Skip to main content
TRV-03 · Travel & Hospitality

Hospitality Staff Copilot

Outcome​

Equip frontline hospitality staff to resolve guest requests on the spot using property knowledge, standard operating procedures, reservation context, and guest preferences, with manager approval for compensation beyond service authority.

Portfolio attributeValue
Primary personaFrontline employee
Primary aggregateGuestServiceRequest
Cosmos DB containers3
Grounding dataPMS and CRS reservations, room state, loyalty and consented preferences, contact-center history, property knowledge, SOPs, housekeeping and engineering work orders, point-of-sale and partner events
Development estimateUSD 60-220 monthly dev estimate
Runtime optionsPython FastAPI and C# ASP.NET Core

Productivity hypothesis​

These measures reflect the workflow, evidence, decision boundary, and industry risk for this use case.

Impact typeMeasureConservative internal assumptionCustomer measurement
ProductivityGuest issue response time15-25% reduction(baseline median guest-issue response minutes - assisted median minutes) / baseline median minutes * 100
Quality or riskSLA breach or compensation leakage rate5-10% reduction(baseline SLA-breach or compensation-leakage rate - assisted rate) / baseline rate * 100

Important: Conservative internal pilot assumptions only; not public benchmarks, commitments, or expected customer results. Validate against a comparable baseline or matched control cohort with mandatory human review.

Rationale: Room, work-order, guest, and SOP context supports faster safe frontline action.

Pilot guardrail: Managers approve compensation above authority limits, and the assistant must not make silent property-system changes.

Customer validation plan​

  1. Baseline the current workflow with representative users and tasks.
  2. Run the accelerator on the same task mix with human review retained.
  3. Measure guest issue response time and sla breach or compensation leakage rate with the documented formulas.
  4. Replace internal assumptions with customer-observed results before a business case.

Why Azure Cosmos DB​

This AI solution needs an operational data plane, not only a vector index:

  • Hybrid vector and full-text retrieval combines semantic recall with exact identifiers, codes, names, and domain language.
  • Hierarchical partition keys and metadata predicates keep tenant and domain boundaries in the normal query path.
  • Operational JSON, source chunks, workflow and agent state, feedback, and audit records remain together for low-latency reads and governed writes.
  • Change feed reacts to source and operational updates without polling.
  • Autoscale throughput, configurable consistency, and multi-region distribution support variable demand and enterprise availability.

Domain workflow​

Domain workflow

StageActorInput → outputApproval
1. Capture the guest requestFrontline employeeGuest message, call, or in-person request → GuestServiceRequest with room and urgencyPolicy controlled
2. Load reservation and guest preferencesStay context serviceReservation identifier and guest profile → Stay context with inclusions and preferencesPolicy controlled
3. Retrieve property knowledge and SOPRetrieval agentRequest kind and property scope → Verified property facts and applicable SOP stepsPolicy controlled
4. Guide the resolution at the point of serviceHospitality copilotSOP steps, inclusions, and preferences → Cited response with the staff member's authority limit shownPolicy controlled
5. Approve compensation beyond authorityDuty managerProposed ServiceRecoveryAction and value → Approved gesture posted to the folio with justificationHuman required

Technical architecture​

Technical architecture

Cosmos DB data model​

ContainerHierarchical partition keyDocumentsSearch paths
guestServiceRequests/tenantId, /propertyIdGuestServiceRequest, ReservationContext, PropertyKnowledgeItem, ServiceRecoveryAction/requestTextVectorEmbedding, /bodyVectorEmbedding, /requestTextVector, /bodyVector
hospitalitySopLibrary/tenantId, /requestKindStandardOperatingProcedure/stepVectorEmbedding, /stepVector
guestProfileStore/tenantId, /guestProfileIdGuestPreferenceRecordOperational fields only

Query contract​

QueryContainerScopePurpose
In-stay service contextguestServiceRequestssingle-partitionRead the open requests and the reservation context for the stay in one partition-aligned query. Guest preferences follow the guest across properties, so they are read from the guestProfileStore container by guestProfileId instead of being duplicated into every property partition.
Currently valid property factsguestServiceRequestssingle-partitionReturn only property knowledge that is in its validity window and recently verified.
Request-kind SOP retrievalhospitalitySopLibrarysingle-partitionKnowledge-container query against hospitalitySopLibrary (partition-aligned to /tenantId + /requestKind): pull the SOP steps and authority limits for the specific request type; the library stores one SOP document per request kind so the lookup is a single partition read.
Pending manager approvalsguestServiceRequestssingle-partitionShow service recovery gestures waiting on duty-manager sign-off.

Python and C# services route each document type to its declared container and hierarchical partition key. ETag-based writes protect mutable aggregates, while append-only records and bounded cross-partition queries preserve scale.

Azure deployment​

Azure deployment

Clone the source repository, authenticate with Azure, and run:

Set-Location accelerators/travel-hospitality/hospitality-staff-copilot
az login
./Deploy.ps1 -Runtime python -Location eastus2

Change python to csharp to deploy the .NET implementation.

Explore the implementation​

Return to all accelerators