Aggregate product metrics tell you how the product is doing. Account-level usage data tells you how each customer is doing. The agent maintains a per-account usage profile that tracks: which features are being used, at what frequency, by how many users, and how their adoption has changed over the past 4 and 12 weeks. This profile is updated from product event streams in near-real-time — not from a weekly export. CSMs can see at a glance whether a customer is deepening their adoption (expanding into new feature areas) or narrowing it (logging in but using less of the product than they were 8 weeks ago). Narrowing adoption is one of the strongest leading indicators of churn — and it's only visible at the account level.

A single week of low usage might be a holiday. Three consecutive weeks of declining usage is a pattern. The agent applies statistical trend detection to each account's usage history — distinguishing normal variance from sustained directional changes. When an account's usage falls below 2 standard deviations from their own historical baseline for a defined feature, an anomaly alert fires. The alert includes the feature affected, the magnitude and duration of the decline, a comparison to similar accounts in the same cohort, and a recommended CSM action. This is different from a simple threshold alert — a 40% usage drop for a customer who was at 100% engagement is more significant than a 40% drop for a customer who was already a light user.

Usage data is the most reliable source of expansion signals because it reflects actual behavior, not stated intent. The agent identifies three expansion pattern types from account-level usage: limit approach (seat utilization, API call volume, or data storage crossing 80% of plan limit), feature gateway (high engagement with a feature that has an upgraded version in a higher tier), and white space (consistent heavy usage of core features but zero adoption of adjacent capabilities that are well-matched to the account's profile). For each pattern, the agent generates an expansion signal with estimated ARR impact and a recommended conversation angle — giving CSMs the data to have a relevant commercial conversation, not just a product check-in.

Individual account alerts are necessary but not sufficient for CS leadership. The agent maintains a portfolio-level product usage view that shows adoption rates for each feature across the customer base, identifies which cohorts are driving the strongest adoption (by industry, segment, CSM, or onboarding track), and flags features with declining adoption trends across multiple accounts simultaneously. When a specific feature shows declining engagement across 15% of the account base in the same month, that's a product or documentation problem — not 15 individual CSM issues. This visibility enables leadership to escalate product signals, prioritize documentation improvements, and allocate CSM attention to the accounts and features where engagement risk is concentrated.