently.
Because it is a compact illustration of the general problem: a customer raises a factual challenge that requires several bounded lookups, a comparison of conditions, and a policy decision, all while the call remains live. The same structure applies to itinerary disruption in travel, unexpected technical issues in customer support, and objections in B2B sales — see use cases.
The routing logic behind this workflow
Every stage above is decided by one component: the Live Decision Broker. It chooses which path a request takes, how long it may run, and when the available evidence is sufficient to answer.