AI Agent Cost Allocation: Chargebacks and Showback for Multi-Department Usage

The Hidden Cost Explosion of Enterprise AI Agents

When AI agents launch, the monthly LLM bill is modest — perhaps a few hundred dollars. Six months later, it reaches five figures and no one can explain where the money went. Finance asks which department is responsible. Engineering points at usage logs that show API calls but not dollar amounts. The support team claims they barely use the agent, while the data shows they generate 60% of the traffic.

Cost allocation solves this by attributing every dollar of AI spending to the team, project, or cost center that generated it. This is not just accounting — it changes behavior. Teams that see their actual costs make smarter decisions about prompt design, model selection, and caching strategies.

Cost Tracking Architecture

Every agent request generates a cost record that captures the LLM provider charges (based on token counts and model pricing), infrastructure costs (compute, memory, storage), and any third-party tool costs.

flowchart LR
    subgraph IN["Inputs"]
        I1["Monthly call volume"]
        I2["Average deal value"]
        I3["Current answer rate"]
        I4["Receptionist cost<br/>per month"]
    end
    subgraph CALC["CallSphere Captures"]
        C1["Missed calls converted<br/>at 24 by 7 coverage"]
        C2["Receptionist payroll<br/>displaced or freed"]
    end
    subgraph OUT["Outputs"]
        O1["Recovered revenue<br/>per month"]
        O2["Operating cost saved"]
        O3((Net ROI<br/>monthly))
    end
    I1 --> C1
    I2 --> C1
    I3 --> C1
    I4 --> C2
    C1 --> O1 --> O3
    C2 --> O2 --> O3
    style C1 fill:#4f46e5,stroke:#4338ca,color:#fff
    style C2 fill:#4f46e5,stroke:#4338ca,color:#fff
    style O3 fill:#059669,stroke:#047857,color:#fff

from dataclasses import dataclass, field
from datetime import datetime
from uuid import uuid4

MODEL_PRICING = {
    "gpt-4o": {"input_per_1k": 0.0025, "output_per_1k": 0.01},
    "gpt-4o-mini": {"input_per_1k": 0.00015, "output_per_1k": 0.0006},
    "claude-sonnet-4": {"input_per_1k": 0.003, "output_per_1k": 0.015},
    "claude-haiku": {"input_per_1k": 0.00025, "output_per_1k": 0.00125},
}

@dataclass
class CostRecord:
    record_id: str = field(default_factory=lambda: str(uuid4()))
    request_id: str = ""
    timestamp: str = field(
        default_factory=lambda: datetime.utcnow().isoformat()
    )
    user_id: str = ""
    department: str = ""
    cost_center: str = ""
    agent_id: str = ""
    model: str = ""
    input_tokens: int = 0
    output_tokens: int = 0
    llm_cost_usd: float = 0.0
    infra_cost_usd: float = 0.0
    tool_cost_usd: float = 0.0
    total_cost_usd: float = 0.0

class CostCalculator:
    def __init__(self, pricing: dict = MODEL_PRICING):
        self.pricing = pricing

    def calculate(
        self,
        model: str,
        input_tokens: int,
        output_tokens: int,
        tool_calls: int = 0,
    ) -> dict:
        model_price = self.pricing.get(model, self.pricing["gpt-4o"])
        llm_cost = (
            (input_tokens / 1000) * model_price["input_per_1k"]
            + (output_tokens / 1000) * model_price["output_per_1k"]
        )
        infra_cost = 0.0001  # base per-request infrastructure cost
        tool_cost = tool_calls * 0.001  # per tool execution cost

        return {
            "llm_cost_usd": round(llm_cost, 6),
            "infra_cost_usd": round(infra_cost, 6),
            "tool_cost_usd": round(tool_cost, 6),
            "total_cost_usd": round(llm_cost + infra_cost + tool_cost, 6),
        }

Department Attribution and Reporting

Each request is tagged with the user's department and cost center from the SSO claims. Aggregation queries produce monthly reports per department, per agent, and per cost center.

class CostReporter:
    def __init__(self, db_pool):
        self.db = db_pool

    async def department_summary(
        self, year: int, month: int
    ) -> list[dict]:
        rows = await self.db.fetch(
            """
            SELECT
                department,
                cost_center,
                COUNT(*) AS total_requests,
                SUM(input_tokens) AS total_input_tokens,
                SUM(output_tokens) AS total_output_tokens,
                ROUND(SUM(llm_cost_usd)::numeric, 2) AS llm_cost,
                ROUND(SUM(infra_cost_usd)::numeric, 2) AS infra_cost,
                ROUND(SUM(tool_cost_usd)::numeric, 2) AS tool_cost,
                ROUND(SUM(total_cost_usd)::numeric, 2) AS total_cost
            FROM cost_records
            WHERE EXTRACT(YEAR FROM timestamp) = $1
            AND EXTRACT(MONTH FROM timestamp) = $2
            GROUP BY department, cost_center
            ORDER BY total_cost DESC
            """,
            year, month,
        )
        return [dict(r) for r in rows]

    async def agent_cost_breakdown(
        self, agent_id: str, days: int = 30
    ) -> dict:
        rows = await self.db.fetch(
            """
            SELECT
                date_trunc('day', timestamp) AS day,
                model,
                COUNT(*) AS requests,
                SUM(input_tokens) AS input_tokens,
                SUM(output_tokens) AS output_tokens,
                ROUND(SUM(total_cost_usd)::numeric, 4) AS daily_cost
            FROM cost_records
            WHERE agent_id = $1
            AND timestamp > NOW() - INTERVAL '%s days'
            GROUP BY day, model
            ORDER BY day DESC
            """ % days,
            agent_id,
        )
        return {"agent_id": agent_id, "period_days": days, "data": [dict(r) for r in rows]}

    async def top_cost_users(
        self, department: str, month: int, limit: int = 10
    ) -> list[dict]:
        rows = await self.db.fetch(
            """
            SELECT
                user_id,
                COUNT(*) AS requests,
                ROUND(SUM(total_cost_usd)::numeric, 2) AS total_cost,
                ROUND(AVG(total_cost_usd)::numeric, 4) AS avg_cost_per_request
            FROM cost_records
            WHERE department = $1
            AND EXTRACT(MONTH FROM timestamp) = $2
            GROUP BY user_id
            ORDER BY total_cost DESC
            LIMIT $3
            """,
            department, month, limit,
        )
        return [dict(r) for r in rows]

Budget Management and Alerts

Departments set monthly budgets. The system tracks spending against budgets in real time and sends alerts at configurable thresholds — typically 50%, 80%, and 100%.

class BudgetManager:
    def __init__(self, db_pool, notifier):
        self.db = db_pool
        self.notifier = notifier

    async def check_budget(self, department: str, cost_center: str) -> dict:
        budget = await self.db.fetchrow(
            "SELECT monthly_budget_usd, alert_thresholds "
            "FROM department_budgets "
            "WHERE department = $1 AND cost_center = $2",
            department, cost_center,
        )
        if not budget:
            return {"status": "no_budget_set"}

        current_spend = await self.db.fetchval(
            """
            SELECT COALESCE(SUM(total_cost_usd), 0)
            FROM cost_records
            WHERE department = $1 AND cost_center = $2
            AND date_trunc('month', timestamp) = date_trunc('month', NOW())
            """,
            department, cost_center,
        )

        utilization = (current_spend / budget["monthly_budget_usd"]) * 100
        thresholds = budget["alert_thresholds"]  # e.g. [50, 80, 100]

        for threshold in sorted(thresholds):
            if utilization >= threshold:
                await self.notifier.send_budget_alert(
                    department=department,
                    cost_center=cost_center,
                    utilization_pct=round(utilization, 1),
                    threshold=threshold,
                    current_spend=round(current_spend, 2),
                    budget=budget["monthly_budget_usd"],
                )

        return {
            "department": department,
            "budget_usd": budget["monthly_budget_usd"],
            "current_spend_usd": round(current_spend, 2),
            "utilization_pct": round(utilization, 1),
        }

Chargeback vs. Showback

Chargebacks transfer actual costs to department budgets. Showback reports costs without transferring them. Most organizations start with showback to build awareness, then move to chargebacks once departments understand their usage patterns and have had time to optimize.

FAQ

How do you handle shared agents used by multiple departments?

Attribute costs to the department of the user making the request. If a shared analytics agent is used by sales, marketing, and finance, each department pays for its own usage. For agents that run background tasks without a user context, allocate costs to the agent owner's department or split proportionally based on historical usage patterns.

What is the best granularity for cost tracking — per request, per session, or per day?

Track at per-request granularity and aggregate upward. Per-request records let you identify expensive individual queries, unusual usage patterns, and optimization opportunities. Daily or monthly aggregates lose this detail. Storage cost for per-request records is minimal compared to the LLM costs you are tracking.

How do you account for cached responses that avoid LLM calls?

Track cache hits as zero-cost LLM requests but include the infrastructure cost (cache storage, lookup time). This gives departments credit for their caching efficiency and incentivizes prompt designs that maximize cache hit rates. The cost report should show both actual spend and estimated savings from caching.

---

#EnterpriseAI #CostAllocation #Chargebacks #FinOps #BudgetManagement #CostTracking #AgenticAI #LearnAI #AIEngineering