Senior Product Manager at a leading tech company shares insights on architecting truly intelligent Customer Experience systems
The customer support industry stands at an inflection point. Gartner forecasts that agentic systems will autonomously resolve 80% of customer service issues by 2029, cutting operational costs by 30%. While many organizations experiment with basic chatbots for customer support, Udit Joshi has architected and deployed intelligent AI-based customer experience platforms supporting more than 100 million support engagements each year. His career path uniquely positions him to revolutionize products in this new generation of AI powered tools: from co-founding Red Baton, a UX startup featured in national press, to being Technology consultant at Deloitte on multi-million dollar transformations, to leading Enterprise platform strategy now at Google. The conversational voice bot that he helped develop at Google Ads, achieved 98.5% customer satisfaction while automating 90% of issue detection, and his technical approach subsequently adopted across other businesses at Google including the hardware products, payment solutions and the new autonomous vehicle arm. Joshi shares how he transitioned from legacy customer support systems to a unified AI-first architecture.
Udit, your Customer Experience platform supports more than 100 million customer engagements each year, delivers 40% cost savings compared to traditional solutions, and saves tens of millions of dollars annually. These results positioned your architectural approach as the internal blueprint subsequently adopted in support operations for autonomous vehicles and payments. What core technical decisions enabled this transition from the industry’s AI-Assistive approach to your AI-first model?
I faced a critical choice: continue layering more automation features onto fragmented legacy systems, multiplying system complexity, or rebuild around a unified AI-first architecture. The second option was necessary. My architecture rests on three pillars. First, a Unified Omnichannel Console consolidates chat, voice, and email into one application, eliminating productivity losses from tool-switching. Second, a Prescriptive RAG Engine provides contextual guidance based on customer situation and history rather than just retrieving information. Third, intelligent routing using Natural Language Understanding to detect complex intent and connect customers to specialists immediately. This delivers 40% cost savings compared to traditional solutions.
Your conversational voice bot for the company’s advertising platform replaced traditional IVR systems. What specific engineering challenges did you overcome to achieve high accuracy in issue detection, and how did you solve them?
The primary challenge was achieving precision for large scale and complex enterprise products. Traditional Natural Language Understanding wasn’t enough because customers describe problems in varied ways. I needed the system to understand intent within context. I engineered a multi-layered detection system analyzing conversational context, customer account status, and historical patterns simultaneously. When a customer says “my ads are not working,” the system must distinguish between technical failures, budget depletion, policy violations, or strategic misconfigurations. Each requires a different specialist routing. The breakthrough was creating a real-time evaluation engine processing these data streams fast enough for live calls while maintaining accuracy. This automated 90% of diagnosis work and achieved routing precision that eliminated customer frustration inherent in traditional IVR’s rigid menu structures.
Customer support systems are typically built for specific use cases with specialized knowledge.. How did you approach designing an architecture that could work across different business contexts without requiring complete rebuilds?
Every customer-facing operation faces the same challenge: consistent, high-quality service at scale without complex navigation. I built the architecture around universal principles rather than product-specific rules. The Unified Console provides complete context regardless of messaging topic—advertising campaigns, payments, or other services. The Prescriptive RAG Engine adapts to different domains: autonomous vehicles, payment protocols, and advertising policies, while maintaining contextual guidance patterns. The critical design decision was defining clear boundaries between autonomous AI handling and when human expertise becomes necessary. This boundary logic transfers across contexts because it’s based on complexity thresholds and confidence levels rather than domain rules. Different business units can plug in their domain knowledge while core routing, context management, and escalation logic remain unchanged.
You co-founded Red Baton during undergraduate studies, served as Technology Consultant at Deloitte on multi-million dollar enterprise software implementations, and now lead enterprise platform strategy at a major tech company. How does combining entrepreneurial, consulting, and corporate experience influence your approach?
Founding Red Baton taught me to identify unmet needs and build unconventional solutions. We redesigned school notebooks to carry advertisements, requiring rethinking value creation in overlooked spaces. That assumption-questioning influenced my decision to re-architect infrastructure rather than improve incrementally. I carried this builder’s mindset into my current role, where I won an award at a company-wide hackathon. This global competition, judged by senior executives, recognized my team’s innovative solution for complex challenges in the enterprise customer engagement platform. This reinforced how combining technical proficiency with entrepreneurial thinking is essential for solving industry problems. Deloitte provided the opposite skillset: enterprise-scale transformation discipline. Leading multi-million dollar SAP CRM implementations taught me to navigate complex stakeholders and implement change in resistant organizations. Combining these mindsets lets me think boldly while executing with the required discipline. When proposing the AI-first customer experience model, I presented a complete transformation strategy with implementation plans and risk mitigation, not just technology.
Your platforms support core revenue-generating products across the organization. What do you see as the next critical challenge for enterprise customer experience transformation, and how should leaders leverage AI to bridge that gap?
Organizations face a “vision gap” – thinking about incremental improvements when they should be using these new technical capabilities to completely re-architect their operational models. True transformation requires moving beyond “how do we add AI and these new tools to our workflow?” to “how do we rebuild systems around AI as primary operator?” Companies have the technology but lack architectural vision and change management discipline to execute a full-scale transformation. My plans focus on sharing the blueprint I have developed through writing, speaking, and mentorship. I’m producing academic articles about customer experience transformation and AI product management, mentoring younger product managers, and participating in technical communities like Hackathon Raptors fellowship and IEEE, where I am a senior member. The American economy’s competitiveness depends on organizations adopting truly intelligent AI-supervised models at scale.
