Excel template · Measure
AI Showback and Chargeback Model
An outline model for reporting or charging AI cost to teams: cost pools, allocation drivers, the monthly split, an unallocated line and a readiness test for moving to chargeback.
Excel (.xlsx) · version 0.1 (2026-10-27) · a proposal from this site, not a standard · CC BY 4.0
- What it is
- A workbook that builds monthly AI cost pools, splits each by a stated method and driver, keeps headroom and unallocated cost as their own lines, and tests whether you are ready to move from showback to chargeback.
- Who it is for
- A FinOps, TBM or ITFM lead setting up cost visibility for AI, with the platform owner who runs shared capacity.
- When to use it
- When AI spend has grown past a single budget line and teams are asking what their share is, or before turning on chargeback for a shared AI platform.
- Not for
- A billing system or a source of truth for invoices. It is an outline to agree the method before it is automated.
AI Showback and Chargeback Model
Excel (.xlsx) · version 0.1 · CC BY 4.0
The full structure
Every sheet in the file, in order. The download adds space to write, the formulas, input shading and drop-down lists.
Instructions
How to use the workbook: cream cells are inputs, grey cells are formulas, blue rows are illustrative.
- Start with showback. Move a pool to chargeback only when the readiness test on the Readiness sheet passes for it.
- Build each pool from its whole cost. A rate based on GPU or token cost alone undercharges, and the shortfall reappears as an unexplained central cost.
- Keep headroom held for a service level and unallocated cost as separate lines. Do not spread them across teams.
Cost pools
Everything it costs to provide each kind of AI capability for a month, by spend type.
Illustrative rows
| Pool | Spend type | Includes | Source record | Monthly cost |
|---|---|---|---|---|
| Model APIs | Direct consumption | Token and request charges from model providers | Provider invoices or billing export | 18,000 |
| Copilot and assistant seats | Licensed seats | Per-user AI licences | Licence register | 24,000 |
| Embedded AI tiers | Embedded tiers | AI add-ons inside SaaS contracts | Contract schedule | 9,000 |
| Shared AI platform | Infrastructure | Reserved GPUs or depreciation, serving stack, platform team, shared overheads | Cloud bill, fixed asset register, team cost | 120,000 |
Allocation drivers
For each pool, the method and the measured driver used to split it.
Illustrative rows
| Pool | Method | Driver unit | Data source | Driver measured reliably? |
|---|---|---|---|---|
| Model APIs | Metered use | Tokens by project key | Gateway logs | Yes |
| Copilot and assistant seats | Metered use | Assigned seats | Licence register | Yes |
| Embedded AI tiers | Fixed share | Agreed percentage | Annual agreement | No |
| Shared AI platform | Hybrid | Reserved GPUs plus metered GPU-hours | Scheduler and labels | Partly |
Monthly allocation
Each team's driver quantity per pool, and the cost that follows from it. Headroom and unallocated cost are calculated, not spread.
Illustrative rows
| Team or cost centre | Model APIs share | Seats share | Embedded share | Platform share | Allocated cost |
|---|---|---|---|---|---|
| Customer operations | 45% | 30% | 40% | 35% | [ ] |
| Engineering | 35% | 45% | 20% | 25% | [ ] |
| Finance | 20% | 25% | 40% | 10% | [ ] |
| Headroom held for a service level | [ ] | [ ] | [ ] | 12.5% | [ ] |
| Unallocated | [ ] | [ ] | [ ] | [ ] | [ ] |
Internal rates
For shared capacity: reserved, metered and batch tiers, each tested against the external price for equivalent quality.
| Tier | What it buys | Internal rate | External equivalent | Internal above external? |
|---|---|---|---|---|
| Reserved | Guaranteed capacity, charged whether used or not | [ ] | [ ] | [ ] |
| Metered | Shared capacity at the fully loaded rate, no guarantee at peak | [ ] | [ ] | [ ] |
| Batch | Spare capacity at a discount, may be paused | [ ] | [ ] | [ ] |
Readiness for chargeback
The four conditions to meet before moving a pool from showback to chargeback. The sheet recommends showback until all four hold.
| Condition | Met? (Yes or No, one column per pool) |
|---|---|
| The driver is measured reliably, and teams can reconcile their share to their own usage data. | [ ] |
| The pool reconciles to the source invoices or the fixed asset register every month. | [ ] |
| Teams have the controls to change their use: reservation sizes, model choice, batch scheduling. | [ ] |
| The internal rate has been compared with the external alternative, and the gap is understood. | [ ] |
Illustrative
Illustrative month
Hypothetical. Every figure is invented and describes no real organisation. They are the blue rows in the workbook. Money is in pounds a month.
| Line | Shared AI platform |
|---|---|
| Monthly pool | £120,000 |
| Charged to teams | £84,000 |
| Headroom held for one service's peak | £15,000 |
| Unallocated | £21,000 (17.5% of the pool) |
Terms used in this template
Read alongside
Who Pays for Idle GPUs?
The four allocation methods, internal tiers and the move to chargeback.
Showback
Reporting cost without moving money.
Chargeback
Moving cost into the consuming team's budget.
TBM and AI
Cost pools and allocation in a TBM model.
All templates
Every downloadable template on the site, by phase.