New research reveals 8.5% of agent interactions end in abandonment because shoppers were forced to spend minutes providing context the agent should have already had.
FOSTER CITY, Calif., [DATE] — Conviva, the intelligence layer across AI agents, websites, and apps, today released new research finding that consumer-facing AI agents waste an average of two minutes and 32 seconds gathering basic context before they can begin to address a customer’s request. The study, based on an analysis of major e-commerce and travel booking sites, finds that context failure — not answer quality — is the leading driver of agent abandonment.
In nearly two-thirds of sessions (65%), customers were asked to re-explain information the platform should have already captured: pages visited, orders placed, tasks previously attempted, and errors encountered. This re-explanation loop required an average of 3.4 back-and-forth exchanges per session and resulted in measurable abandonment at every stage — before, during, and after context was established.
“Brands are shipping AI agents and calling it a customer experience upgrade, but if the agent opens with ‘How can I help you today?’ after a customer just spent 10 minutes failing to complete a task on their own, that’s not an upgrade,” said Keith Zubchevich, President and CEO of Conviva. “Agents can’t see what happened outside the conversation they’re currently in, and that blind spot is costing brands.”
Key Findings
- 2:32 — Average time consumed in context-gathering exchanges before the user’s actual request could be addressed.
- 3.4 — Average number of back-and-forth exchanges required for the agent to gather context before addressing the customer’s request.
- 65% — Share of sessions where the user had to explain what the platform should have already known: pages visited, tasks attempted, order and booking details, errors encountered.
- 8.5% — Shoppers who abandoned the session while establishing context or immediately after — before receiving any substantive help.
There Are Two Kinds of Context. Most Agents Only Have One.
Conviva’s research distinguishes between two fundamentally different context types: response context and behavioral context.
Response context is what’s happening in the current agent session. It includes the system prompt, the conversation history, and retrieved documents. Most AI agents can acquire response context in milliseconds.
Behavioral context is what happened before the conversation started. It includes the pages a user visited, the tasks they attempted, the self-service flows they tried and failed, the frustration signals they generated across the app or website over multiple sessions. Most AI agents never acquire behavioral context at all.
Conviva found agents spend two minutes and 32 seconds asking a customer to reconstruct, in conversation, the behavioral context the agent was never built to capture. The distinction matters because no amount of model improvement closes an architectural gap. An agent that cannot see what happened outside the current conversation will always have to ask.
About the Research
This analysis is based on customer-facing AI agent sessions across e-commerce and travel booking sites, captured and scored using Conviva’s full-census, client-side telemetry. The full report is available at conviva.ai.
About Conviva
Conviva is the digital intelligence platform for apps, websites, and AI agents — connecting real consumer interactions to real business outcomes. Unlike traditional analytics tools that track events and funnels, Conviva maps the full sequence of consumer behavior over time, surfacing the patterns that predict revenue, retention, and risk. With full-census, client-side telemetry and Stateful Pattern Analytics, Conviva sees what session-based and funnel-based tools cannot. Learn more at conviva.ai and follow the company on LinkedIn.
Media Contact
Liesse Jayalath
[email protected]