Sajjan Kanukolanu argues that attaching a chatbot to an unchanged sales stack misses important context about a buyer's research, history and intent. His architecture separates signal collection, buyer intelligence and actions rather than treating the conversation window as the whole system.
He demonstrates agents that consolidate visitor signals, enrich records, apply ideal-customer criteria, compare contacts with a CRM and route follow-up actions. A context graph connects people, accounts, deals and prior interactions so messages and sales alerts can reflect that history. The demonstration is a vendor case study, not an independently verified conversion result.
The talk addresses changing customer criteria, excessive alerts, imperfect visitor identification and human approval bottlenecks. Kanukolanu recommends separating customer fit from intent, keeping policies auditable and returning outcomes to the knowledge base. He does not claim every visitor can be identified reliably.
Watch on YouTube
