AI Chargeback and Showback: How to Allocate AI Cost Across Business Units

Once you're measuring work-item-level AI cost, the next question is governance: who pays for it? In a large organization with multiple AI agents, that's not academic. A healthcare system might have a claims AI (revenue cycle), a prior authorization AI (clinical operations), a risk assessment AI (underwriting), and a patient engagement AI (patient care). Do they split the $200K/month AI infrastructure budget equally? Proportionally to volume? Or should each team have a dedicated budget and cost center?

This is where chargeback and showback come in. Both are allocation mechanisms, but they have different enforcement models.

Showback: Visibility Without Accountability

Showback means you measure and report each business unit's AI spend, but you don't actually charge them for it. The cost stays on a central P&L (finance, technology, shared services) but you publish a report showing each team their proportional share. It's transparent but non-binding.

Showback is the cautious step. You get the benefits of visibility—each team can see the cost of their AI—without forcing the organizational politics of who pays for shared infrastructure. A typical showback report looks like:

| Business Unit | AI Spend | % of Total | Cost Per Work Item | YoY Trend | | --- | --- | --- | --- | --- | | Revenue Cycle (Claims) | $85,000 | 42.5% | $0.45/claim | +12% | | Clinical Ops (Prior Auth) | $48,000 | 24% | $0.80/auth | +8% | | Risk & Analytics | $38,000 | 19% | $1.20/model | +5% | | Patient Engagement | $29,000 | 14.5% | $0.38/interaction | +18% | | Total | $200,000 | 100% | - | +10.5% |

This lets each team understand their consumption without forcing them to reduce it. It's most useful in Stage 2 or 3 of the 5-Stage Maturity Curve—when you have visibility but haven't yet built the operational accountability to enforce cost ownership.

Chargeback: Allocating Cost to P&Ls

Chargeback means you actually debit each business unit's P&L for their AI consumption. The revenue cycle team is charged $85K, clinical operations is charged $48K, and so on. Each team's margin and ROI calculation must account for the cost they were charged.

Chargeback creates accountability. A team that's paying $48K for their prior authorization AI will ask harder questions: "Is this actually saving us money? Should we adjust the automation rate? Should we switch vendors?" Chargeback incentivizes real decision-making, not just awareness.

Chargeback is more common in Stage 4 (Optimized) of the maturity curve, where business units have the autonomy and accountability to manage their AI spend as part of their P&L.

How to Allocate: The $50K Example

Here's a concrete example: a mid-market healthcare company has $50,000/month in AI infrastructure cost (shared GPU resources, embedding database, observability platform, gateway infrastructure) and wants to allocate it across four business units: Revenue Cycle ($85K), Clinical Operations ($48K), Risk & Analytics ($38K), and Patient Engagement ($29K).

There are three common allocation methods:

Method 1: Proportional to volume (work items processed) Revenue Cycle processes 100,000 claims/month, Clinical Ops processes 60,000 prior auths/month, Risk processes 32,000 risk assessments/month, Patient Engagement processes 78,000 interactions/month. Total: 270,000 work items.

• Revenue Cycle: (100,000 / 270,000) * $50,000 = $18,518
• Clinical Ops: (60,000 / 270,000) * $50,000 = $11,111
• Risk & Analytics: (32,000 / 270,000) * $50,000 = $5,925
• Patient Engagement: (78,000 / 270,000) * $50,000 = $14,444

Method 2: Proportional to direct spend (API cost only) Revenue Cycle's API cost is $85,000 (visible), Clinical Ops is $48,000, Risk is $38,000, Patient Engagement is $29,000. Total: $200,000. Now allocate the $50K infrastructure overhead proportionally:

• Revenue Cycle: ($85,000 / $200,000) * $50,000 = $21,250
• Clinical Ops: ($48,000 / $200,000) * $50,000 = $12,000
• Risk & Analytics: ($38,000 / $200,000) * $50,000 = $9,500
• Patient Engagement: ($29,000 / $200,000) * $50,000 = $7,250

Method 3: Fixed allocation plus usage-based overflow Allocate $40K of the $50K equally ($10K per team) as a "floor," then allocate the remaining $10K based on overage:

• Revenue Cycle: $10,000 + ($85K / $200K) * $10K = $14,250
• Clinical Ops: $10,000 + ($48K / $200K) * $10K = $12,400
• Risk & Analytics: $10,000 + ($38K / $200K) * $10K = $11,900
• Patient Engagement: $10,000 + ($29K / $200K) * $10K = $11,450

Which method you choose depends on your organization's priorities. Method 1 (volume-based) is fairest if all work items have similar infrastructure cost. Method 2 (API-cost-based) aligns with the principle "you pay for what you use." Method 3 (fixed plus usage) creates shared ownership while incentivizing efficiency.

Handling Shared Components

Some infrastructure components are genuinely shared and hard to allocate: a central embedding database serving all four teams, a unified observability platform, a shared rate-limit gateway. There are a few options:

Absorb the cost centrally. Keep shared components on a central P&L (technology, finance, shared services). Charge each business unit only for their direct AI costs (API, direct integrations, team-specific infrastructure). This is the simplest model and most common in early stages.

Allocate using a utilization metric. Measure the utilization of the shared component (embedding queries per team, observability log volume per team, gateway throughput per team) and allocate proportionally. This is more accurate but requires instrumentation.

Use a hybrid model. Fixed allocation of shared components (each team pays a monthly "infrastructure tax") plus usage-based allocation of variable components. This splits the difference between simplicity and accuracy.

Governance: Who Owns the Allocation Methodology?

One practical question: who decides how to allocate shared costs? Finance usually wants allocation to be rigorous and audit-friendly (Method 2: proportional to direct spend). Engineering often wants it to be simple (Method 1: volume-based). Operations teams want it to incentivize good behavior (Method 3: fixed plus usage-based).

The answer: your finance and engineering leaders should align on the method before you start charging. If you don't have alignment, you'll spend more time resolving allocation disputes than you save from cost awareness.

A practical approach: start with showback using a simple method (Method 1 or 2), run it for 2 months, gather feedback from business unit leaders, then switch to chargeback with the same method. This gives teams time to adjust and builds trust in the allocation numbers before real money starts moving.

From Showback to Chargeback: The Maturity Path

Most organizations start with showback: "Here's what you're spending on AI." After 2-3 months of visibility, they move to chargeback: "Here's what you're being charged for AI, and it affects your P&L."

The transition requires two things: (1) agreement on the allocation method (you want consensus before you start charging), and (2) clear communication that the charge is based on usage, not arbitrary. If teams understand they can reduce their charge by reducing their AI spend or improving cost per work item, they'll optimize rather than resent the charge.

The goal of chargeback is not revenue-neutral (it's not a profit center for shared services) but rather accountability. Each team becomes responsible for their AI ROI, not just the company as a whole. That accountability is what drives the move from Stage 3 (Allocated) to Stage 4 (Optimized) on the maturity curve.

Once chargeback is in place, teams have every incentive to optimize: reduce AI volume, switch to cheaper models, improve automation rates, reduce human review overhead. That's when the real economic gains start flowing.

Learn more about how to structure multi-tenant AI cost allocation for SaaS companies, or return to the pillar article on AI cost attribution.