At SaaStr AI Day, Pylon's founders Marty Kausas and Advith Chelikani presented a new vision for customer support AI, revealing why traditional deflection metrics fail to capture true efficiency and how their agentic support model enhances human work rather than replacing it.
- Deflection rate can be misleading in measuring support team efficiency.
- Pylon emphasizes human-AI collaboration over full automation.
- Precomputed context layers improve AI-augmented ticket handling.
What happened
During SaaStr AI Day, Pylon’s co-founders, Marty Kausas and Advith Chelikani, critiqued the traditional focus on ticket deflection rates in customer support. They highlighted a case where a large company with a 1,000-person support team deployed automated resolution agents that deflected about half of incoming tickets, yet the overall headcount remained constant. This illustrated that deflecting tickets alone does not necessarily reduce workload, as the more complex issues require human intervention.
Pylon shared insights from their launch of an agentic customer support product, which prioritizes AI augmentation of human agents instead of outright automation. Their approach involves utilizing AI to precompute comprehensive customer context—such as past interactions, sentiment, related issues, and backend status—to assist support reps in managing escalated tickets more efficiently.
Why it matters
The widespread industry reliance on deflection rates as a key performance indicator may paint an incomplete or overly optimistic picture of support efficiency. Easy-to-resolve tickets constitute the majority of deflected work, which often consume minimal time, leaving human reps to handle the more context-heavy and relationship-driven requests. This results in unchanged staffing needs despite apparent improvements in ticket deflection.
Pylon’s approach signals a paradigm shift in AI deployment across support teams. By emphasizing the augmentation of human agents rather than replacement, they address the nuanced challenges of B2B customer support, where contextual understanding and tailored responses are critical. Their method also suggests potential cost savings and faster resolution through better AI inference efficiency and enhanced team control.
What to watch next
Operators should critically assess their support metrics by comparing deflection rates alongside key indicators such as headcount and median handle time to understand the true impact of AI integrations. The evolution of AI in support is likely to focus on hybrid models, combining automated front-line ticket handling with deep AI augmentation for human analysts.
Companies developing AI support tools need to consider implementing precomputed context layers to reduce repetitive data processing and improve AI quality and speed. The market may see growing demand for solutions that empower support agents with contextual intelligence, enabling them to resolve complex issues more rapidly while maintaining meaningful customer relationships.