Grit

The Truth Behind Automation Claims in Customer Support | Cresta CEO Ping Wu

February 9, 2026

Key Takeaways Copied to clipboard!

  • Real-world automation success in customer support depends on three constraints: conversation complexity, IT infrastructure debt, and customer demographics, not just model advancement. 
  • Cresta's philosophy views automation and human support as a single, unified system, using AI to augment experts in high-emotion/high-value conversations while automating low-value interactions. 
  • The current AI paradigm shift is perceived as more transformative than previous ones, like the dot-com era, but enterprise adoption is slower due to the time required to clean data and modernize infrastructure. 

Segments

AI Progress Unimaginable Since 2019
Copied to clipboard!
(00:00:00)
  • Key Takeaway: The technological leap in AI capabilities over the last six years (since 2019) is considered unimaginable, even by the creators of foundational models like the Transformer.
  • Summary: Bringing 2001 technology back to 1995 would not surprise people, but bringing today’s capabilities back to 2019 is unimaginable. This rapid advancement includes people who authored the Transformer paper. Ping Wu views automation in the call center as one integrated system, not a separate entity.
Host Introduction and Guest Background
Copied to clipboard!
(00:00:56)
  • Key Takeaway: Ping Wu, CEO of Cresta, previously spent nine years at Google building conversational AI before applying generative AI to the contact center.
  • Summary: Joubin Mirzadegan hosts ‘Grit,’ exploring challenges in building history-making companies. Ping Wu brings generative AI to the contact center to enable real-time human-agent collaboration. The episode will cover why customer service is hard to automate and the future of the hybrid workforce.
Talent Motivation: Tech vs. Customer Problems
Copied to clipboard!
(00:01:49)
  • Key Takeaway: Talent at Cresta is self-selected to be excited by solving hard technical problems that directly drive measurable customer value and ROI, avoiding the trap of only solving ‘cool, sexy’ technical puzzles.
  • Summary: The discussion addresses whether engineering talent prioritizes technical challenges over solving hard customer problems. Ping Wu notes that contact center metrics like efficiency, attrition, and customer satisfaction make ROI demonstration very clear and quantifiable. This environment attracts people who want to apply technology to real-world customer value.
Board Member Insights and Roles
Copied to clipboard!
(00:04:15)
  • Key Takeaway: Board members like Doug Leoni provide insights on company building and SaaS success patterns, while technology experts like Sebastian focus on long-term technology trends.
  • Summary: Carl Icahn (formerly on the board) provided deep operational experience, focusing on go-to-market strategy and scaling through channel partners. Doug Leoni focuses on company-building failure and success patterns due to his extensive SaaS experience. Sebastian offers advice on technological trends, given his background in AI and Waymo.
Cresta’s Unified Contact Center Platform
Copied to clipboard!
(00:06:23)
  • Key Takeaway: Cresta serves major clients like United Airlines, deploying AI assistants to 9,000 agents across chat and voice channels to automate workflows, take notes, and provide insights.
  • Summary: United Airlines is a deep, multi-year partner where Cresta’s AI helps agents with note-taking and workflow automation. United also uses Cresta’s insights-to-action team to fix root causes of calls, aiming to eliminate calls entirely rather than just automating them. The core belief is building one AI platform that provides visibility across the entire contact center data and workflows.
Three Buckets of Contact Center Interactions
Copied to clipboard!
(00:08:49)
  • Key Takeaway: Contact center interactions should be categorized into three buckets: those that should be eliminated (due to broken processes), those that should be automated (low-emotion, fast resolution), and those requiring human augmentation (high-emotion/high-value).
  • Summary: The first bucket involves calls caused by failed processes or confusing bills; AI’s role here is observability to fix the root cause, as seen with United Airlines. The second bucket involves simple tasks like password resets where both parties prefer fast, automated resolution. The third bucket involves high-emotion situations (e.g., insurance claims, lost luggage) where AI should augment the human expert to allow them to be more emotionally available.
Constraints on Automation Potential
Copied to clipboard!
(00:12:36)
  • Key Takeaway: The achievable level of automation is determined by conversation complexity, the modernity and API availability of the IT stack, and customer demographics.
  • Summary: Simple e-commerce interactions built on modern APIs can likely achieve 100% automation. Conversely, complex businesses like global airlines face complicated interactions and legacy systems (often homegrown without APIs) that are optimized for human GUIs. Older customer demographics often prefer human interaction, limiting automation potential even if the technology allows it.
Current LLM Shortcomings for AI Agents
Copied to clipboard!
(00:15:36)
  • Key Takeaway: Current LLMs struggle with complex instruction following, multimodality, and taking action on screens without explicit APIs, necessitating significant context engineering.
  • Summary: A major hurdle is the need to manually engineer context by bringing the right information into the model, which is difficult when knowledge bases are scattered and uncleaned. Furthermore, tribal knowledge not documented in the knowledge base remains inaccessible to AI. Voice interactions introduce more error propagation, and human distrust remains a factor.
Value Beyond Conversation Automation
Copied to clipboard!
(00:18:20)
  • Key Takeaway: Significant value exists in automating non-conversational workflows like quality assurance (QA), after-call work, data entry, and summarizing interactions for human agents.
  • Summary: AI can automate QA processes, which traditionally cover only 1-2% of calls manually, providing real-time compliance checks and insights at scale. For complex calls, AI can automate the first 20% (authentication) and handle all after-call work, including data entry and summaries, freeing up human agents.
Enterprise AI Adoption Pace and ROI
Copied to clipboard!
(00:21:22)
  • Key Takeaway: The initial widespread AI piloting in large enterprises is expected to converge onto use cases with concrete, undeniable ROIs, such as coding and CX transformation.
  • Summary: Concrete value drivers include automating quality management at scale, increasing manager span of control, and generating voice-of-customer insights that can replace survey spending. Ping Wu believes that while bubbles exist, companies that deliver undeniable value will emerge as long-term winners.
Capital Strategy and Staying Lean
Copied to clipboard!
(00:24:09)
  • Key Takeaway: Cresta prioritizes staying lean, viewing excessive valuation as a risk that mortgages the future by setting growth expectations that are difficult to meet, especially concerning inflated salary costs.
  • Summary: Cresta recently raised a Series D but remains cautious about raising capital simply because it is available. The framework involves balancing the cost of capital (dilution) against the equity value that can be created with existing funds. High private valuations must be carefully weighed against public market comparisons and the risk of over-committing to future growth.
CEO Transition and Platform Strategy
Copied to clipboard!
(00:31:19)
  • Key Takeaway: Ping Wu’s transition to CEO was a natural continuum of his mission, solidified by serving as interim CEO, and he drove the strategy to evolve Cresta into a multi-product platform early on.
  • Summary: Wu previously co-founded Google Contact Center AI, making the mission at Cresta a consistent journey. He advocated for expanding into a multi-product platform to create synergy, where data from human assistance feeds back to improve AI agent automation, forming a powerful feedback loop.
Advice for Engineering Leaders Becoming CEOs
Copied to clipboard!
(00:36:46)
  • Key Takeaway: An engineering leader transitioning to CEO must possess a genuine interest in the business side, including sales, convincing partners, and managing diverse human interactions.
  • Summary: The surprise for new CEOs often lies in the necessity of engaging with the business aspects rather than just solving enduring technical problems. While many business skills can be learned from mentors and advisors, the fundamental interest in the human and commercial side of the business is crucial for success in the CEO role.
Future of Customer Experience Transformation
Copied to clipboard!
(00:38:14)
  • Key Takeaway: The customer service domain is poised for massive transformation over the next five to ten years, evidenced by the common search query: ‘how do I get to a human agent’.
  • Summary: The current technological advancements are so far beyond what was imaginable just six years ago that the transformation will likely be bigger than previous paradigm shifts. Enterprise saturation of current model capabilities may take five to ten years as infrastructure is modernized and data is cleaned. Cresta is actively hiring across functions to build out this future.