What Is Seat Behavioral Classification?

The central challenge of enterprise AI governance is not understanding what AI costs in aggregate. Most organizations can get that number from a vendor invoice. The challenge is understanding what each seat costs, what each seat produces, and whether the plan tier matches the actual consumption pattern.

This is the problem seat behavioral classification solves.

The framework uses five categories, applied per seat over a rolling 30-day measurement window:

GHOST

Definition: Token consumption below 5% of plan capacity. What it means: This seat is provisioned but effectively unused. The user may have left the organization, switched to a different tool, or simply never adopted the product. The seat is billing at full plan rate and generating zero measurable output. Action: Deprovision or downgrade immediately. Confidence is high.

UNDERUTILISED

Definition: Token consumption between 5% and 20% of plan capacity. What it means: The user is active but consuming a fraction of their provisioned capacity. A seat on Max 5× consuming 8% of plan capacity is doing the work of a Pro seat at 5× the cost. Action: Review within 30 days. Downgrade if pattern persists across two consecutive measurement periods.

NORMAL

Definition: Token consumption between 20% and 80% of plan capacity. What it means: Healthy utilisation. The plan tier is appropriate for the usage pattern. No action required. Action: Monitor quarterly.

POWER USER

Definition: Token consumption above 80% of plan capacity. What it means: This seat is heavily used and approaching plan limits. If the user hits their monthly ceiling, productivity stops. Action: Monitor for limit events. Proactive upgrade conversation before they hit the ceiling mid-project.

AT-RISK

Definition: Consumption trajectory indicates the seat will exhaust plan capacity before the end of the billing period. What it means: This is not just a Power User — this is a Power User who will run out of tokens before month-end based on current burn rate. Action: Immediate upgrade or workload redistribution.

WHY THIS MATTERS MORE THAN TOKEN PRICING

Token prices vary by model and change with vendor releases. Behavioral classification is stable — a Ghost seat is a Ghost seat whether the underlying model costs $0.80 per million tokens or $5.00. The classification tells you what action to take regardless of pricing changes.

This is why seat behavioral classification is the core signal in AI FinOps, not cost per token. The pricing data tells you the size of the problem. The behavioral data tells you where to act.

The five-category seat behavioral classification framework (Ghost, Underutilised, Normal, Power User, At-Risk) described in this article is the subject of a pending US patent application filed by PromptKing Inc. (USPTO ref: 2-01025019).

The metric that matters: what percentage of your seats fall outside the Normal band right now? If you don't know, you don't have the foundation for AI spend governance.