Guide — Hospitality
Voice AI for Hotels: The Complete Guide for Operators (2026)
The short version: Enterprise-grade voice AI is no longer experimental. Marriott, Hilton, and Hyatt are already running it at scale. The question for independent hotels and mid-market chains is not whether the technology works — it's whether you can access a version of it that fits your operation and your budget.
Published 18 February 2026 — by Paul Gosnell, POSTMAN
Why 2026 Is the Inflection Point for Hotel Voice AI
For years, voice AI in hospitality was a proof-of-concept industry. Pilots ran in a handful of business hotels, results were mixed, and adoption stalled. That changed in 2024 and accelerated sharply into 2025.
The clearest signal is PolyAI. The London-based voice AI company closed an $86M Series D in December 2025 — bringing their total funding to over $200M at a $750M valuation. That is not a company operating on promise. That is institutional capital following demonstrated results at the largest hospitality brands in the world.
PolyAI's "EVA" platform (Evolution Virtual Agent) is deployed at Marriott, Hilton, Hyatt, and Choice Hotels. At these properties, EVA handles 100% of incoming calls — not as a triage layer that routes callers to humans, but as the full resolution layer. Reservation inquiries, booking modifications, loyalty queries, room-level requests: all resolved without a human agent touching the call.
Forrester's commissioned study on PolyAI customers found a 391% ROI and an average of $10.3M in cost savings per enterprise customer. These numbers are not from a vendor press release — Forrester's methodology requires real customer data.
The technology works. The ROI is documented. The enterprise chains have moved. What remains is the mid-market — independent hotels, boutique chains, regional groups — catching up with a deployable, integrated version they can actually access.
The catch: PolyAI sells to enterprise. Their contracts start at £100k+, their sales cycles run 6–12 months, and their implementation requires dedicated procurement teams. If you run one hotel or a small group of properties, they will not return your call.
The opportunity is real. The access gap is real. This guide covers both.
What Hotel Operations Can Be Automated by Voice AI
Before deciding on a vendor or platform, it is worth being precise about what voice AI can and cannot handle well. The technology excels at structured, repeatable conversations with defined outcomes. It struggles with genuinely complex negotiations, serious complaints, and anything that requires empathy at scale.
Here is what your front desk currently handles that a well-built voice agent can take over:
Reservations and Availability
- Availability inquiries for specific dates and room types
- Rate quotes across room categories
- Booking confirmations and reference number lookups
- Booking modifications (date changes, room upgrades)
- Cancellations within policy parameters
- Group booking inquiries and handoff to sales
Pre-Arrival and Check-In
- Check-in and check-out time confirmation
- Early check-in requests and availability checks
- Late check-out requests (with real-time room availability lookup)
- Directions, parking, and transport logistics
- Pre-arrival preference capture (pillow type, floor preference, etc.)
In-Stay Requests
- Room service orders (menu read-out, order placement, delivery tracking)
- Housekeeping requests (fresh towels, turn-down service, do not disturb removal)
- Maintenance tickets (creating, logging, and confirming resolution timeframes)
- Wake-up call scheduling
- In-room amenity queries (minibar, safe, TV remote, Wi-Fi password)
Multilingual Concierge
This is where mid-market hotels often feel the most acute pain. International guests call in a language your front desk staff may not speak fluently. A voice agent built on a modern LLM handles this natively.
- Restaurant recommendations and reservation assistance
- Taxi and transfer bookings
- Local attraction information and activity recommendations
- Shopping and pharmacy directions
- Real-time responses in the caller's language — no hold, no escalation required
Loyalty and Accounts
- Points balance inquiries
- Tier status and benefit explanations
- Points redemption options
- Member rate access
What Voice AI Should Not Handle (Yet)
Be realistic about the boundaries. A voice agent should escalate to a human for: formal complaints that require empathy and compensation decisions, safeguarding situations, complex negotiation (large group contracts), and anything requiring physical verification. A well-designed agent knows these limits and hands off cleanly.
PMS and Systems Integration
A voice agent that cannot talk to your Property Management System is a voice agent that cannot do its job. This is where many hotel technology discussions stay superficial — and where operators need to ask hard questions before signing anything.
The Major PMS Platforms
| PMS | Market Position | API Access | Integration Complexity |
|---|---|---|---|
| Opera (Oracle Hospitality) | Dominant in full-service and luxury hotels | OHIP (Oracle Hospitality Integration Platform) REST API | Medium — well-documented, requires Oracle partner certification |
| Mews | Modern cloud PMS, fast-growing in boutique and lifestyle hotels | Comprehensive REST API, good documentation | Low — developer-friendly, active marketplace |
| Apaleo | API-first platform, popular with tech-forward operators | Full open API, GraphQL and REST | Low — designed for integrations from day one |
| Protel (Sertifi/dormakaba) | Mid-market European hotels | XML-based API, partner programme | Medium-High — older architecture |
Channel Managers and Booking Engines
If your voice agent is going to check availability and take reservations, it needs real-time data from your channel manager — not just your PMS. This typically means integrating with platforms like SiteMinder, Booking.com Connectivity, or your OTA manager of choice. A well-built integration layer reads live inventory and writes confirmed bookings back to prevent double-booking.
How Integration Actually Works
In practice, a voice agent connects to your systems via API calls that happen mid-conversation. When a caller asks "do you have a sea-view room available for the 14th of March?", the agent pauses — imperceptibly — queries your PMS for real-time availability, and reads back an accurate answer. When the caller confirms a booking, the agent writes the reservation back to your PMS and triggers a confirmation email.
The quality of this integration is what separates a useful voice agent from a glorified FAQ bot. When evaluating any vendor, ask specifically: which PMS do you have live integrations with, how is data written back (real-time vs batch), and what happens when your PMS is offline?
For operators on Opera: Oracle's OHIP programme has expanded significantly. Most serious voice AI vendors now offer Opera integration as standard. Verify this in writing — "we can integrate" and "we have a live, tested integration" are very different claims.
Cost and ROI: Real Numbers
The enterprise ROI numbers from Forrester are compelling but they do not translate directly to a 50-room independent hotel. Let us work with realistic mid-market numbers.
The Cost of Doing Nothing
| Cost Component | Range | Notes |
|---|---|---|
| Front desk call handling | £8–15 per call | Labour, management overhead, error cost |
| Missed calls (out-of-hours) | £60–200 per missed booking | ADR-dependent; most hotels miss 15–30% of calls after hours |
| Staff turnover in front desk roles | £2,000–5,000 per replacement | Hospitality turnover is high; recruitment costs accumulate |
| Multilingual gap | Variable | Lost bookings from international callers who cannot communicate clearly |
What a Voice Agent Costs
| Approach | Build Cost | Monthly Running Cost | Typical Contract |
|---|---|---|---|
| PolyAI (enterprise) | Included in contract | £8,000–20,000+/mo | Multi-year, £100k+ ACV |
| VAPI / Bland.ai (self-serve) | £0 platform fee, but you build it | Usage-based (~£0.05–0.10 per minute) | None — but requires technical team |
| Custom build (POSTMAN) | From £5,000 | Hosting + API costs, typically £200–800/mo | No lock-in; you own the agent |
ROI Calculation for a Mid-Market Hotel
Take a 120-room hotel averaging 400 inbound calls per month. Current cost: 400 x £10 average = £4,000/month in call handling labour.
A voice agent handles 70% of those calls autonomously (reservations, FAQs, standard requests). That is 280 calls resolved without staff time. At £10 per call: £2,800/month saved.
Add out-of-hours coverage. If the agent captures 5 additional bookings per month that previously went to voicemail or a competitor — at an average £120 ADR for 2 nights — that is £1,200/month in recovered revenue.
Total monthly benefit: ~£4,000. Against a running cost of £500/month. Payback on a £5,000 build in approximately 2 months.
These are conservative numbers. Hotels with higher call volumes, international guests, or significant out-of-hours demand see faster returns.
Choosing Your Approach
There are three realistic paths to a hotel voice agent. Each suits a different type of operator. Here is an honest breakdown.
PolyAI
Enterprise chains
- ✓Proven at Marriott/Hilton scale
- ✓Deeply integrated hospitality vertical
- ✓Dedicated CS and account management
- ✗£100k+ contracts minimum
- ✗6–12 month sales and implementation cycle
- ✗Not accessible to independent operators
Best for: 50+ property chains, branded hotels with central procurement
VAPI / Bland.ai
Self-serve platforms
- ✓Low platform costs
- ✓Fast to prototype
- ✓No long-term commitment
- ✗Requires a technical team to build and maintain
- ✗No PMS integration out of the box
- ✗You are on your own when it breaks
Best for: Operators with an in-house tech team who want to own the build entirely
Custom Build (POSTMAN)
Mid-market hotels
- ✓Built for your systems — Opera, Mews, Apaleo
- ✓Multilingual from day one
- ✓You own the agent — no ongoing licence fees
- ✓From £5,000 build cost
- ✓Accessible to independent operators
- ✗Requires clear brief on workflows and integrations
Best for: Independent hotels, boutique chains, groups of 2–15 properties
The decision comes down to your size and internal capability. Enterprise chains have the procurement budget and centralised IT to make PolyAI work. Self-serve platforms require a developer on staff who can build and maintain the integration. Custom builds — what we do at POSTMAN — sit in between: a fully built, integrated voice agent delivered to your specification, without the enterprise price tag or the requirement to build it yourself.
See also: AI voice agents vs traditional call centres — a cost comparison and our voice AI platform comparison guide for a deeper look at the underlying technology choices.
A Real-World Example
We are currently building a voice agent for a luxury UK property company experiencing rapid growth. They operate a small collection of high-end properties and had a specific problem: inbound call volume had outpaced their front desk capacity. During peak periods — weekend enquiries, holiday booking windows — calls were going unanswered or callers were waiting several minutes to speak to someone. For a property positioning itself at the luxury end of the market, that was a brand problem as much as an operational one.
The agent we are building handles the initial inbound call in the caller's language. It qualifies booking intent — dates, room type, party size, occasion — and checks live availability against their reservation system. If the caller is ready to book, the agent takes the reservation. If they have questions about the property that go beyond the agent's knowledge, or if they want to discuss a complex request, it transfers to a human with a warm handoff that includes the conversation transcript.
Key features of this particular build:
- English, French, German, and Mandarin — covering the majority of their international enquiry traffic
- Live availability lookup from their PMS, so the agent never quotes inaccurate room status
- Reservation write-back — bookings taken by the agent appear in the PMS immediately
- Out-of-hours handling — the agent operates 24/7, capturing enquiries that previously hit voicemail
- Escalation path — any call the agent cannot confidently resolve goes to a human queue with full context
This is not a pilot. It is going into production as the primary inbound call handler, with the human front desk team shifted from answering routine enquiries to handling complex requests and high-value bookings that require genuine hospitality skills.
Frequently Asked Questions
Most modern implementations are transparent — the agent introduces itself as an AI assistant for the hotel. In our experience, guests care far less about whether they are speaking to a human than about whether they get a fast, accurate answer. Disclosure is also good practice and increasingly an expectation. The agent should identify itself clearly while still sounding natural and helpful.
Modern voice agents built on large language models handle 30+ languages natively. For practical hotel deployments, we recommend identifying your top 4–5 caller languages and testing those specifically. Automatic language detection — where the agent detects what the caller is speaking and responds in kind — is standard in well-built agents.
Opera (via OHIP), Mews, and Apaleo all have well-documented APIs that a custom-built agent can connect to. Protel and some older PMS platforms require more work. The integration is not trivial — it requires a developer to build and test the data layer — but it is entirely achievable. This is exactly what we do at POSTMAN. If you are on a PMS not listed here, ask specifically about their API availability before committing to any build.
A basic voice agent handling FAQs and call routing: 1–2 weeks. A fully integrated agent with PMS connectivity, multilingual support, and reservation write-back: 4–8 weeks depending on API access and testing cycles. Enterprise deployments via PolyAI: 6–12 months. If a vendor promises PMS-integrated, multilingual hotel voice AI in 3 days, ask to see a live reference deployment before signing anything.
A well-designed agent has clear escalation paths. When it encounters a request outside its confidence threshold — a complex complaint, an unusual request, or anything emotionally sensitive — it transfers to a human with a brief summary of the conversation so far. The caller does not have to repeat themselves. This warm-handoff pattern is critical for maintaining service quality.
Yes — and any vendor who waves this away is one you should not work with. Voice calls may be recorded for training and QA. Callers should be informed. Personal data (name, booking details) processed by the agent needs to be handled in line with your existing GDPR obligations. A responsible build specifies data retention policies, minimises what data is stored, and ensures callers are informed. Ask your vendor explicitly about data residency and processing agreements.
In February 2026, Anthropic announced WebMCP — a standard that allows AI agents to interact with web-based tools and services via a browser context. For hotel voice agents, this is worth watching: it points toward a future where a voice agent can interact with booking engines, OTA extranet portals, and web-based PMS interfaces that do not have formal APIs. That lowers the integration barrier significantly for hotels on older systems. It is early-stage, but the direction of travel is clearly toward simpler AI-to-system connectivity.
Yes. For hotel groups or management companies, a single voice agent can be configured to recognise which property the caller is contacting (via the inbound number) and switch context accordingly — different room types, different rates, different FAQs, different escalation paths. This is more efficient than building and maintaining separate agents per property and gives central oversight of call handling quality across the portfolio.
Next Steps
If you are a hotel operator actively evaluating voice AI, here is what we would suggest:
- Audit your current call volume. Look at your PMS and phone system logs. How many calls per month? What percentage are routine enquiries vs complex requests? What is your out-of-hours miss rate? This gives you the baseline for any ROI calculation.
- Identify your PMS and its API status. Call your PMS provider and ask specifically: do you have a REST API, what does it expose, and do you have an approved integration partner programme? This determines what is technically possible.
- Be clear about your language requirements. Which languages do your international callers speak? This shapes the voice model and testing requirements.
- Define your escalation policy. What calls must always reach a human? What is the maximum wait time you are comfortable with for escalated calls? Build this into the brief.
- Talk to someone who has built this before. Not a platform vendor trying to sign you to a SaaS contract — someone who has designed and shipped a hotel voice agent, understands PMS integration, and can tell you where the real complexity sits.
We have built voice agents for hospitality operators. We know where the integrations are straightforward and where they are not. We know what a caller-facing agent needs to handle to actually reduce front desk load rather than just reroute it.
If you want a straightforward conversation about what a voice agent would look like for your property — including an honest assessment of timeline, cost, and the limitations of the technology — get in touch. No pitch deck. Just a direct conversation about your operation.
You can also read our hospitality AI overview or explore how voice AI stacks up against traditional call centres for more context before you commit to any direction.