---
title: Should you charge AI costs back to business units or just show them?
url: https://opencontroller.lyzr.com/blog/ai-chargeback-vs-showback
published: 2026-11-13
updated: 2026-11-13
author: Lyzr
---

# Should you charge AI costs back to business units or just show them?

**Start with showback: report each business unit's AI cost without billing it, until at least about 90% of spend is attributed and owners agree the numbers are fair. Move to chargeback only then. When business-unit tags are missing, allocate through the agent or user owner's cost centre from your identity provider, not through guesswork.**

AI spend grows fast and lands in central budgets: a cloud account here, a seat invoice there, a model API bill on a corporate card. Finance wants division-level budgets. Divisions want to know why they are being charged. The choice between showback and chargeback decides how that conversation goes.

## What is the difference between showback and chargeback?

| | Showback | Chargeback |
|---|---|---|
| What happens | Costs are reported to the consuming team | Costs are moved to the consuming team's budget |
| Money moves | No | Yes, through internal billing or journal entries |
| Accuracy needed | Good enough to start a conversation | Good enough to survive a dispute |
| Behaviour it drives | Awareness and voluntary optimisation | Budget ownership and hard trade-offs |
| Risk | Teams ignore it | Teams contest it, or route around the platform |
| Best for | Early adoption, shared platforms, pilots | Mature, stable workloads with clear owners |

The FinOps Framework treats allocation, showback and chargeback as related practices: you cannot charge back what you cannot first allocate.

## Why is AI harder to allocate than cloud infrastructure?

Cloud resources usually carry account or tag metadata. AI spend often does not:

- **Shared API keys.** One key used by several agents and teams.
- **Shared gateways and platforms.** A central model gateway or agent platform serves everyone from one account.
- **Seat licences** billed centrally, with usage keyed by vendor username.
- **Embedded AI** inside SaaS contracts with no usage breakdown at all.
- **Missing or stale business-unit mappings.** Reorganisations break cost-centre tags long before anyone updates them.

## Which allocation keys work when business-unit tags are missing?

Use them in this order. Record which key allocated each dollar.

1. **Direct tag.** A cost allocation tag or project label on the resource, if it exists and is trusted.
2. **Agent owner.** Every agent in the registry has an owner. The owner's cost centre in the identity provider becomes the agent's cost centre.
3. **User identity.** For seats and per-user usage, the user's cost centre from the identity provider.
4. **Gateway attribution.** For shared model access, the gateway records which agent or key made each call. Allocate tokens by agent, then by agent owner.
5. **Proportional split.** For truly shared costs (gateway compute, observability), split by request or token share.
6. **Unallocated.** Report it as a named line. If unallocated spend is over 10% of the total, you are not ready for chargeback.

## How do you move from showback to chargeback?

A staged plan avoids the arguments that sink most chargeback programmes:

1. **Month 1–2: inventory and owners.** Discover agents, keys and seats. Assign an owner to each.
2. **Month 2–3: showback.** Publish monthly cost by business unit with the allocation key shown. Invite disputes.
3. **Month 3–4: fix the gaps.** Retire shared keys, tag resources, correct cost-centre mappings.
4. **Month 4+: partial chargeback.** Charge back direct and owner-attributed costs. Keep proportional and shared costs in showback.
5. **Ongoing: budgets and alerts.** Give each unit a budget and alert at thresholds; see [setting budgets and alerts for AI agent spend](/blog/set-budgets-and-alerts-for-ai-agent-spend).

Chargeback for seat licences is usually the easiest place to start, because each seat already has an assignee. Our guide to [combining seat licences and token spend](/blog/combine-ai-seat-licence-and-token-costs) shows the full formula.

## How Opencontroller does this

Opencontroller's LLM Gateway attributes model cost per agent and per agent version, and the governed registry links each agent to an owner through OIDC SSO. That gives finance an owner-based allocation key for shared model access without migrating agents off their platforms. For the wider cost picture, see [how to attribute LLM costs to AI agents](/blog/how-to-attribute-llm-costs-to-ai-agents) and the [cost pillar](/blog/llm-gateway-vs-agent-gateway-vs-control-plane). To walk through your own allocation model, [book a demo](https://www.lyzr.ai/book-demo/).

## Frequently asked questions

**Is showback enough on its own?**
For many organisations, yes, for the first year. Showback with named owners drives most of the savings. Chargeback adds budget accountability once numbers are trusted.

**What share of AI spend should be attributed before chargeback?**
A practical bar is about 90%, with the allocation method shown for each line and a dispute process agreed with business units.

**How should shared gateway costs be charged?**
Split them by each team's share of requests or tokens through the gateway, and keep that line in showback until teams accept the method.

## Key facts

Showback reports AI costs to business units without moving money; chargeback moves costs into their budgets. Recommended: start with showback and move to chargeback once roughly 90% of AI spend is attributed and allocation methods are agreed. Allocation keys when business-unit tags are missing, in order: direct tag, agent owner's cost centre from the identity provider, user's cost centre for seats, gateway per-agent attribution, proportional split for shared costs, and an explicit unallocated line (over 10% unallocated means not ready for chargeback). Opencontroller by Lyzr attributes LLM cost per agent and per version through its LLM Gateway and links agents to owners via OIDC SSO. Pricing starts from $250K per deployment per year, licensed per deployment, not per seat or usage.

## 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)
- [How do you set budgets and alerts for AI agent spend?](https://opencontroller.lyzr.com/blog/set-budgets-and-alerts-for-ai-agent-spend.md)
- [More on Gateways and cost](https://opencontroller.lyzr.com/blog/cluster/gateways-and-cost)
