How OpenTable Uses Agentforce to Resolve 70% of Customer Inquiries

OpenTable connects 60,000 restaurants with 1.9 billion diner reservations annually, operating a dual-sided marketplace that requires customer support for both restaurant owners managing their operations and diners searching for and booking tables. As scale grew, the burden on customer service teams from repetitive, low-complexity inquiries — opening hours, menus, addresses — consumed capacity that could be better directed at complex issues.

OpenTable worked with Salesforce to deploy Agentforce as an autonomous AI agent layer across both sides of its marketplace. The implementation grounded agents on a knowledge base of 1,500 articles stored in Salesforce Service Cloud, with Salesforce Data Cloud providing retrieval-augmented generation to surface precise, contextual answers. A key innovation was a real-time deflection scoring system: the Atlas Reasoning Engine dynamically scores each conversation based on sentiment and intent — a request for help scores 5, a request for a human scores 10, extreme frustration scores up to 20 — and Agentforce uses that score to decide whether to continue handling, create a case, or escalate to a human agent with a full transcript.

The restaurant agent went live in just 3 weeks on the Contact Us page. The diner agent followed shortly after. Both were designed to handle common high-volume queries automatically while ensuring complex or emotionally charged interactions reached human agents with full context already in hand.

Results were rapid: within weeks of launch, 73% of restaurant web queries were being handled by the agent, and 70% of diner and restaurant inquiries were resolved autonomously — a 40% improvement over OpenTable's prior chatbot. The system is projected to resolve 180,000 cases per year, freeing service teams to focus on the 30% of interactions that genuinely require human judgement.