Prism for travel and hospitality

Take your guests off hold

Prism answers and finishes the calls your guests and travelers make: reservation changes, cancellations, rebooking, and loyalty requests. They get an answer in seconds, and your staff get back to the people in front of them.

A traveler on a phone call at an airport gate, feet up on his suitcase
Prism · cancellation call
0:45

GuestYeah, I need to cancel a hotel reservation.

PrismHappy to help. What's the confirmation number for the booking?

tool calls
find_reservation · King room, Fri 12th
check_cancellation_policy · inside the window, no fee
cancel_reservation · AB4729

Hospitality on hold: what 71 calls tell us about the guest phone experience

We sent a personal agent to cancel 54 hotel bookings, the way a guest would. Real hotels, real phone lines, and the agent said it was AI on every call. This is what it heard before anyone said hello.

Read the full report

The method, every outcome, what the brands publish (and don't) about agents calling on a guest's behalf, and what it means for your phone line. No form.

Download the report (PDF)

The calls Prism takes off your queue

Press play to hear each one.

A man at home on a phone call
Play
0:45

Reservation changes and cancellations

Verifies the booking, checks the policy, and makes the change in your booking system while the guest is still on the line.

A traveler silhouetted at an airport window as a plane takes off
Play
0:40

Disruption rebooking

When a flight cancels or a property oversells, Prism calls travelers first, offers options, and rebooks them.

A guest walking through a quiet hotel lobby
Play
0:35

After-hours and overflow

Picks up when the front desk or contact center can't, so a busy night doesn't turn into a voicemail or a lost booking.

A business traveler with a suitcase on a train platform
Play
0:30

Loyalty member service

Looks up the member, answers points and status questions, and treats your best customers like it knows them.

An open suitcase packed with sweaters and headphones
Play
0:30

Pre-trip confirmations

Outbound calls before arrival or departure to confirm plans, collect details, and offer upgrades.

A phone left on a cafe table beside coffee
Play
0:50

New bookings

Answers rate and availability questions and books the room on the first call, before the caller gives up and books through an OTA.

Your people, back at the front desk

Every call Prism finishes is time your team spends with the guest in front of them. The phone stops pulling staff away from the lobby, and hospitality goes back to happening in person.

A front desk agent checking in a smiling guest

Prism speaks first in half the time

0.57 seconds to first audio, measured on the same harness as the other speech-to-speech models. Public benchmarks are where we start; the pilot measures Prism on your own calls.

0.4 s 0.6 s 0.8 s 1.0 s 1.2 s 1.4 s 20% 30% 40% 50% 60% 70% Time to first audio · ◀ faster τ-Voice task completion · better ▶ Faster and more accurate Prism56.6% · 0.57 s GPT-Live-1 (Sol, low)59.3% · 1.24 s · reasoning on GPT-Realtime-2.1 High45.7% · 1.21 s · reasoning on Gemini 3.8 Live30.1% · 1.18 s · reasoning off Reasoning offReasoning on 0.4 s 0.6 s 0.8 s 1.0 s 1.2 s 1.4 s 20% 30% 40% 50% 60% 70% Time to first audio · ◀ faster τ-Voice task completion · better ▶ Faster and more accurate Prism56.6% · 0.57 s GPT-Live-1 (Sol, low)59.3% · 1.24 s GPT-Realtime-2.1 High45.7% · 1.21 s Gemini 3.8 Live30.1% · 1.18 s Reasoning offReasoning on

Each dot is a speech-to-speech model scored on τ-Voice, the voice agent benchmark developed by Sierra. Up is more accurate, left is faster to start speaking, and rings mark models that had reasoning turned on. Prism gets to the top-left corner without it.

One model, with a checkpoint before every word

Prism listens to the caller's audio and writes its reply as text before it speaks. Your rules check that text, so nothing off-policy reaches a guest.

Guardrails you can read

Policy rules run on the text checkpoint, and every decision is logged, so you can see exactly why the agent said what it said.

Tuned on your calls

RL fine-tuning on your own calls and rubrics, so Prism learns your policies, properties, and the way your guests talk.

Runs where you need it

On-prem, in your VPC, or in the cloud. HIPAA and SOC 2. Works with LiveKit, Pipecat, and Twilio.

Start building

Get an API key
reservation_agent.py
[Snippet from the Prism quickstart: one session, a hotel system prompt, and one tool definition (cancel_reservation)]

Questions

Doesn't a text step make it a cascade?
No. A cascade reasons over a transcript and loses the caller's tone. Prism reasons over the audio and emits text so your rules can review every word before it becomes speech. The text is the output checkpoint, not the input to the reasoning.
Where does Prism run?
On-prem, in your VPC, or in the cloud. HIPAA and SOC 2. Telephony-ready.
What does a pilot look like?
We measure your current calls, fine-tune Prism on them, and measure again. The pilot succeeds when Prism beats your current setup on your own rubrics.

Run a pilot on your own calls

We measure your current calls, fine-tune Prism on them, and measure again. The pilot succeeds when Prism beats your current setup on your own rubrics.