---
title: How do you combine AI seat licences and token spend into one cost view?
url: https://opencontroller.lyzr.com/blog/combine-ai-seat-licence-and-token-costs
published: 2026-11-13
updated: 2026-11-13
author: Lyzr
---

# How do you combine AI seat licences and token spend into one cost view?

**Put every AI cost on one person- or agent-level ledger: the fixed monthly cost of each seat, plus the metered model spend that person or agent generated. Join both on a single identity key, then roll up to team and business unit. Without that join, seat-based tools look free and API-based agents look expensive, and neither picture is true.**

Most enterprises now pay for AI in two shapes at once. Coding assistants and chat workspaces are bought per seat. Agents built on cloud platforms and model APIs are billed per token. Finance sees two invoices. Engineering sees two dashboards. Nobody sees what a team actually spends on AI.

## Why do seat costs and token costs end up in separate reports?

They come from different systems with different units:

- **Seat licences** arrive as a fixed price per user per month on a vendor invoice. Usage inside the seat is often "free" at the margin until an allowance runs out.
- **Token spend** arrives as metered charges from model providers or cloud bills, usually tagged by account, project or API key rather than by person.
- **Overage** sits in between. Some seat plans meter usage above an allowance. For example, Anthropic's usage analytics for seat-based Enterprise plans note that the spend export covers only spend above the seat allotment.

So a developer with a coding-assistant seat and an agent running on a cloud model can appear in three places, under three identifiers.

## What is the formula for total AI cost per team?

Use one formula at the lowest level you can identify (a person or an agent), then sum upwards:

**Total AI cost = Σ seat cost (prorated) + Σ metered model spend + Σ overage + shared platform cost (allocated)**

| Component | Unit | Where it comes from | Identity key |
|---|---|---|---|
| Seat cost | Price per user per month, prorated by days assigned | Vendor invoice, seat-assignment API | User email or IdP ID |
| Metered model spend | Tokens × price | Gateway logs, provider billing, cloud bills | API key, agent ID, project tag |
| Overage on seat plans | Usage above allowance | Vendor spend export | User email |
| Shared platform cost | Gateway compute, vector stores, observability | Cloud bill | Allocated by usage share |

Prorate seats by days assigned. A seat assigned on the 20th of a 30-day month should carry about a third of the monthly price, or your per-team numbers will jump every time someone joins.

## What data do the main vendors expose for this?

Each vendor gives you a different slice. Check the current docs before you build:

1. **GitHub Copilot** exposes seat assignments per organisation (with a `last_activity_at` field) and user-level usage metrics reports per enterprise and per organisation through the REST API. GitHub has also added AI credits consumed per user to the usage metrics API.
2. **ChatGPT Enterprise** workspace analytics provides a Users report as CSV with email, department, groups, message counts and last-active dates for ranges up to 12 months.
3. **Claude Team and Enterprise** usage analytics provides a spend CSV where each row is one person's usage of one model, with email, request count, tokens and spend, plus an Analytics API that the Primary Owner can enable.
4. **Model APIs and cloud platforms** bill by account, project or key. Here you need a gateway or tagging to get to the agent or person.

## What does a worked example look like?

Here is an illustrative month for one platform team. The prices are placeholders, not vendor list prices.

| Line | Calculation | Cost |
|---|---|---|
| 40 coding-assistant seats | 40 × $39 | $1,560 |
| 12 chat workspace seats | 12 × $60 | $720 |
| Chat overage above allowance | from vendor spend export | $210 |
| Two production agents on a cloud model | gateway token logs | $3,400 |
| Share of shared gateway and tracing | 8% of $5,000 by request share | $400 |
| **Total** | | **$6,290** |

Two things usually surprise teams. First, seats are often a minority of total spend once agents run in production. Second, about one seat in five may show no recent activity, which is pure waste. Our guide to [finding unused AI seat licences](/blog/find-unused-ai-seat-licences) covers how to spot those.

## How do you join everything on one identity?

The join is the hard part. Seat systems know users by email or a vendor username. Model spend is keyed by API key or project. Do this in order:

1. Make your identity provider the source of truth for person, team and cost centre.
2. Map vendor usernames to IdP identities. Some, like GitHub, may not expose an email at all. See [how to attribute GitHub Copilot usage to employees](/blog/attribute-github-copilot-usage-to-employees).
3. Give every agent an owner and a cost centre in a registry, then attribute its model spend through the gateway. Our post on [attributing LLM costs to AI agents](/blog/how-to-attribute-llm-costs-to-ai-agents) explains the tagging.
4. Report unmatched spend as its own line. Never spread it silently.

## How Opencontroller does this

Opencontroller's LLM Gateway attributes model cost per agent and per agent version, and its governed registry links each agent to an owner through OIDC SSO. Connectors to Claude Enterprise and OpenAI Enterprise workspaces bring workspace usage into the same estate view, so seat-based and metered AI appear side by side. See the [capability reference](/capability-reference) or the [gateway comparison](/blog/llm-gateway-vs-agent-gateway-vs-control-plane) for how the pieces fit. To see it on your own data, [book a demo](https://www.lyzr.ai/book-demo/).

## Frequently asked questions

**Should seat costs be allocated by usage or by assignment?**
By assignment. A seat costs the same whether it is used or not. Allocate it to the person it is assigned to, then use usage data to decide whether to keep it.

**What if a vendor only reports usage above the seat allowance?**
Then add the fixed seat price from the invoice and the overage from the export. Do not treat the export as total cost.

**How often should the combined view refresh?**
Daily is enough for most teams. Many vendor exports already lag by a day, and finance reviews monthly.

## Key facts

To combine AI seat licences and token spend: total AI cost = prorated seat cost + metered model spend + overage above seat allowance + allocated shared platform cost, joined on one identity key from the identity provider. GitHub Copilot exposes seat assignments with last_activity_at and user-level usage metrics reports; ChatGPT Enterprise workspace analytics exports a Users CSV with last-active dates for ranges up to 12 months; Claude Enterprise exports a per-user, per-model spend CSV (on seat-based plans it covers only spend above the seat allotment). Opencontroller by Lyzr attributes model cost per agent and per version through its LLM Gateway and connects Claude Enterprise and OpenAI Enterprise workspaces. Opencontroller pricing starts from $250K per deployment per year.

## Related

- [How do you attribute LLM costs to individual AI agents?](https://opencontroller.lyzr.com/blog/how-to-attribute-llm-costs-to-ai-agents.md)
- [What AI spend reports does a CFO actually need?](https://opencontroller.lyzr.com/blog/ai-spend-reporting-for-finance-leaders.md)
- [Where do AI agent costs usually leak?](https://opencontroller.lyzr.com/blog/where-ai-agent-costs-leak.md)
- [More on Gateways and cost](https://opencontroller.lyzr.com/blog/cluster/gateways-and-cost)
