Showback or chargeback: how to allocate AI costs by team.

01Showback and chargeback, defined

Both are ways of making a shared bill visible to the people who caused it. Showback is a report: here is what your team spent on AI this month, against the budget you were given. Chargeback is an accounting entry: that amount is moved from a central budget to yours.

The words come from cloud cost management, where they have been used for a decade. They carry over to AI with one difference that matters: the cost of a model call depends on choices a team may not control, and on prices that change more often than cloud prices ever did.

02Why AI makes the choice harder than cloud did

• Spend arrives through shared credentials, so the first job is attribution, before any allocation is fair.
• Cost depends on the model chosen, the length of the prompt and how often an agent loops. A team charged for a model it did not choose will argue the bill, and it will be right.
• AI arrives inside products you already license, as features switched on by the vendor. Those costs sit in software budgets and never reach the AI line.
• Model prices move during the year. A charge calculated at January prices and settled in June needs a rule for which price applies.

The cost of getting it wrong is visible across the Tasman. In research for Elastic among more than 500 senior decision-makers at Australian organisations, reported by CFOtech in June 2026, one in three had exceeded their AI budget in the last financial year. Thirty-two percent had paused, cancelled or wound back deployments because the output could not justify the cost, and 8 percent were tracking whether AI delivered revenue or savings.

The FinOps Foundation's guidance on AI token costs, published in June 2026, sets the order: showback to application and team owners first, chargeback later, where organisational maturity supports it.

The common mistake is to switch on chargeback because spend has grown, before anyone trusts the numbers. The first month of disputed bills costs more goodwill than the spend it was meant to curb.

03When to use which

• Start with showback for every team, as soon as calls are attributed. It changes behaviour on its own when it arrives during the month, while there is still time to act on it.
• Move a team to chargeback when its AI spend is large enough to matter in its own budget, its attribution has matched the invoice for two or three months, and it chooses its own models and agents.
• Keep costs central where no team controls them: the platform, shared evaluation, experiments under a fixed cap, and agents that serve the whole company.

04Allocation rules that hold up

1. Allocate by the call, using the team, project and workflow recorded when the call ran. Headcount and revenue splits are guesses that teams learn to ignore.
2. Charge the team that caused the call, which is often different from the team that owns the key.
3. Split shared platform cost on measured usage, and say in advance what the measure is.
4. Give experiments their own central line with a cap, so a team does not avoid a useful test because it would land on its budget.
5. Fix the price rule: calls are costed at the price in force when they ran.
6. Publish the rules before the first report. A rule written after a dispute looks like a ruling.

05What actually changes behaviour

Allocation tells a team what it spent. It does not stop the spend. Teams change what they do when the number arrives while the month is open, and when the budget they are measured against holds at a cap they can see coming. Chargeback without a cap moves an overrun from one budget to another. Showback with a cap, approved by a named person, usually prevents it.