Agentic systems can only learn from what they can see. If the event stream contains only a narrow slice of user behavior, then the system’s view of the product experience is fragmented and incomplete. This paper explains why including the full range of user events - core and peripheral alike - is critical for effective learning. By treating every action as a potential signal, even those that seem only loosely connected to business goals, you enable agents to trace the paths users actually take, not just the destinations they occasionally reach.
The Problem With Only Tracking “Core” Events
When product teams think about what events to share, they often default to only the ones that represent the business’s top priorities: a purchase, a subscription renewal, a completed checkout. These core events are critical to the business, but if they are the only events provided, they leave agents effectively blind until the very end of the process.
Imagine trying to play basketball blindfolded, with the blindfold lifted only in the split second when you make a basket. You’d know you scored, but you’d have no sense of how you got there, what moves worked, or why you lost the ball on the last possession. That’s what it’s like for an agent learning only from core events. Success and failure exist in sharp, isolated moments, but the game itself - the path that led there - remains invisible.
Agents learn by recognizing patterns and making adjustments. If a user does one thing and the outcome is only weakly correlated with the final goal, then does something else that correlates more strongly, the agent has found a directional signal. Neither action may be the business’s ultimate objective, but together they illuminate the path toward it.
Peripheral events document what happens between baskets. They tell the story of the process, not just the outcome. Without them, an agent cannot reconstruct how users progress (or stall) on their journey, and it cannot learn how to guide them more effectively.
Top-Only Is as Problematic as Bottom-Only
Focusing exclusively on the top of the funnel creates the same blind spot as focusing only on the bottom. In both cases, you’re cutting out a crucial part of the user journey and rendering it invisible to the agent. If you provide only pre-checkout or onboarding events, the system learns nothing about what happens once users actually start engaging with the product. If you provide only conversion or purchase events, the system is blind to the moments when users explore, hesitate, or drift away. Either way, you strip out the continuity that makes learning possible.
Agents Need to See Movement, Not Just Milestones
Think of the user journey as a path leading to a gate. The gate itself—the purchase, the signup, the completed action—is important. But if you only record events that occur at the gate, you deprive agents of all information about the path that led there. Did the user walk halfway down and turn around? Did they circle the gate three times before entering? Did they stop at every landmark or rush straight through? Without the full trail of events, the agent has no way to learn what progress looks like.
When agents have access to a wide variety of events, including those that may feel peripheral to business goals, they can:
Detect Progress Early: Recognize when a user is on the right track, even before they reach the final step.
Learn Effective Sequences: See which chains of actions lead most reliably to desired outcomes.
Differentiate User States: Distinguish between a disengaged user and one who is actively exploring but not yet converting.
Adapt Interventions: Adjust nudges and recommendations in real time, based on the specific stage a user is in.
This makes the agent more than a scoreboard watcher - it can learn to understand the flow of the game. On the other hand, when only core events are instrumented:
Learning is sparse: The agent only sees a small fraction of user behavior.
Signal is delayed: Feedback arrives too late to help adjust the course.
Optimization is brittle: Strategies are tuned to rare, extreme moments rather than the continuous dynamics of engagement.
The result is agents that struggle to help users build momentum, because they lack the evidence needed to recognize momentum in the first place.
The Problem With Only Tracking “Core” Events
When product teams think about what events to share, they often default to only the ones that represent the business’s top priorities: a purchase, a subscription renewal, a completed checkout. These core events are critical to the business, but if they are the only events provided, they leave agents effectively blind until the very end of the process.
Imagine trying to play basketball blindfolded, with the blindfold lifted only in the split second when you make a basket. You’d know you scored, but you’d have no sense of how you got there, what moves worked, or why you lost the ball on the last possession. That’s what it’s like for an agent learning only from core events. Success and failure exist in sharp, isolated moments, but the game itself - the path that led there - remains invisible.
Agents learn by recognizing patterns and making adjustments. If a user does one thing and the outcome is only weakly correlated with the final goal, then does something else that correlates more strongly, the agent has found a directional signal. Neither action may be the business’s ultimate objective, but together they illuminate the path toward it.
Peripheral events document what happens between baskets. They tell the story of the process, not just the outcome. Without them, an agent cannot reconstruct how users progress (or stall) on their journey, and it cannot learn how to guide them more effectively.
