Priceline
agentic genAI booking agent
In 2026, Priceline made its Penny agent fully agentic on the Claude models; travelers who use it save on average nearly 10 minutes per trip and convert more.
Objective
Move travel search from a stack of filters and tabs to a single conversation where the traveler describes their need and the agent compares, weighs trade-offs, and books.
The deployment
Penny is Priceline's AI travel agent. In 2026 it becomes agentic: it understands a complex request (comparing flights to several cities over a given week), evaluates price and availability live, surfaces the trade-offs, and lets travelers book without leaving the conversation. The reasoning relies on Anthropic's Claude models on the Priceline stack hosted on Google Cloud, with OpenAI for voice. Priceline positions Penny on preference memory and its deals technology, not just the model.
Results Proof C
Quantified official Priceline press release, a concordant mention in the Booking Holdings Q1 2026 earnings call, and specialist press coverage (Skift); the hard public figure (time saved) remains a brand claim.
How it works
Documented architectureThe stack in detail
- llm Anthropic Claude Claude models for Penny's conversational reasoning and agentic planning (multi-city comparison, trade-offs).
- llm OpenAI Realtime Real-time voice block for Penny's voice interface.
- infra Google Cloud Hosting for Priceline's AI stack.
- plateforme Stack Priceline in-house Deals engine, traveler preference memory, real-time inventory and pricing, booking flow integrated into the conversation.
How it runs, concretely
For ops teams-
1Stating the need customer
The traveler describes their trip, including complex multi-destination requests, by text or voice.
-
2Reasoning and search AI
The agent plans, queries inventory, and compares options across several cities and dates.
-
3Presenting the trade-offs AI
It surfaces price, availability, and compromises, drawing on preference memory and Priceline deals.
-
4Booking in the thread customer
The traveler confirms and pays without leaving the conversation.
The expressed travel intent, crossed with live inventory and pricing and the traveler's preference memory. Without a reliable price and availability feed, the proposed trade-offs become wrong and conversion drops.
How your customers perceive this type of use
Sourced studiesLes consommateurs n'acceptent pas les chatbots par defaut : 64% prefereraient que les entreprises n'utilisent pas d'IA dans leur service client (Gartner, 2024) et pres d'un utilisateur sur cinq du service client par IA n'en retire aucun benefice (Qualtrics, 2025). L'acceptation se construit sur trois conditions mesurees par Salesforce : savoir qu'on parle a une IA, pouvoir escalader vers un humain, comprendre la logique de l'agent.
Acceptance conditions
- Etre informe qu'on parle a une IA et non a un humain (pres de 75% le demandent, Salesforce 2024)
- Un chemin d'escalade clair vers un agent humain (45% plus enclins a utiliser l'agent IA, Salesforce 2024)
- Une logique de l'agent clairement expliquee (44% plus enclins, Salesforce 2024)
Red lines
- Rendre l'humain injoignable : c'est la premiere inquietude des consommateurs sur l'IA dans le service client (Gartner 2024) et 50% craignent que l'IA les coupe du contact humain (Qualtrics 2025)
- Remplacer le service client par l'IA sans alternative : 53% envisageraient de partir chez un concurrent (Gartner 2024)
Sources: Salesforce 2024 · Gartner 2024 · Qualtrics 2025
How to replicate
Inference, not sourcedData prerequisites
- real-time inventory and pricing via API
- per-user preference memory
- booking history
Org prerequisites
- product AI team
- in-thread payment integration
- governance of the proposed trade-offs
Possible stack
- commercial LLM for reasoning
- existing travel search engine
- real-time voice block
The plan, step by step
- Step 1Frame a narrow use case (flight comparison) and connect a conversational assistant to inventory and pricing via API.Deliverable: Conversational prototype that answers with real inventory data.
- Step 2Add per-user preference memory and trade-off presentation (price, availability, compromises).Deliverable: Agent able to compare several cities and dates with persisted preferences.
- Step 3Integrate booking and payment into the conversation thread, with payment compliance (PSD2/SCA in the EU).Deliverable: Transactional booking flow accessible without leaving the conversation.
- Step 4Add the real-time voice block and open a beta to a group of users.Deliverable: Voice interface in beta with measurement of time saved per trip.
- Step 5Compare agent users to non-users (engagement, conversion, customer-service contacts) and generalize.Deliverable: Quantified read vs the classic journey and a full rollout plan.
First step: Connect a conversational assistant to inventory for a narrow use case (flight comparison), then add in-thread booking.
The same pattern, proven in other industries
Amazon
genAI shopping assistant
Burberry
genAI copilot for consultants (search, image recognition, complete-the-look)
Flipkart
conversational genAI shopping assistant
Sources
- S1 Priceline's Penny Goes Fully Agentic Primary archive pending
- S2 Booking Holdings AI Assistants Slash Costs and Boost Bookings Established press archive pending
- S3 Priceline Gives Penny a Multi-Agent Makeover Established press archive pending
An error, newer info, a source?
This page lives on its accuracy. If a figure has moved, if the deployment has changed, or if you have a higher-quality source, tell us. Every sourced correction is verified before publication.