Customer Self-Service

Customer support agents handle routine questions in chat or voice, and route unresolved issues to live staff.

Business impact

• Response latency — DoorDash reports response latency with Anthropic's Claude 3 Haiku at 2.5 seconds or less.
• Topics managed in the bot — Rabobank reports 300 to 400 topics across its two text-based agents.
• Website URLs used as knowledge source — Rabobank describes more than 8,000 URLs on its website as part of the knowledge-source challenge.

Data requirements

• Help-center content (text) — DoorDash uses help-center content to answer routine support inquiries.
• Customer questions and conversation intents (text) — The agents analyze what customers ask in chat or voice to route or answer the request.
• Website knowledge sources and URLs (text) — Rabobank uses its website content as a knowledge source, including a large set of URLs.
• Conversation and call data (audio) — Voice interactions are processed through speech-to-text and text-to-speech services.

AI methods and techniques

• Intent recognition — Used to identify why a customer is calling or chatting and route the request to the right path or team.
• Generative AI — Used in fallback handling when the customer inquiry cannot be solved by the standard path.
• Retrieval from knowledge content — DoorDash uses help-center content to provide answers to routine inquiries.

Industries

Real-world evidence

4 documented case studies on record.

Companies using this: DoorDash, Ford, Salesforce, Rabobank.