Most consumer-facing AI agents open every conversation the same way: they ask. How can I help you, who are you, what did you try, what went wrong. As you can imagine (or have experienced first-hand) this is extremely frustrating for the human on the other side of the chat.
Conviva analyzed AI agent interactions across major e-commerce and travel booking sites and found that consumer-facing agents spend an average of two minutes and 32 seconds establishing basic context before they can address a customer’s actual request. In nearly two thirds of sessions, customers were asked to re-explain information the agent should have already known, including pages visited, tasks attempted, and errors encountered. 8.5% of shoppers abandoned mid-session, either during context-setting or immediately after.
The agents themselves aren’t the issue. What’s missing is the layer underneath them.
Why Do AI Agents Seem Stupid?
Gartner published a report in March 2026 identifying what’s absent from most agentic deployments: a dedicated context layer. According to Gartner®, “The absence of a dedicated context layer, an emerging component in agentic architectures, leads to unreliable agent behavior and higher costs, especially when LLMs process poorly prioritized or noisy data without effective context filtering.”
The Gartner research identifies three components a robust context layer must include: semantic understanding (organizing information based on meaning, not just keywords), operational state (right-time data so agents act on current information), and provenance (mechanisms to track data sources through to decisions and outcomes).
When that layer is missing — which it is in most consumer-facing deployments today — agents default to asking. They greet every customer as a stranger.
Conviva’s research quantifies what that costs:
- 3.4 turns wasted per session on average, just establishing context
- 30% of sessions ended in failure because the agent misread the user’s intent from the start
- 5% of users abandoned mid-explanation, before they ever stated their actual request
- 3.5% completed context-setting, received a response, and still left — driven out by the friction of re-explaining
The root cause, in every case, is that web analytics, app telemetry, and the chat agent are disconnected systems. The agent simply cannot see what happened before it opened.
Why the Context Layer Is Hard to Build for Consumer Agents
Gartner notes that “this context layer cannot be bought off the shelf; it must be engineered to fit your organization’s unique needs.” We feel that observation is especially true for consumer-facing agents, where the context challenge is fundamentally different from enterprise agentic AI.
Enterprise agents operate with known users, structured roles, and bounded task sets. Consumer-facing agents operate with anonymous users, infinite session variability, millions of simultaneous interactions, and behavioral signals that play out across channels — app, web, voice, chat — in real time.
For a consumer context layer to work at decision time, it needs three capabilities that most analytics platforms weren’t built to provide:
- Full-census behavioral capture. Sampling doesn’t work when you need to know this specific customer’s session right now. Conviva captures behavior at full census — every user, every action, no aggregation or sampling.
- Stateful, sequential pattern intelligence. What matters isn’t a page view or a click — it’s the sequence of how those things happened, over what time period. A customer who visited the baggage policy page, opened the mobile app, hit an error, and then opened a chat within a two-minute timespan is communicating something precise. Conviva’s stateful pattern analytics reads those sequences as behavioral fingerprints.
- Real-time injection at decision time. Context is most valuable if it reaches the agent before the first message is sent. Conviva’s Consumer Context Graph injects pre-chat behavioral context — pages visited, tasks attempted, self-service failures, frustration signals like rage clicks — directly into the agent’s system prompt before the conversation opens.
The result: the agent already knows why the customer is there before they say a word. The re-establishment loop is eliminated at the source.
What Changes When AI Agents Start With Context
Based on Conviva’s analysis, a ~70% reduction in the context gap is achievable for sessions where a true behavioral signal is available. That translates to:
- Context re-establishment rate dropping from 65% to ~20%
- Aggregate time wasted reduced by 70%
- Wrong-intent session failure rate falling from 30% to ~9%
As Gartner® concludes: “The context layer is foundational for AI success. D&A leaders must prioritize the development of a robust context layer to empower AI agents with the knowledge required for consistently reliable, cost-efficient and contextually relevant decision making.”
For consumer brands, that work starts with knowing what your customers did before they asked for help.
Want complimentary access to the Gartner® research? Find it here.
Gartner®, “The 3 Core Components of the Context Layer for AI Agents,” Andrés García-Rodeja, Michael Gonzales et al., 10 March 2026. GARTNER is a registered trademark and service mark of Gartner, Inc. and/or its affiliates in the U.S. and internationally and is used herein with permission. All rights reserved. Gartner does not endorse any vendor, product or service depicted in its research publications, and does not advise technology users to select only those vendors with the highest ratings or other designation. Gartner research publications consist of the opinions of Gartner’s research organization and should not be construed as statements of fact.