The 37× Problem: Why Your AI Invoice Doesn't Match Your Seat Count
The CFO called again. The Microsoft 365 Copilot line item is 40% higher than last quarter. Nobody on the IT team can explain it. The licence count hasn't changed. The seat count is flat. But the invoice isn't.
The answer is not in the invoice. It never was.
Seats explain licences. Sessions explain cost.
PromptKing's Session Type Classifier categorises every AI interaction into four types:
- Interactive — user-initiated chat, predictable cost
- Scheduled — cron-triggered workflows, moderate cost
- Background — system-initiated processing, often invisible
- Agentic — autonomous multi-step agents, highest cost variance
The multiplier between interactive and agentic sessions is not 2× or 5×. In production Copilot Studio deployments, we routinely see 37× — a single agentic execution involving Graph grounding, Dataverse writes, and two automated actions costs 37 times what a simple interactive chat costs.
Seat-based FinOps cannot detect this. You can have 200 seats at $30/month and a single HR onboarding agent generating $142/month in Studio credits alone — billed separately, invisible in the seat count.
The four bills hiding inside one AI invoice
Microsoft's AI cost structure in 2026 has four layers:
- M365 Copilot licence — $30/seat/month, flat, predictable
- Copilot Studio credits — 25 credits per autonomous trigger, 10 more per Graph query
- Azure OpenAI consumption — token usage beyond Studio credits, different invoice
- Dataverse storage — agent deployments push past default 15GB
Finance sees two or three line items and cannot connect them. The CFO is not confused because AI is complex. The CFO is confused because the same AI workload generates four invoices, none of which tell the same story.
What the Agentic Cost Panel shows
PromptKing's Agentic Cost Panel breaks cost down by session type, agent, and vendor layer — per execution, not per month. It answers the question seat dashboards cannot:
- Which agents are driving the spike?
- What session type dominates spend?
- Which cost layer (licence, Studio credits, Azure tokens, Dataverse) is growing fastest?
This is the CFO view that closes the board conversation — not "we have 200 Copilot seats" but "our agentic sessions cost $4.20 each vs $0.11 for interactive chat, and HR Onboarding Agent ran 164 times last month."
From visibility to control
Cost visibility without policy enforcement is monitoring. When an agent's 30-day spend hits 3× its 7-day average, PromptKing's Policy Engine triggers SPEND-002, routes to the Decision Log, and — on admin approval — can quarantine the agent via Microsoft's management API.
The cost insight and the governance action live in the same control plane. That is what separates an AI FinOps dashboard from an AI control layer.
See your organization's AI spend data
PromptKing connects to your AI vendors and surfaces exactly this analysis — for your seats, your vendors, your budget.