Customer Support Automation

AI automates customer support by answering routine questions, surfacing account or case context, routing work, and escalating to humans when needed.

Business impact

• Inquiries resolved without human intervention — Airbnb reported 45% of user inquiries resolved without human intervention.
• Share of CX cases solved by AI — Cisco reported 25% of CX cases solved by AI.
• Customer satisfaction (CSAT) — CoreWeave reported CSAT scores of 4.9-5.0.
• Support cost — AT&T reported call-center analytics cost decreased by 84%, and P&G’s vendor-reported result was a 25% overall customer-service cost reduction.
• Response accuracy — AT&T reported up to 40% improvement in AI-agent response accuracy after post-training.
• Resolution volume — Expedia reported 143 million conversations a year handled by the AI customer-service agent, and over 50% of traveler queries resolved by it.
• Manual tasks automated — P&G’s vendor-reported result said previously manual customer-service rep tasks were automated by over 80%.

Data requirements

• Customer inquiries (text) — The systems process support questions and chat-style requests from customers.
• Ticket context (text) — CoreWeave’s workflow gathers context from support tickets.
• Support interaction history (text) — AT&T and Bank of America describe workflows that summarize or evaluate prior interactions.
• Knowledge-base content (text) — Ford’s agent queries knowledge bases before answering.
• Operational system data (tabular) — AMD’s chatbot retrieves SAP data on order scheduling and deliveries.

AI methods and techniques

• Retrieval-augmented answering — Used to pull relevant information from knowledge bases or operational systems before generating a reply.
• Summarization — Used in live-agent tools such as EricaAssist to condense the interaction and highlight what matters next.
• Classification and routing — Used to identify the right queue or resolution path for a support case.
• Agentic workflow automation — Used in support workflows that can gather context, recommend next steps, or complete routine actions with limited human input.
• Fine-tuning — AT&T’s process includes fine-tuning customer-care AI agents on curated training data.
• Evaluation — AT&T evaluates responses before deployment to check performance and quality.
• Data curation — AT&T curates training data to improve the support agents it deploys.

AI models and model families

Mistral 7B

Industries

Real-world evidence

15 documented case studies on record.

Companies using this: AT&T, AMD, Airbnb, Bank of America, Cisco, CoreWeave, Expedia, Ford, Intel, Komatsu, Pinecone, P&G, Salesforce, Philips, WestJet.