Travel & Hospitality AI agent · Guest services
Guest-Services Copilot AI Agent
Answer routine guest requests from the record they can be answered from, recognise the questions a copilot may not resolve, and put those in front of a named colleague on shift.
- Typical delivery
- 4–6 weeks
- Deployment
- Your stack
- Person named
- Advice only
- After launch
- Agent Care
What this agent does
Drafts the answer, or fetches the person
In
- 01
Ingesting the request, the stay record and the property's own escalation rules from supported systems.
- 02
Normalising what is actually being asked, so a welfare question is not read as a housekeeping one.
Reason
- 03
Drafting an answer only where a record on file supports it, and naming that record beside the answer.
- 04
Recognising the questions a copilot may not resolve — safety, security, medical, police, refusal.
- 05
Raising any such question to a named colleague on shift, with the clock running and the routing logged.
Decide
- 06
Writing an access or health note into a classified field rather than into a free-text note.
- 07
Holding the drafted answer for the colleague who sends it, on the property's own terms.
Out
- 08
Retaining the request, the record, the flags, the amendments and the words actually shown.
- 09
Executing write actions only inside the approval boundaries agreed during implementation.
Product statement
The agent drafts an answer or fetches a person; the property and its staff keep the duty of care, and the colleague on shift decides.
Example workflow
One request, record to answer
- 1Request receivedMessage, call, in-app note, kiosk session or a word at the desk
- 2Stay assembledFolio, stay notes, preferences, access needs and the escalation list for this property
- 3Answer draftedDraft answer, its source record, confidence
- 4Controls appliedEscalation rules, disclosure limits, classified-field checks and confidence threshold
No human action required
Stages 1 to 4 run unaided, and nothing is sent at any of them — the agent is drafting, and the colleague's lane opens at the confidence gate.
Goes to the colleague on shift to send.
Adds a duty-manager read first.
Colleague sends it
The answer is held with the record behind it, its flagged lines and the confidence.
Send · Amend · Escalate to the duty manager
Sent — the guest has the answer
- 6Property systems updatedOnly where write access and approval policy allow it
- 7Outcome evaluatedAnswer corrections, flagged-line outcomes, escalation timings and repeat contacts
Amendments
Every colleague amendment is counted in the evaluation.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required
8 items
- Answering a safety, security or medical question.
- Deciding whether to let a person into a room.
- Confirming that a named person is staying here.
- Responding to a request from law enforcement.
Automation boundary
Agent acts unaided
- ✓Answer from the record on file, and show which record.
- ✓Recognise a question the agent may not resolve, and stop.
- ✓Route the request to a named colleague, with the clock on for the named owner.
- ✓Log the exact words shown to the guest and to the colleague.
Any write happens inside the boundaries agreed at implementation, never ahead of approval.
- Refusing a guest, or bringing a stay to an end.
- Making a judgement about a guest's behaviour.
- Granting a refund, a waiver or a goodwill credit.
- Changes to escalation, retention or answer rules.
Example output
One request, annotated
Everything the agent drafts is attached to the stay record it was drawn from.
Guest-services output · single request
Illustrative example- Request
- Welfare concern
- Drafted line
- Not answered — raised to the duty manager named on the roster
- Time to a person
- 6 min
- Source of record
- Property escalation list
- Confidence
- 89%
- Attribution
- Property name and duty roster
As received
Taken from the stay record and the property's escalation list — nothing on this side is written by the agent.
Source records used
- Stay record on file
- Property escalation list
- Duty roster on shift
Why this wording
It fetches a person rather than answering and a person still decides.
What the score decides
Below the configured threshold the answer picks up a duty-manager read before.
Action
SendAmendEscalate to the duty managerValue
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 request
From the stay record
Draft from the record
Draw on the folio, the stay notes on file and the escalation list configured for the property.
Answer it, or fetch a person
Routine requests reach the colleague already drafted and already sourced.
Send review to the raised signals
Every signal the agent raises is put in front of a named colleague, so the read starts where risk concentrates.
Build an evidence trail
The request, the record it was answered from and the colleague who took it stay on the stay.
Integrations
Typical integrations
Five system groups connect to the same agent. Which of them are in scope is decided in discovery.
Property management
- Opera Cloud
- Infor HMS
- Mews
- Cloudbeds
- Apaleo
Guest messaging
- SMS
- in-app
- Kipsu
- Zingle
- Akia
Service and tasks
- HotSOS
- Quore
- Alice
- Knowcross
Agent
Guest-services copilot
- Reads the record
- Drafts the answer
- Fetches a person
Profiles and CRM
- Salesforce
- Revinate
- Cendyn
- Amperity
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 guest
Every layer wraps the next. What none of them catches is in the map below.
L6 · Outermost — last line of defence
Inward → L1 · closest to the model
- L1 – L2
- Test whether an answer may stand
- L3
- Puts the sending in a colleague's hands
- L4 – L5
- Hold the answer and the record behind it
- L6
- Falls back to signposting when signals degrade
How Nestack evaluates it
Evaluate the request workflow — not only the paragraph at the end.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the answer the guest reads
- E1Final-output evaluationWas the answer supported by a record the property actually holds?
- E2Step-level evaluationDid the agent recognise a question it was not permitted to resolve?
- E3Tool evaluationDid it read and write the correct stay and the correct field?
- E4Confidence calibrationDo low-confidence drafts actually attract more colleague amendments?
- E5Slice evaluationHow does performance change across specific request types?
- E6Business outcomeHow many raised signals reached a named person, and how quickly?
Floor — the duty of care the property owes
Failure modes
Where each failure originates in the agent
Seven ways a request goes wrong, by stage.
Agent lifecycle
Direction of processing →
01 · Retrieval
1 mode
- BY-03
Superseded stay record
An answer is read from a folio the stay has already moved past.
Stage gathers
Requests, folios, stay notes and escalation rules
02 · Reasoning
2 modes
- BY-04
Advice given as a decision
A judgement about a guest is written as though it were made.
- BY-06
Policy read out flat
A rule is quoted back with no modification considered at all.
Stage proposes
Draft answer, its source record and confidence
03 · Tool / write
2 modes
- BY-02
Signal logged, nobody named
An observation is recorded with no colleague set to work it.
- BY-05
Sensitive note in free text
A health or access need is written to an unclassified field.
Stage writes
Only where write access and approval policy allow it
04 · Output
1 mode
- BY-01
Guest confirmed to a caller
An unauthenticated caller learns that a named guest is here.
Stage returns
The answer the colleague sends and the property owns
05 · Change / Version
1 mode
- BY-07
Silent escalation regression
A model or rule change widens what the agent will answer.
Stage tracks
Model, prompt, escalation rules and retention config
- Sev-1 · a held question answered
- Sev-2 · a wrong answer reaches a guest
- Sev-3 · source degrades, request goes to a person
Affected slices
Overall quality can hide one bad cohort
An aggregate answer-quality figure and the figure for one slice are different measurements The cohorts that carry it are named, not averaged away Nestack reports it by slice rather than in aggregate..
Slice performance — reported separately, not only in aggregate
Illustrative example| Slice | Failure rate | Lift | Lift vs. threshold | Status |
|---|---|---|---|---|
| Safety and welfare signals | 4.2% | 3.5× | Review | |
| Accessibility requests | 2.8% | 2.3× | Review | |
| Complaint and redress asks | 2.3% | 1.9× | Watch | |
| Routine service requests | 1.1% | 0.9× | Normal |
Bar: colleague-amendment-rate lift vs. routine-request baseline · scale 0–4.0× · tick marks the 2.0× review threshold
2 of 4 slices over threshold
Evidence-linked improvement
Every cycle ends by adding a test
The loop shuts when the miss is a case in the suite, not when it has been explained. That suite is what the next guest request is measured against.
Improvement cycle · five stages
Switchback — the path turns at Improve and returns at Learn
- 01
Detect
Colleague-amendment rate rises in a request slice.
- 02
Diagnose
The colleague who picked the request up is asked what they saw, and the records are read behind them until the cause narrows to one.
- 03
Improve
Whatever changes ships against a version, with the requests that prompted it attached.
- 04
Verify
Nothing ships until the affected cases pass a second time.
- 05
Learn
The suite grows by one case; so does the escalation list.
Learn → Detect
The return edge. Detection next time runs against a suite one case longer.
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, answering workflow, evaluation, integration, then production validation and handover.
Workstream
Week 1Week 2Week 3Week 4Week 5Week 6
- 01Guest-request workflow discovery and boundary setting.Week 1
- 02Property and CRM source assessment.Week 1, Week 2
- 03Escalation, retention and answer rule mapping per site.Week 1, Week 2
- 04Request ingestion and normalisation.Week 2
- 05Answer drafting and source binding.Week 2, Week 3
- 06Confidence scoring and escalation routing.Week 3
- 07Duty-manager escalation workflow.Week 3, Week 4
- 08Property-system integration.Week 5
- 09Safety-signal regression cases.Week 4, Week 5
- 10Guardrails and escalation controls.Week 4
- 11Request-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 property, one team | ProductionProduction property systems | AdvancedMultiple properties / brands |
|---|---|---|---|
| Introduced at Pilot | |||
| Drafting to your records and rules | ✓ | ✓ | ✓ |
| Named-colleague escalation | ✓ | ✓ | ✓ |
| Answer-quality baseline | ✓ | ✓ | ✓ |
| Introduced at Production | |||
| Reporting by request type | — | ✓ | ✓ |
| Approval workflow in your systems | — | ✓ | ✓ |
| Approved write-back | — | ✓ | ✓ |
| Property-system integration | — | ✓ | ✓ |
| Introduced at Advanced | |||
| Multi-brand escalation rules | — | — | ✓ |
| Multi-stage duty approvals | — | — | ✓ |
| High request volume | — | — | ✓ |
| Multi-property service controls | — | — | ✓ |
| Build price | From $5,000 | From $8,000 | Custom quote |
- Final build price
- Confirmed after discovery based on integrations, workflow complexity, request 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 stay records and the notes you keep
→Request ingestion and record mapping
Week 102Requests you have already answered
→Drafting baseline, source binding and answer scoping
Week 203Your escalation list and who is on it
→Escalation, retention and answer rule mapping
Week 104Access to relevant APIs, feeds or exports
→PMS, messaging and CRM assessment, then integration setup
Week 205Answers you would not want given
→Signal cases and failure-mode testing
Week 406What must reach a person on the property
→Confidence scoring, signal routing, guardrails and escalation controls
Week 307Named colleagues to take what is raised
→Escalation 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
The bands follow real work rather than a plan, so evaluation and pilot genuinely share the fifth week.
Phase
W1
W2
W3
W4
W5
W6
- DiscoveryW1
- BuildW2 – W3
- EvaluateW4 – W5
- Pilot & LaunchW5 – W6
Week focus
W1Request workflow discovery, escalation mapping and the automation boundary
W2Source integration and the drafting baseline
W3Answering workflow, confidence logic and escalation controls
W4Safety-signal cases, held-act guardrails and failure-mode testing
W5Property-system integration, pilot stays and targeted corrections
W6One stay cycle answered under the duty manager, then Agent Care handover
- Reading the band
- A bar covers the weeks its work is named in, and nothing else. The week 5 overlap is real, not padding.
- At the end of W6
- Once the cycle validates, Agent Care owns the running agent.
- Duration
- Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.
Next step · Travel & Hospitality AI agent
Build a guest-services copilot around the person it has to fetch.
A copilot that sees a safety signal has made a record, and that record gets read against the property long afterwards, on a standard of what you should have known. So when your copilot raises something at three in the morning, whose name does it reach? Tell us that, and the escalation is built around it.
- Typical delivery
- 4–6 weeks
- Deployment
- Your records
- Person named
- Advice only
- After launch
- Agent Care available
Nestack Agents · Guest services
AGT-TH-03 · Agent Care available after launch