Table of Contents
AI adoption is no longer the hard part. AI economics is.
Case in point: Uber burned through its entire 2026 AI budget in 4 months. Its own COO couldn’t say what the spend actually bought.
That’s the problem finance leaders are waking up to: Almost every enterprise can now see what it spends on AI, and almost none can control it. Visibility is not governance. Dashboards show what happened; they don’t forecast, optimize or bring that spend into the plan. And the bill arrives before anyone has accounted for it.
At Uber, spend surged as Claude Code spread across engineering. By March, 84% of engineers were using the tool, and AI was generating roughly 70% of committed code. But adoption also was fueled by an internal AI “tokenmaxxing” culture, complete with a leaderboard that employees gamed by consuming tokens on low-value tasks. Forbes ran the whole thing under the headline, “AI Costs More Than The People It Replaced.”
Uber isn’t alone.
- Microsoft reportedly moved developers away from Claude Code as costs climbed.
- Nvidia Vice President Bryan Catanzaro told Axios that for his team, “the cost of compute is far beyond the costs of the employees.”
- Tesla recently told staff it’s capping AI spend at $200 a week per employee, with higher usage requiring approval.
The moment AI shows up in the work, it changes what the work costs; and finance is usually the last to find out. Workforce planning used to mean salaries, benefits and subscription fees. Now it also means the cost of every prompt, response, workflow and agent consuming tokens, compute and employee time. Left unmanaged and without a governed process to plan for it, AI spend spreads faster than anyone can forecast it, assign it or connect it back to real measurable value.
Seeing last month’s AI bill is the easy part. Getting ahead of next month’s spend is where almost every finance team is stuck. That’s where AI spend management comes into play.
What Is AI Spend Management?
AI spend management is the process of planning, forecasting, governing, allocating and optimizing the costs associated with AI usage across a business. It includes token consumption, AI software subscriptions, cloud compute, infrastructure costs, model usage, financial oversight and the labor costs tied to AI-enabled work.
The more important questions for finance leaders are what is driving demand, which teams and use cases are responsible, how that demand affects labor economics and whether the resulting productivity gains justify the cost.
Mike Zack, general manager of Acterys, describes this as a move from AI spend visibility to planning-led control. He explains how the Acterys platform’s Microsoft Power BI write-back capabilities help organizations address their AI spend management challenges:
“Write-back and planning in Power BI gives finance leaders the ability to not only see who’s using AI and where token consumption is happening, but take the next step of optimization: the ability to plan and forecast in Power BI and get ahead of that usage. That’s how organizations move from simply tracking tokens to forecasting, and actually controlling, spend.”
What is AI Token Consumption, and How Do You Plan for it?
AI token consumption is the volume of data an AI model processes when users or applications submit prompts and generate responses. As AI usage scales from individual productivity to enterprise workflows, token consumption rises. Because many AI providers charge based on the number and type of tokens used, token consumption is a key driver of AI operating costs.
Planning for AI token consumption means forecasting the cost of AI usage across tools, models, workflows and teams. It enables business leaders to assign cost ownership, identify the biggest drivers of spend and manage AI costs before they become difficult to predict or control.
For finance teams, the most important shift is that planning for AI token usage doesn’t need to, nor should it, sit apart from workforce planning. It should be connected to how the business plans headcount, role mix, productivity and labor cost.
What is AI Token Forecasting?
AI token forecasting is the process of estimating how much an organization’s AI tools and applications will be used by its workforce and translating that usage into projected costs.
Whether a company plans to hire more developers or simply expand AI access across the sales team, those workforce decisions may directly affect token consumption. Similarly, if a company expects AI to improve productivity, finance needs a way to model whether that productivity offsets the cost of increased AI usage.
Being able to plan and forecast AI token usage gives finance leaders a way to control and optimize growth, instead of reacting to it after the fact (or after they receive the bill).
At a foundational level, an AI token forecast must answer five questions upon which to base estimations:
- Who is using AI?
- How much are they using?
- What is that usage costing?
- How does usage vary by role, team or workflow?
- What business value is being created?
As Zack explains,
“Enterprises today want to be able to understand if a single individual used, let’s say, a million tokens, what did that actually create and how do you link that back to efficiencies and productivity?”
Dashboards can show spend after the fact, but dashboards do not forecast, model, approve or push decisions back into the plan. Finance teams today must connect token consumption to headcount, productivity, workflows and outcomes.
A simple AI token forecasting model may include:
Input | What it measures | Why it matters for workforce planning |
Number of AI users | Employees, teams or systems with AI access | Establishes the workforce population driving demand |
Token consumption per user | Average prompt and response volume | Helps forecast AI cost by role or department |
Cost per token or model call | Pricing by model, vendor or platform | Connects employee usage to actual spend |
Workflow frequency | How often AI-enabled workflows run | Captures recurring costs tied to how work gets done |
Department or cost center | Where usage originates | Enables allocation, chargeback and labor cost planning |
Business outcome | Time saved, revenue impacted, risk reduced or process improved | Connects workforce productivity to AI investment |
The idea is to track usage as well as allow finance teams to forecast how that usage may change as more employees gain access, more workflows are automated and more AI agents begin operating across the business.
That means building assumptions around adoption rates, usage intensity by role, model selection, workflow automation, productivity improvement and business value.
5 Questions Finance Should Ask Before Expanding AI Access
- Which departments, roles or employee groups will receive expanded AI access?
- How will AI usage change the fully loaded cost of each role?
- What workflows will employees use AI for?
- How much token or model consumption is expected per employee or workflow?
- How will the business measure whether AI usage improved productivity, output or decision quality?
The goal is not to restrict AI adoption. The goal is to understand the AI cost per employee, make AI adoption financially sustainable and increase workforce planning accuracy.
What is AI Cost per Employee?
AI cost per employee is the estimated annual cost of AI usage, tools, infrastructure and governance associated with each employee. It helps finance and HR teams understand how AI changes the all-in cost of labor.
This is a vital new metric to watch as AI becomes part of daily work.
The AI Cost-per-Employee Formula
AI cost per employee = AI subscriptions + token usage + cloud compute + governance/admin costs
Zack frames this as a blind spot in how companies think about labor cost:
“A lot of people are not thinking about the all-in costs of an employee. Of course, we think about the salary, the variables and the benefits. But if this person comes aboard, and we use an average of 500,000 tokens per employee, how does that factor into workforce planning?”
That question is becoming more urgent as AI access expands across the workforce. A developer building AI-assisted applications may consume far more tokens than a finance analyst using AI for variance commentary. A sales operations team may use AI for forecasting, account research and workflow automation. A customer support team may use AI agents that generate recurring model consumption at scale.
That means finance leaders need a more nuanced view of labor costs.
| Role or team | Potential AI cost drivers | Workforce planning consideration |
| Software development | Code generation, testing, documentation, AI-assisted application development | Higher AI cost may be justified if development velocity improves |
| Finance | Forecast commentary, variance analysis, scenario modeling, reporting automation | Usage should connect to faster cycles, better forecasts and reduced manual work |
| Sales | Account research, outreach personalization, pipeline analysis | Usage should connect to productivity, conversion or revenue outcomes |
| HR | Job descriptions, workforce planning, policy support, employee service workflows | Usage should connect to time savings and better employee experience |
| Operations | Process automation, anomaly detection, workflow recommendations | Usage should connect to efficiency, throughput and cost reduction |
The key is to avoid treating AI cost as a generic IT expense. If employees are using AI to perform work, then AI consumption becomes part of the cost of that work.
For workforce planning, finance and HR teams can start by modeling AI cost per employee in tiers:
| Usage tier | Example user type | Workforce planning assumption |
| Low AI usage | Occasional AI-assisted writing, search or summarization | Minimal incremental labor cost |
| Moderate AI usage | Regular use in reporting, analysis, research or workflow support | Per-user monthly AI cost |
| High AI usage | Developers, data teams, AI-heavy operators or agent managers | Higher token and compute forecast by role |
| Automated/agentic usage | AI agents running recurring workflows | Cost tied to workflow frequency, not just headcount |
This provides a more accurate view of how headcount growth may affect AI spend.
AI Chargeback Modeling and Spend Optimization for Finance and IT
As AI usage expands, finance and IT teams need a shared model for managing ownership, allocation and optimization. Chargeback is an important part of that model, but assigning costs alone won’t prevent AI spend from becoming unpredictable.
If AI spend sits in one centralized IT or finance budget, business units may have little incentive to optimize usage. If costs are allocated without context, departments push back. If spend is not assigned at all, finance struggles to explain why AI costs are rising. And if finance and IT evaluate spend separately, organizations may miss opportunities to partner to optimize costs and usage through changes to models, workflows, access levels or infrastructure.
AI chargeback creates accountability by assigning AI costs to the teams, departments, workflows, roles or cost centers that consume them. Finance and IT can then use that visibility to optimize spending, establish usage thresholds and determine whether higher consumption reflects greater productivity or inefficient deployment.
This is especially important for workforce planning because AI costs scale differently across employee groups. Two departments often have the same headcount but very different AI consumption profiles. One team may rely heavily on AI-enabled workflows, while another uses AI only occasionally. One role becomes more productive with AI, while another consumes tokens without measurable improvement.
Zack shared an example of a large bank that could see the total number but lacked visibility into ownership: “Finance knew because they know the total number, but they didn’t see where it was coming from.” The next step, he explained, was being able to reassign budget “to a specific department or person.” Once that ownership is established, finance and IT can work together with other business leaders to determine whether the underlying usage should be funded, optimized or reconsidered.
A shared AI chargeback and optimization model helps answer:
- What departments are consuming the most AI tokens?
- Which roles or employee groups have the highest AI cost per employee?
- Which workflows are driving the highest recurring costs?
- Where should AI spend be allocated or charged back?
- Which business units are creating measurable value from AI usage?
- Which models, tools or workflows could be redesigned to reduce consumption?
- Which costs should remain centralized because they support enterprise-wide capabilities?
- Where should additional AI investment be approved based on expected business impact?
There are several possible models.
Chargeback model | How it works | Best for |
Centralized budget | AI costs remain in IT or a corporate innovation budget | Early-stage AI adoption or experimentation |
Department allocation | Costs are assigned by department or business unit | Companies with broad AI adoption across functions |
Usage-based chargeback | Costs are assigned based on actual token, model or platform consumption | Mature organizations with reliable usage data |
Workforce-based allocation | Costs are assigned based on role, headcount or AI usage tier | Companies integrating AI into workforce planning |
Workflow-based chargeback | Costs are assigned to specific AI-enabled processes or applications | Companies using AI in operational workflows |
Hybrid model | Shared infrastructure remains centralized while variable usage is allocated | Organizations balancing governance and accountability |
A hybrid model is often the most practical starting point. Core AI platforms, governance, security and infrastructure may remain centrally funded, while variable usage costs can be allocated to departments based on consumption, headcount or workflow intensity. Finance can establish budgets, forecasts and approval thresholds, while IT evaluates the technical drivers of spend and identifies opportunities to improve efficiency via model selection, architecture, workflow design or access controls.
The value of chargeback isn’t just cost recovery. It’s better planning and optimization. When departments understand the cost of AI usage, they can make smarter decisions about hiring plans, productivity targets, workflow automation and AI access. When finance and IT review that usage together, they can also determine whether rising costs reflect productive adoption or inefficient consumption.
Finance and IT should work together to define:
- Which AI costs are fixed versus variable
- Which costs are enterprise-wide versus department-specific
- How token consumption will be measured by role, team or workflow
- How AI cost per employee will be incorporated into workforce plans
- How shared AI infrastructure will be allocated
- What usage or cost thresholds trigger review
- How departments will justify incremental AI budget
- Which technical changes could reduce token consumption or model costs
- How business value and productivity will be evaluated alongside usage
- Who can approve new tools, models, agents or high-cost workflows
The goal is not to punish high AI usage. High consumption may be a positive signal when it is tied to workforce productivity and business value. Finance and IT need a shared way to distinguish productive consumption from tokenmaxxing and unmanaged consumption to shape AI spend before it becomes a budget problem.
How to Measure AI Token ROI
AI usage does not automatically equal AI value. An employee can use a million tokens and create meaningful business impact. Another employee or workflow can consume significant AI resources with little measurable return. When token usage is incentivized internally, that second pattern leads to tokenmaxxing, the same leaderboard behavior that ran up Uber’s bill. It’s exactly what optimization is built to catch: consumption that isn’t tied to value. Finance teams need a way to connect consumption to outcomes.
What Is AI Token ROI?
AI token ROI measures the business value created by AI usage relative to the cost of that usage. It helps organizations determine whether AI consumption is improving productivity, revenue, margin, forecast accuracy, decision velocity, customer experience or operational efficiency.
The AI Token ROI Formula
AI token ROI = business value generated from AI usage ÷ total AI usage cost
The hardest part of the equation is defining value.
AI value can be quantitative, qualitative or both.
AI usage metric | AI value metric |
Tokens consumed | Hours saved per employee |
Model calls | Revenue influenced |
Number of prompts | Forecast accuracy improved |
Active users | Cycle time reduced |
Cost by department | Manual work eliminated |
Cost by workflow | Margin improvement |
AI tool adoption | Employee productivity gains |
Workflow frequency | Risk reduced or compliance improved |
AI cost per employee | Output per employee improved |
Finance teams should avoid relying only on adoption metrics. High adoption may show that employees are using AI, but it does not prove that AI is improving the business.
A stronger approach is to connect AI usage to specific workflows, roles and KPIs. For example:
Workflow | Potential ROI measure | Workforce planning relevance |
Financial forecasting | Shorter planning cycles, improved forecast accuracy | Finance teams can support more strategic analysis without proportional headcount growth |
Variance analysis | Faster close process, reduced manual reporting time | Analysts can spend more time on insight and less on reconciliation |
Sales planning | Better quota allocation, improved pipeline visibility | Sales operations can support growth more efficiently |
Workforce planning | More accurate labor cost forecasts | HR and finance can better model headcount, productivity and AI consumption |
Customer support automation | Reduced handle time, improved response speed | Support teams can scale service levels without linear hiring |
Software development | Faster release cycles, reduced rework | Engineering capacity can increase without relying only on headcount growth |
Finance leaders should also gather feedback from the users doing the work. Usage data alone may show who consumed tokens, but only the employee or team doing the work may know whether AI reduced effort, improved output or changed the workflow.
Finance teams should use questions like these to evaluate AI value:
- Which workflows generate measurable ROI?
- Which roles are becoming more productive because of AI?
- Which AI use cases save the most employee time?
- Which teams consume the most tokens but show limited business impact?
- Which AI-enabled workflows improve forecast accuracy or decision speed?
- Where are employees repeatedly asking the same questions that could be turned into a standardized workflow?
The purpose of AI token ROI is not to reduce AI usage across the board. It is to direct AI investment toward the highest-value use cases and improve how the business plans for labor, productivity and growth.
Why Bad Data Makes AI More Expensive
AI costs do not only rise because more people use AI. They also rise when AI has to work harder to produce the right answer.
Bad data can make AI more expensive because it creates inaccurate outputs, repeated prompts, manual validation, rework and lost trust. When users cannot rely on AI-generated answers, they often ask the same question multiple ways, check the output manually or revert to old processes. That means the business pays for the AI interaction but does not capture the expected value. This is tokenmaxxing without the leaderboard. No one is gaming anything, but the tokens still get burned on answers that don’t add value.
This has direct implications for workforce planning.
If employees use AI but still need to manually validate every answer, the organization may end up paying twice: once for the AI consumption and again for the employee time required to correct or verify the output. That undermines the productivity assumptions finance teams may be building into workforce plans.
Zack describes this risk in practical terms:
“Not only do employees waste a bunch of tokens coming up with answers that aren’t accurate, but now they’re frustrated by the waste of time. They lose a level of trust in the AI.”
In other words, bad data can create both hard costs and soft costs.
Hard costs include additional token consumption, model calls, compute resources and platform usage. Soft costs include wasted employee time, duplicate work, slower decisions and reduced confidence in AI-enabled processes.
This is why AI economics begins with data quality.
If enterprise data is fragmented, inconsistent or poorly governed, AI may return different answers depending on where it pulls information from. A revenue number in a CRM may not match a revenue number in an ERP. A workforce planning assumption may live in a spreadsheet that is disconnected from the official forecast. A business definition may vary by department.
Without a trusted data foundation and shared business context, AI can accelerate confusion.
Data issue | AI cost impact | Workforce planning impact |
Fragmented data | More prompts and manual reconciliation | Employees spend more time validating outputs |
Inconsistent definitions | Conflicting answers | Productivity assumptions become unreliable |
Poor data quality | Incorrect outputs | Rework offsets AI efficiency gains |
Lack of governance | Unclear source of truth | Workforce plans rely on inconsistent assumptions |
Disconnected workflows | Repeated AI interactions | Higher consumption without clear labor productivity gains |
This is where the semantic layer becomes critical. A semantic layer organizes business data into shared meaning. It defines metrics, relationships, hierarchies, calculations and business rules so people and AI systems understand the context behind the data.
For AI spend management, this matters because better business context can reduce unnecessary consumption. If AI understands which revenue number to use, which forecast version is current and which workforce plan is approved, users are less likely to waste tokens asking follow-up questions or validating outputs manually.
A simple way to frame it:
Bad data leads to bad answers.
Bad answers lead to repeat prompts.
Repeat prompts lead to wasted tokens.
Wasted tokens lead to higher AI costs.
Higher costs without better workforce productivity weaken AI ROI.
Finance leaders should view data quality as an AI cost optimization strategy and a workforce productivity strategy.
Scenario Planning for AI Cost Volatility
AI costs can change quickly. Usage may grow as more employees adopt tools. Model pricing may shift. New AI agents may run recurring workflows. Cloud infrastructure demands may increase. Business teams may move from experimentation to production. A new product, acquisition or planning cycle may introduce new AI workloads.
Scenario planning helps finance teams prepare for that volatility.
Instead of building one static AI budget, finance teams should model multiple scenarios based on adoption, usage intensity, model mix, workflow automation, workforce plans and business value.
This is where forecasting becomes central to optimization. As Zack explained,
“It’s like anything in a grocery store. You buy things in bulk, you save money… If you’re able to forecast this out and you’re able to then have control around it, it just makes things more efficient, which leads back to optimization.”
Common AI cost scenarios include:
Scenario | Description | Workforce planning implication |
Base case | AI usage grows at expected adoption rates | Budget reflects current rollout and headcount plan |
High adoption case | More employees use AI more frequently than expected | Token and subscription costs increase by role or department |
Hiring growth case | Headcount expands in AI-heavy roles | AI cost rises alongside labor cost |
Agentic workflow case | AI agents begin running recurring processes | Costs shift from user-driven to workflow-driven |
Premium model case | Teams rely on more expensive models for complex work | Cost per employee or workflow rises |
Optimization case | Better workflows, data and governance reduce unnecessary usage | Cost per outcome improves |
Productivity gain case | AI reduces time required for recurring work | Workforce capacity increases without proportional hiring |
Pricing change case | Vendor or model pricing changes | Forecast needs sensitivity analysis |
Scenario planning allows finance leaders to ask better questions before costs hit the budget.
Example planning prompts:
- What happens if AI adoption doubles in the next six months?
- What happens if token consumption per employee increases by 25%?
- What happens if high-usage teams move to more expensive models?
- What happens if AI-heavy roles grow faster than expected?
- What happens if AI agents run daily workflows instead of occasional employee prompts?
- What happens if better data quality reduces repeat prompts by 15%?
- What happens if AI improves forecast accuracy or reduces planning cycle time?
- What happens if AI reduces the need for incremental headcount in a specific workflow?
Scenario planning should also connect AI cost to workforce value. A high-cost scenario may still be desirable if it improves productivity, increases capacity or reduces the need for additional hiring. A low-cost scenario may not be effective if it limits adoption of high-value use cases.
Finance teams should evaluate AI scenarios across both cost and outcome.
Scenario question | Cost view | Workforce value view |
What if usage increases? | Higher token and compute costs | More productivity or automation potential |
What if model costs rise? | Higher cost per workflow or employee | Need to prioritize higher-value use cases |
What if adoption stalls? | Lower spend | Lower productivity transformation impact |
What if data quality improves? | Lower wasted consumption | Higher employee trust and better decisions |
What if AI agents scale? | More recurring usage | More automated processes and faster execution |
What if headcount grows? | Higher AI consumption by role | Need to include AI cost in labor planning |
The goal is not to predict AI costs perfectly. The goal is to give finance leaders a planning model that can adapt as AI usage, hiring plans and productivity assumptions change.
Visibility vs. Control: The Core Shift in AI Spend Management
Many organizations are still in the visibility stage of AI spend management. They are trying to understand how much they are spending, who is using AI and where costs are coming from.
Visibility is essential, but it is only the first step.
The next stage is control. Control means finance and business leaders can forecast AI spend, allocate it, govern it, optimize it and connect it to outcomes. It also means they can incorporate AI usage into workforce planning, so labor cost models reflect the full cost and value of AI-enabled work.
This is where write-back becomes critical. By adding write-back capabilities to Power BI, the Acterys platform enables organizations to move beyond reporting on AI token spend and begin forecasting, allocating and controlling it within the same Power BI environment they already use. Finance teams can work from current usage data, model how consumption may change and adjust plans before costs scale.
Zack describes the opportunity as “moving from usage and tracking to optimization.” That is the central shift finance leaders need to make: from knowing AI spend happened to planning, forecasting and managing it before it becomes a budget problem.
Visibility – Reactive | Control – Proactive |
Shows what was spent | Forecasts what will be spent |
Reports usage after the fact | Plans usage before it scales |
Identifies high-consuming teams | Determines whether usage is creating workforce productivity |
Tracks tokens, tools and cloud costs | Allocates and optimizes spend by workflow, team, role or outcome |
Helps explain the bill | Helps manage the business driver |
Shows AI cost separately | Connects AI cost to headcount, labor plans and productivity |
This is the heart of AI spend management. Finance teams should not be left explaining AI costs after they appear. They should be part of the planning process that determines where AI investment goes, how usage is governed, how workforce plans change and how value is measured.
Conclusion: AI Spend is a Workforce and Business Driver
AI is becoming part of how work gets done. That means AI costs will increasingly show up in workforce planning, IT budgets, operational forecasts, product development, sales planning, customer support and financial reporting.
The organizations that treat AI spend as an unmanaged technology expense will struggle to explain rising costs. The organizations that treat AI spend as a workforce and business driver will be better equipped to forecast usage, govern investment and optimize value.
Finance leaders have a critical role to play in this shift.
They can help the business move from AI experimentation to AI economics. They can connect usage to outcomes. They can bring structure to token forecasting, workforce planning, chargeback, ROI measurement and scenario planning. And they can help ensure AI adoption grows in a way that is financially sustainable.
Zack describes the broader impact as “downstream,” noting that AI costs may start as large technology expenses but ultimately roll down to individuals, teams and how work gets done. “There’s always a bottom-up. It always impacts individuals, and then it rolls up,” he said.
The next era of workforce planning will not only account for people, compensation and capacity. It will also account for the AI resources employees use to get work done and the value those resources create.
AI spend should not be managed like a surprise IT bill. It should be planned and controlled like any other business driver.
