Conference Presentation, Keynote
Cresta: The Path to 90% Service Automation: Pitfalls, Secrets & Lessons from Enterprise Deployments
- Speaker and Context: Ping Wu, CEO of Cuesta, an AI platform designed to transform contact centers by increasing revenue and service efficiency.
- Core Thesis: While AI has automated high-skill tasks like coding and marketing, automating contact center roles is harder due to data complexity, legacy infrastructure, and the lack of verifiable outcomes.
- Coding vs. Support Automation Comparison:
- Data Source: Coding relies on public, evergreen datasets (e.g., GitHub), whereas customer support relies on private, rapidly changing, and company-specific rules.
- Contextual Complexity: Coding context is contained within the code and digital tools; support requires multimodal context including screen visuals, CRM data, and real-time agent actions.
- Tooling: Coding utilizes standardized tools with modern APIs; Fortune 500 support often relies on legacy systems without API access.
- Verification: Code output is instantly verifiable and testable for reinforcement learning; customer service outcomes are difficult to verify automatically in real-time.
- Redefining "X% Automation":
- Automation percentage refers to task contribution within a conversation, not the percentage of calls handled end-to-end.
- A 90% automation metric implies AI performs 90% of the sub-tasks (e.g., typing, data entry, ticket creation) within a conversation, even if a human is present.
- Current Industry Landscape:
- Most contact centers currently operate on the "no AI" or "human-led" end of the spectrum due to legacy on-premise infrastructure.
- Current adoption is slowly moving toward AI-assisted summaries and after-call work, with only a minority utilizing fully AI-led chatbots or voicebots.
- Future Outlook (5–10 Years):
- Projections suggest the majority of contact center tasks will shift to AI, freeing humans from repetitive actions.
- Critical enablers for this shift include improved instruction-following models, better knowledge base hygiene, and API-first integration for AI actions.
- Cuesta's Three-Component Solution:
- Conversation Intelligence Platform: Analyzes all interactions (human and AI) to identify automation opportunities and learn agent workflows; currently deployed at United Airlines.
- Agent Assistant: A "Copilot" for human agents that suggests actions, transcribes in real-time, and automates sub-tasks like data entry while learning from human behavior.
- AI Agent: Native agents capable of complex, turn-by-turn troubleshooting across multiple channels (e.g., voice and chat), continuously learning from mistakes and human recovery steps.
- Strategic Shift in CX:
- Reaching 70–90% automation will lower interaction costs, enabling a shift from "defensive" CX to proactive, high-volume customer engagement.
- The goal is an "age of abundance" where AI allows businesses to interact with customers more frequently rather than reducing contact.
- Handling Nuance and Transparency:
- Cuesta plans to explicitly inform callers they are interacting with an AI agent, relying on natural voice generation to build trust.
- Agents can present consumers with a choice: wait X minutes for a human or resolve the issue immediately with AI.
- High-Value Customer Escalation:
- The system can identify premium customers based on phone numbers or purchase history and bypass AI agents to fast-track human interaction.
- User habits and trust in AI are expected to evolve over time as agent capabilities improve.