Accor doubles guest booking intent and lifts CSAT to 89% across multiple hotel properties

Accor’s problem became visible the moment it went live on chat. Within weeks of deploying web chat and WhatsApp, human agents were fielding more volume than the channels were built to absorb. Accor hadn’t miscalculated. It had simply made itself easier to reach, and guests responded. 

The volume itself was manageable. What wasn’t manageable was the nature of the questions. A guest asking about subscription card benefits needed an answer tailored to their membership level. A guest asking about room availability needed information accurate to a specific property, on a specific date, under a specific brand. None of that could come from a static FAQ. And none of it should have been going to a human agent in the first place. This created several compounding problems: 

• Complexity that scaled with the portfolio: With 45+ brands and policies that differed by property, a rigid chatbot couldn’t stay accurate across the full range of guest questions. What was true for a Rixos property was not true for a Fairmont property, and a canned response that ignored that distinction frustrated guests rather than helping them. 
• Multi-part questions with no clean path: Guests rarely asked single, tidy questions. “Do you have meeting rooms, and is the wifi free?” required pulling from two separate pieces of property information simultaneously. A keyword-matching system would either fail or send the conversation to a human agent unnecessarily. 
• Human agents doing pre-booking work: Inquiries about amenities, room types, loyalty card comparisons, and local activities occupied the same queue as complaints, member service issues, and upsell conversations. The team’s expertise was going toward the lowest-value work in the queue. 
• An in-house AI investment with nowhere to go: Accor had built its own large language models for translation and other use cases. Without a platform that could accommodate them, there was no way to put that investment to work in service of guest conversations. 

Accor and Quiq built Allison to handle guest inquiries across web chat and WhatsApp. Rather than decision trees or keyword matching, Allison uses large language model reasoning to evaluate each guest message in context and generate a response grounded in accurate, real-time property information. The system was designed to: 

• Apply agentic reasoning with Process Guides: For each guest interaction, Allison selects the appropriate “Process Guide” — a set of reasoning steps, policies, and best practices matched to the type of inquiry. A loyalty card question routes to a different guide than a room availability question. Allison applies the right framework before generating any response, which is what allows it to handle nuanced, multi-part questions accurately rather than defaulting to an escalation. 
• Pull real-time property data through the Properties API: Allison connects directly to Accor’s Properties API, giving it access to live information on room types, availability, activities, and services. Responses are not pulled from a static document. They reflect what is actually available at the specific property the guest is asking about, at the time they are asking. 
• Integrate Accor’s own large language models through AI Studio: Accor had already built its own large language models for language translation. Quiq’s AI Studio allowed Accor to plug those models directly into Allison’s workflow. Accor retained control over its own AI investment rather than having it replaced by a third-party model, and can continue to evolve both independently. 
• Hand off fully qualified conversations to human agents: When a guest conversation genuinely requires human involvement, Allison does not pass a blank chat thread to an agent. It passes the full conversation history, the guest’s stated intent, their expressed preferences, and relevant booking context. The human agent picks up a qualified conversation, not an empty one. 

From a guest’s perspective, the difference is in how specific the answer is. A few examples from real use:

• Subscription card comparisons: When a guest asked which of Accor’s subscription cards would work best for a family of four traveling through Europe with young children, Allison drew on knowledge across the portfolio to explain the differences between relevant card options, including discount tiers, status nights, and room availability guarantees. It then offered to connect the guest with a specialist for more personalized guidance. 
• Multi-part upgrade and availability questions: When a guest asked “What are the best rooms for a family with kids, and can we upgrade at check-in?” Allison understood both parts of the question, considered property-specific options, and provided a coherent response covering room types and the upgrade process. Rather than two disconnected answers or an escalation, the guest got everything they needed in a single reply.
• Qualified handoffs to human agents: When a guest asked about family room availability and mentioned traveling with two children, Allison gathered their preferences before escalating. The human agent who received the conversation saw the full history alongside a summary of what the guest had already shared, and could open with specific options rather than starting from the beginning and making the guest repeat themselves. 

For Accor’s human agents, Allison changed what the job looks like. Before Allison, every escalated chat arrived with no context. Agents would spend the first several exchanges gathering information the guest had already provided. Now, human agents inherit conversations that Allison has already qualified, and they spend their time resolving rather than catching up. 

For the operations team, Quiq’s AI Studio provides a centralized view of every interaction across channels and properties. Accor can track what guests are asking about, identify gaps in property knowledge, and refine Allison’s Process Guides based on what is and isn’t working. When a property updates its room configurations or policies, the change flows through to Allison’s responses without requiring manual updates to individual bot scripts. 

By deploying Allison across four hotel properties, Accor turned a guest service bottleneck into a conversion advantage. Guests who previously waited for a human agent to answer a pre-booking question now get an accurate, property-specific answer in seconds. Human agents who previously handled FAQ-level inquiries now focus on guests who genuinely need them. 

Key outcomes: 

• 2x increase in booking intent: measured by click-outs on booking links, meaning guests who interacted with Allison went on to book at twice the rate of those using the previous approach 
• 89% CSAT for AI-assisted interactions across Rixos properties: up from 67%, reflecting guests’ confidence in Allison’s accuracy and relevance 
• 80% AI accuracy rate: up from 46%, meaning the large majority of responses now give guests correct property information without requiring an escalation
• Fully qualified human agent handoffs: with every escalation including full conversation history and guest context, eliminating the cold-start problem that had been adding time and friction to every escalated interaction 
• Accor’s own large language models are now active in guest conversations: putting an existing investment to work rather than replacing it