When AI spend grows large enough to allocate, organizations reach for one of two models. Chargeback, where teams are billed for what they consume, or showback, where teams see their consumption without being charged. Both depend entirely on the same foundation: reliable attribution. Get the tracking right and either model works; get it wrong and neither does.
What each model needs
Chargeback puts AI cost against a team's budget, which creates a direct financial incentive to manage it. Showback surfaces the same cost without the budget consequence, which encourages awareness without discouraging use. The choice is cultural, but the plumbing underneath is identical.
The attribution both require
Both models require every dollar of AI spend to map to a team or cost center. At the API layer, that means one key per team with a registry mapping keys to owners. At the SaaS layer, it means seats assigned to named users in named teams. Without this mapping, chargeback produces disputes over accuracy and showback produces reports teams cannot act on.
The monthly allocation run
Once attribution exists, allocation is mechanical. Pull the vendor usage by key or seat, look up the owner in the registry, sum by team, and post to the right budget or report. This runs monthly and, once the registry is built, takes minutes rather than the days a manual reconciliation would.
Choosing between them
A practical pattern is to start with showback while attribution matures and teams learn their numbers, then move to chargeback for production workloads once the data is trusted. Match the model to the accountability culture you actually have. If exceeding a budget carries real consequences, chargeback will drive behavior. If it does not, showback delivers the same awareness at lower friction.
Chargeback and showback are two answers to the same question, and both rest on tracking that attributes cleanly. Build the attribution first, and the choice between them becomes a straightforward matter of culture.