Travel & Hospitality AI agent · Voice
Voice & Contact-Centre AI Agent
Take guest calls with all-party consent captured before audio reaches a model, disclose the automation on the first turn, and keep a person one utterance away in each language answered.
- Typical delivery
- 4–6 weeks
- Deployment
- Your stack
- Human handover
- All-party
- After launch
- Agent Care
What this agent does
Consent first, then the conversation
In
- 01
Capturing all-party consent at the top of the call, before audio reaches any model at all.
- 02
Disclosing on the first turn that the caller is speaking with an automated system, not a person.
Reason
- 03
Answering the question 'am I talking to a robot?' truthfully, in the same turn it is asked in.
- 04
Routing audio only to models with no path to shared training, as a tested and evidenced control.
- 05
Detecting the language spoken, and handing to a person where it is one the suite has not covered.
Decide
- 06
Holding relay, TTY and long-silence calls for a person rather than timing the caller out.
- 07
Escalating to a named colleague in one utterance, in each language the agent answers in.
Out
- 08
Retaining the call, the disclosure, the consent state and the words actually spoken.
- 09
Executing write actions only inside the approval boundaries agreed during implementation.
Product statement
The agent handles the call; a named colleague takes over on request, and the operator owns every word the agent said.
Example workflow
One call, consent to handover
- 1Call receivedReservations line, property switchboard, service number or a callback
- 2Consent and context assembledConsent state, disclosure given, language detected and the booking or stay on file
- 3Answer spokenSpoken answer, its source, confidence
- 4Controls appliedConsent checks, disclosure checks per market, language-coverage checks and confidence threshold
No human action required
Stages 1 to 4 run unaided, and nothing is committed at any of them — the agent is answering, and the colleague's lane opens at the confidence gate.
Stays with the agent, handover one word away.
Hands to a colleague on the floor.
Colleague takes over
The call is passed with its transcript, its flagged lines and the confidence.
Take over · Coach · Escalate to the floor lead
Handled — the caller has an answer
- 6Contact systems updatedOnly where write access and approval policy allow it
- 7Outcome evaluatedHandover rate, flagged-line outcomes, language coverage and post-call corrections
Handovers
Every colleague handover is counted in the evaluation.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required
8 items
- Recording a call before consent has been captured.
- Continuing in a language the suite does not cover.
- Telling a caller it is a person when it is asked.
- Placing an outbound call without prior consent.
Automation boundary
Agent acts unaided
- ✓Capture consent and disclose the automation on.
- ✓Answer from the booking or stay record, and name that record.
- ✓Detect the language and say which ones have been evaluated.
- ✓Flag what a person must weigh, and pass the call to them.
Any write happens inside the boundaries agreed at implementation, never ahead of approval.
- Deciding that a relay or a TTY call may be dropped.
- Settling a complaint, a refund or a goodwill waiver.
- Sending guest audio anywhere it may train a model.
- Changes to consent, retention or language rules.
Example output
One call, annotated
Everything the agent says on a call is attached to the consent and the record behind it.
Voice output · single call
Illustrative example- Call
- Inbound, Spanish
- Spoken line
- Confirmed the arrival date held on the booking, then offered a colleague
- Time to a person
- 22 sec
- Source of record
- Booking record on file
- Confidence
- 87%
- Attribution
- Property and automated system
As received
Taken from the consent state and the booking record on file — nothing on this side is written by the agent.
Records used
- All-party consent state
- Booking record on file
- Evaluated language set
Why this wording
It says which record the date came from, and offers a person rather than improvising.
What the score decides
Below the configured threshold the call is passed to a colleague before the answer is spoken.
Action
Take overCoachEscalate to the floor leadValue
Where AI adds value
The same four claims, placed at the point in the workflow where each one applies.
Where the value lands
Value 01 – 04
Every call
From the consent state
Answer from the record
Draw on the booking or stay on file, your approved answers, and the disclosure each market you dial into requires.
Disclose it before you record
Routine calls are handled with the disclosure given and the consent already captured.
Send handover to the hard calls
Relay calls, unevaluated languages and low-confidence turns go to a colleague — disability access and language access are different duties, and neither stands in for the other.
Build an evidence trail
The call, the disclosure and consent recorded for it and the person who took over stay together.
Integrations
Typical integrations
Five system groups connect to the same agent. Which of them are in scope is decided in discovery.
Contact platforms
- Genesys
- NICE
- Five9
- Amazon Connect
Telephony
- Twilio
- SIP trunks
- Session border controllers
Reservations and property
- Amadeus
- Sabre
- Opera Cloud
- Mews
Agent
Voice & contact centre
- Captures consent
- Answers the call
- Hands to a person
Recording and QA
- Verint
- Calabrio
- Call-recording archives
Observability & evaluation
- OpenTelemetry
- Langfuse
- Supported monitoring/evaluation sources
Integration availability depends on the client's existing systems and API access.
Agent controls
Six layers between the model and the caller
The controls are nested. What survives all of them appears in the map underneath.
L6 · Outermost — last line of defence
Inward → L1 · closest to the model
- L1 – L2
- Test whether a turn may be spoken
- L3
- Puts the call in a colleague's hands
- L4 – L5
- Show the call and the consent recorded for it
- L6
- Holds the call for a person when signals degrade
How Nestack evaluates it
Evaluate the calling workflow — not only the sentence at the end of it.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the answer the caller hears
- E1Final-output evaluationWas the answer right, and right in the language it was spoken in?
- E2Step-level evaluationDid the agent take consent, disclose, and route audio only where permitted?
- E3Tool evaluationDid it read and write the correct booking and the correct field?
- E4Confidence calibrationDo low-confidence turns actually attract more colleague handovers?
- E5Slice evaluationHow does performance change across specific languages and call types?
- E6Business outcomeHow many calls needed a handover or a correction after they ended?
Floor — the call the operator answers for
Failure modes
Where each failure originates in the agent
Seven ways a call goes wrong, set at its own stage.
Agent lifecycle
Direction of processing →
01 · Retrieval
1 mode
- CJ-03
Consent state unread
A call proceeds before the consent state has been resolved.
Stage gathers
Consent, disclosure, language and the booking
02 · Reasoning
2 modes
- CJ-04
Translated claim drifts
A rate or a rule changes meaning in translation.
- CJ-06
Human identity implied
The agent lets a caller believe a person is speaking.
Stage proposes
The turn the caller hears and relies on
03 · Tool / write
2 modes
- CJ-02
Relay call timed out
Silence tuning drops a relay or a TTY caller.
- CJ-05
Audio on a shared path
Call audio reaches a route that could train a model.
Stage writes
Only where write access and approval policy allow it
04 · Output
1 mode
- CJ-01
Disclosure not given
The first turn passes without the automation being named.
Stage returns
The words the caller hears and acts on
05 · Change / Version
1 mode
- CJ-07
Silent language regression
A model or rule change widens which languages the agent answers.
Stage tracks
Model, prompt, consent rules and language config
- Sev-1 · audio taken without consent
- Sev-2 · a wrong answer reaches a caller
- Sev-3 · language degrades, call to a person
Affected slices
One language can carry the whole rate
Break the call-quality figure down before you trust it. Split it by language, by consent state and by how the caller reached the line The cohorts that carry it are named, not averaged away..
Slice performance — reported separately, not only in aggregate
Illustrative example| Slice | Failure rate | Lift | Lift vs. threshold | Status |
|---|---|---|---|---|
| Relay and TTY calls | 5.1% | 3.9× | Review | |
| Unevaluated languages | 2.9% | 2.2× | Review | |
| Code-switched speech | 2.2% | 1.7× | Watch | |
| Plain English enquiries | 0.8% | 0.6× | Normal |
Bar: colleague-handover-rate lift vs. plain-English baseline · scale 0–4.0× · tick marks the 2.0× review threshold
2 of 4 slices over threshold
Evidence-linked improvement
The cycle ends in a standing test
Explaining a bad call does not close the cycle. A standing regression case does, and that suite is what the next call handled is measured against.
Improvement cycle · five stages
Switchback — the path turns at Improve and returns at Learn
- 01
Detect
Colleague-handover rate rises in a call slice.
- 02
Diagnose
Within the same shift, the recordings, the transcripts and the consent states behind them are read until the cause narrows to one.
- 03
Improve
The fix is versioned against the calls that produced it.
- 04
Verify
The release stops until every affected case is passing.
- 05
Learn
The case stays in the suite, and the calling rules are amended.
Learn → Detect
The return edge. The following cycle begins against a longer suite.
Typical build scope
Twelve workstreams across six weeks
The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, sources, calling workflow, evaluation, integration, then production validation and handover.
Workstream
Week 1Week 2Week 3Week 4Week 5Week 6
- 01Call handling workflow discovery and boundary setting.Week 1
- 02Contact-platform and telephony assessment.Week 1, Week 2
- 03Consent, disclosure and language-coverage rule mapping.Week 1, Week 2
- 04Call ingestion and consent capture.Week 2
- 05Answering logic and record binding.Week 2, Week 3
- 06Confidence scoring and handover routing.Week 3
- 07Colleague handover workflow.Week 3, Week 4
- 08Contact-platform and telephony integration.Week 5
- 09Disclosure and language cases.Week 4, Week 5
- 10Guardrails and consent controls.Week 4
- 11Call-trail instrumentation.Week 5, Week 6
- 12Deployment, documentation and Agent Care handover.Week 6
12 workstreams · 6 weeks · bar shows the weeks a workstream is active — several run in parallel
Final scope and sequence confirmed in discovery
Engagement tiers
What each tier includes
Rows are the capabilities named in each tier's scope. Higher tiers include everything below them.
| Capability✓ in scope · — not at this tier | PilotOne line, one language | ProductionProduction contact platforms | AdvancedMultiple markets / languages |
|---|---|---|---|
| Introduced at Pilot | |||
| Answering from your records | ✓ | ✓ | ✓ |
| Colleague handover | ✓ | ✓ | ✓ |
| Call-quality baseline | ✓ | ✓ | ✓ |
| Introduced at Production | |||
| Reporting by language | — | ✓ | ✓ |
| Approval workflow in your systems | — | ✓ | ✓ |
| Approved write-back | — | ✓ | ✓ |
| Contact-platform integration | — | ✓ | ✓ |
| Introduced at Advanced | |||
| Multi-market consent rules | — | — | ✓ |
| Multi-stage floor approvals | — | — | ✓ |
| High call volume | — | — | ✓ |
| Multi-market calling controls | — | — | ✓ |
| Build price | From $5,000 | From $8,000 | Custom quote |
- Final build price
- Confirmed after discovery based on integrations, workflow complexity, transaction volume, approval controls and deployment requirements.
- Separate from build
- Build pricing is separate from recurring Agent Care, which covers managed monitoring, evaluations, incidents and verified improvements after launch.
What we need from you
What you bring, and what we build with it
Each input maps to a piece of build scope and a week in the delivery timeline.
You bring
We build with it
01Your consent posture and recording policy
→Call ingestion and consent-state capture
Week 102Calls you have already handled
→Answering baseline, language scoping and record binding
Week 203The languages you need, and who covers the rest
→Consent, disclosure and language-coverage rule mapping
Week 104Access to relevant APIs, feeds or exports
→Contact, telephony and property assessment, then integration setup
Week 205Calls you would not want replayed
→Language cases and the evaluation suite
Week 406What audio may never be used for
→Confidence scoring, handover routing, guardrails and consent controls
Week 307Named colleagues to take handed-over calls
→Handover workflow, then pilot and production validation
Weeks 5–6
Nothing else is required
Deployment, documentation and Agent Care handover are ours.
Delivery timeline
Four phases across six weeks
Phases occupy the weeks the work really needs, so week 5 runs evaluation and launch side by side.
Phase
W1
W2
W3
W4
W5
W6
- DiscoveryW1
- BuildW2 – W3
- EvaluateW4 – W5
- Pilot & LaunchW5 – W6
Week focus
W1Call workflow discovery, consent mapping and the automation boundary
W2Source integration and the answering baseline
W3Calling workflow, confidence logic and handover controls
W4Language cases, consent guardrails and failure-mode testing
W5Contact-platform integration, pilot calls and targeted corrections
W6One contact period run under the centre team, then handover
- Reading the band
- The bars follow real work rather than a plan, which is why week 5 carries two kinds rather than padding.
- At the end of W6
- The period closes validation and Agent Care picks up the monitoring.
- Duration
- Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.
Next step · Travel & Hospitality AI agent
Build a voice agent around consent, architecture and the person behind it.
A court let eavesdropping claims run against a cloud contact-centre AI because it could use call audio to improve a model, with nothing alleged about whether it ever did. So bring what a contract cannot supply: a diagram of where the audio actually goes, the states your callers dial from, and the languages you would stand behind. California's bot statute reaches only what happens online; Maine's reaches a voice, and the European transparency duty is in force now.
- Typical delivery
- 4–6 weeks
- Deployment
- Your call stack
- Human handover
- All-party
- After launch
- Agent Care available
Nestack Agents · Voice contact centre
AGT-TH-06 · Agent Care available after launch