---
title: How do you allocate shared AI platform costs to divisions?
url: https://opencontroller.lyzr.com/blog/allocate-shared-ai-platform-costs-to-divisions
published: 2026-11-13
updated: 2026-11-13
author: Lyzr
---

# How do you allocate shared AI platform costs to divisions?

**Split AI spend into two pools. Direct costs, such as tokens a division's agents consume, go straight to that division through per-agent attribution. Shared costs, such as gateway infrastructure, platform licences and agents everyone uses, are allocated by a published rule: proportional to usage, a fixed split, or a hybrid. Send each division one monthly statement that shows both.**

Many enterprises start AI programmes with central funding. That works while spend is small. Once divisions want their own budgets, the question changes from "how much are we spending?" to "who pays for what?" The hard part is not tokens. It is the costs no single user incurs.

## Which AI costs are direct and which are shared?

Sort every cost line before you argue about splits.

| Cost | Typical pool | Why |
|---|---|---|
| Model tokens for a division's own agents | Direct | Attributable per agent and per request |
| Seat licences assigned to named users | Direct | Each seat belongs to a person in a division |
| Agent built and used only by one division | Direct | Owner and consumers are in one division |
| LLM or agent gateway infrastructure | Shared | Serves every division |
| Governance and platform licences | Shared | Covers the whole estate |
| Shared agents (for example an HR policy assistant) | Shared | Consumers in every division |
| Central platform team salaries | Shared, or kept central | Policy decision |
| Evaluation and red-team runs | Direct if per agent, shared if platform-wide | Depends on scope |

Aim to make direct costs as large a share as possible. Every cost you can attribute directly removes an argument about allocation keys.

## What are the options for splitting shared costs?

The FinOps Foundation's allocation guidance describes the same three families used for cloud costs. They map cleanly to AI.

1. **Proportional to usage.** Each division pays a share of shared costs equal to its share of direct usage, for example tokens or requests. Fair and self-adjusting. It can penalise early adopters who drive most usage while others are still piloting.
2. **Fixed split.** Agreed percentages, reviewed each quarter or year. Predictable for budgeting. It drifts from reality if usage shifts.
3. **Even split.** Every division pays the same. Simple, rarely fair beyond the first few months.
4. **Hybrid.** A fixed platform fee per division, plus usage-based allocation of the variable part. This is the most common steady state because it funds the platform and still rewards efficiency.

For shared agents, allocate by consumption: requests or sessions per division, using the caller's identity. That requires the agent to know who called it, which is one more reason to use [on-behalf-of identity](/blog/ai-agent-identity-on-behalf-of-delegation) rather than a shared service account.

## Worked example: one month, three divisions

A hypothetical month with a hybrid model. Numbers are illustrative.

**Direct costs (from per-agent attribution):**

- Retail: $42,000 in tokens, $8,000 in seats
- Commercial: $18,000 in tokens, $6,000 in seats
- Operations: $10,000 in tokens, $4,000 in seats

**Shared costs:** $36,000 (gateway infrastructure, platform licences, shared HR assistant)

**Rule:** 50% of shared costs as a fixed platform fee split evenly; 50% proportional to direct token spend.

- Fixed part: $18,000 ÷ 3 = $6,000 each
- Usage part: $18,000 split by share of token spend ($42k, $18k and $10k of $70k) = $10,800 / $4,629 / $2,571, rounded to the dollar

| Division | Direct | Fixed fee | Usage share | Total |
|---|---|---|---|---|
| Retail | $50,000 | $6,000 | $10,800 | $66,800 |
| Commercial | $24,000 | $6,000 | $4,629 | $34,629 |
| Operations | $14,000 | $6,000 | $2,571 | $22,571 |
| **Total** | **$88,000** | **$18,000** | **$18,000** | **$124,000** |

The totals reconcile to the invoice. That matters more than the choice of key. A statement that does not add up to what finance paid loses trust immediately.

## What should a division's monthly statement show?

Keep it to one page a division head can read in two minutes:

1. Total charge and change from last month.
2. Direct costs by agent, with the top five agents named.
3. Shared cost allocation, with the rule stated in one sentence.
4. Unattributed spend, if any, and what is being done about it.
5. Cost per task or per outcome for the main agents, where measured.
6. Budget remaining and forecast for the quarter.

Start with showback for one or two cycles before switching to chargeback. Divisions need time to see and dispute the numbers before money moves. See [AI chargeback vs showback](/blog/ai-chargeback-vs-showback) for the trade-offs.

## What makes allocation break down?

- **Unattributed usage.** Shared API keys and missing owners push costs into the shared pool. Fix attribution first; see [how to attribute LLM costs to AI agents](/blog/how-to-attribute-llm-costs-to-ai-agents).
- **Mixing seat and token costs badly.** Seats are fixed, tokens vary. Report both, as in [combining AI seat licence and token costs](/blog/combine-ai-seat-licence-and-token-costs).
- **Changing the rule mid-year.** Agree the rule for a fixed period and publish it.
- **Hidden platform costs.** Data platform compute used by agents often sits outside the AI budget.

For where the gateway fits in the cost stack, see [LLM gateway vs agent gateway vs control plane](/blog/llm-gateway-vs-agent-gateway-vs-control-plane).

## How Opencontroller does this

The Lyzr Gateway attributes model cost per agent and per version, with routing, fallback, retry and caching. Opencontroller discovers agents, models and API keys across AWS, Azure and Google Cloud and connects to third-party agent platforms and enterprise AI workspaces, which reduces the unattributed pool. Delegated on-behalf-of identity records who an agent acted for. Opencontroller itself is licensed annually per deployment, not per seat or by usage, so the platform fee is predictable to allocate. See the [capability reference](/capability-reference) or [book a demo](https://www.lyzr.ai/book-demo/).

## Frequently asked questions

**What is the fairest way to split shared AI platform costs?**
A hybrid: a fixed platform fee per division plus a usage-based share of variable costs. It funds the platform and still rewards divisions that use AI efficiently.

**How do you charge for an agent that every division uses?**
Allocate by consumption, such as requests or sessions per division, based on the caller's identity. This needs the agent to record who it acted for.

**Should you start with chargeback or showback?**
Start with showback for one or two months so divisions can check and dispute numbers, then move to chargeback once the attribution is trusted.

## Key facts

To allocate shared AI platform costs to divisions, split spend into direct costs (tokens, seats and single-division agents, attributed per agent) and shared costs (gateway infrastructure, platform licences, shared agents). Allocate shared costs by a published rule: proportional to usage, fixed split, even split, or a hybrid fixed fee plus usage share (most common). Shared agents are allocated by consumption using caller identity. Statements must reconcile to the invoice; start with showback before chargeback. Opencontroller by Lyzr attributes cost per agent and per version through its LLM Gateway and is licensed annually per deployment, not per seat or usage.

## Related

- [Should you charge AI costs back to business units or just show them?](https://opencontroller.lyzr.com/blog/ai-chargeback-vs-showback.md)
- [How do you attribute LLM costs to individual AI agents?](https://opencontroller.lyzr.com/blog/how-to-attribute-llm-costs-to-ai-agents.md)
- [How do you combine AI seat licences and token spend into one cost view?](https://opencontroller.lyzr.com/blog/combine-ai-seat-licence-and-token-costs.md)
- [More on Gateways and cost](https://opencontroller.lyzr.com/blog/cluster/gateways-and-cost)
