When IT costs are centralised and owned by a single budget, technology feels like a free and unlimited resource to every business unit. The result is predictable: uncontrolled consumption, no accountability, and a CFO who cannot explain why the technology bill keeps growing while no one claims ownership.
The challenge has intensified dramatically. Cloud, SaaS, and now AI workloads have made IT spending highly variable. A single team's decision to deploy a premium LLM model or spin up GPU clusters can shift the monthly bill by tens of thousands of dollars — with no approval workflow and no budget impact to the team that made the decision.
This is the problem chargeback and showback models exist to solve.
These two models are related but fundamentally different in their financial mechanics and behavioural impact.
Research consistently shows that chargeback policies drive more conscientious spending than showback alone. Understanding the consequences and actually facing them are two different things.
Despite their clear logic, the majority of enterprise chargeback implementations underperform or are abandoned within 18 months. The failure patterns are consistent:
We recommend a phased approach that builds trust before enforcing accountability, scales from simple to sophisticated, and embeds governance from the start.
Establish baseline visibility. This phase is showback-only. No money moves. Every department sees what they would be charged, understands how the numbers are calculated, and has time to ask questions and clean up waste before accountability begins.
- Build a service catalogue mapping IT services to business consumption
- Implement automated cost ingestion from cloud, SaaS, and infrastructure
- Deploy resource tagging standards across all environments
- Publish monthly showback reports to every department head
- Establish a cross-functional FinOps governance board
Transition from showback to chargeback on the highest-volume, most clearly attributable services first. Start where attribution is unambiguous — cloud compute and SaaS licences.
- Activate chargeback for cloud compute, storage, and SaaS licences
- Implement tiered pricing: standard, premium, and innovation tiers
- Deploy approval workflows for high-cost services (premium AI models, GPU clusters)
- Create budget thresholds and anomaly alerts per department
- Begin quarterly chargeback review meetings with business unit leaders
Extend chargeback to shared services, AI workloads, and cross-functional infrastructure. Introduce cost-per-outcome KPIs and continuous optimisation.
- Allocate shared infrastructure costs using weighted consumption models
- Implement AI token chargeback by use case, team, and model tier
- Introduce cost-per-outcome KPIs alongside cost-per-unit metrics
- Run monthly optimisation reviews with automated recommendation engines
- Benchmark internal rates against external market pricing
Not every service should use the same pricing approach. The right model depends on how the service behaves and how it is consumed.
AI workloads present unique challenges that traditional chargeback models were not designed to handle:
- Variable consumption: A single AI workflow can consume anywhere from 1,000 to 1,000,000 tokens depending on input complexity, model selection, retry patterns, and caching efficiency.
- Model selection impact: The same task processed by a budget model versus a premium model can differ by 50× in cost.
- Multi-model orchestration: Modern AI applications route queries through multiple models in sequence. Attribution requires tracing the full chain.
- Shared infrastructure: Fine-tuned models, vector databases, and embedding services are often shared across teams.
The solution is not to avoid charging back AI costs — it is to build the observability and attribution infrastructure that makes fair allocation possible. This requires token-level tracking by workflow, model, team, and use case.
CFO reporting should include four views that together tell the complete story of IT economic performance:
| Report View | What It Answers |
|---|---|
| AI Spend by Business Unit | Who is consuming, how much, and is it growing faster than value creation? |
| AI ROI by Use Case | Which deployments generate measurable business outcomes — and which are cost centres? |
| Cost-per-Outcome Trend | Is the organisation getting more efficient over time, or is consumption outpacing value? |
| High-Cost / Low-Value Workloads | Which deployments should be optimised, migrated to cheaper models, or retired? |