The CFO’s Guide to What AI Actually Costs Your Business

If you’ve been asked to approve an AI budget recently — or you’re trying to build one from scratch — you’ve probably noticed that most of the information available focuses on one thing: token costs. How much does it cost per thousand words processed? What does a model call run?

Those numbers are real. But for a company with 100 or more employees running AI across multiple departments, token costs are not your primary budget line. They’re not even close.

The real cost of AI operations at a mid-sized business is a layered number — subscriptions, seat licenses, API usage, implementation, ongoing management, and governance — and most organizations are only seeing one or two of those layers clearly when they build their first AI budget. The rest show up later, usually as surprises.

This post lays out all of the layers, gives you realistic ranges for what each one costs, and walks through what a fully loaded AI budget looks like for a company operating at meaningful scale. All pricing ranges referenced here are directional based on publicly available information and typical contract structures as of mid-2025; actual costs will vary by vendor tier, contract terms, and negotiated rates, and should be validated before finalizing any budget.

The question CFOs need to be asking is “What does AI operations cost in total, and what does that buy us?” 

First: Are Those Horror Stories Even About You?

Probably not — and it’s worth saying that plainly, because the gap between the scary headlines and everyday business AI use is substantial.

The organizations that blow through massive AI budgets unexpectedly are almost always doing one of a few specific things: running AI-assisted coding tools across a large engineering team where the tool is making model calls on every keystroke, processing millions of documents through automated pipelines, or running AI agents that are making dozens of model calls per task without any monitoring in place.

If your business is using AI to draft communications, summarize documents, answer customer questions, or support your team’s daily work, you are operating in a completely different cost environment. The mechanics are the same, but the scale is not. Understanding that distinction will save you from making either of two expensive mistakes: spending recklessly because you don’t understand the cost structure, or avoiding AI entirely because you’ve conflated your situation with someone else’s.

Can AI costs spiral? Yes. What you need to know is “Under what circumstances does that happen, and is my business in that situation?”

The Four Cost Layers of AI Operations

Think of your AI cost structure as four distinct layers, each with different drivers, different predictability, and different levers for control.

Layer 1: Platform Subscriptions and Seat Licenses

This is the largest and most predictable cost layer for most businesses. Every AI-powered platform your teams use — whether or not you think of it as an “AI tool” — carries a per-seat or per-tier subscription cost. For a company with 150 employees running a full operational stack, that means:

• Work management platform: Seat-based pricing typically runs $10–$25 per user per month at business tiers, scaling to $20–$40+ at enterprise tiers with advanced automation and AI features. For 150 seats: $1,500–$6,000/month.
• AI assistant platform (e.g., Claude, Microsoft Copilot, or similar): Business and team plans typically run $20–$40 per user per month. For 150 seats: $3,000–$6,000/month. Not every employee may need a seat — licensing for active users only can reduce this meaningfully.
• Customer service platform: Agent-seat pricing for customer service teams typically runs $15–$80 per agent per month depending on tier and AI feature inclusion. For a team of 15–20 agents: $225–$1,600/month.
• Accounting and ERP system: Ranges widely by platform and company size — from $300–$1,500/month for mid-market solutions to significantly higher for enterprise ERP with AI modules.
• Order management and fulfillment platform: $200–$2,000/month depending on order volume and integration complexity.
• Marketing automation platform: $500–$3,000/month for mid-sized operations with multi-channel automation and AI content features.

Subscription and seat license total for a 150-person company running a full stack: roughly $6,000–$20,000 per month, before any AI-specific API costs are added.

Subscription costs are your most predictable AI budget line. They’re also where scope creep lives — unused seats, underutilized tiers, and redundant tools that nobody has reviewed in 18 months. A periodic stack audit pays for itself quickly.

Layer 2: API and Token Costs

This is the layer most people think of when they hear “AI costs,” and it’s genuinely important — but it’s usually not the dominant number for a business operating primarily through packaged platforms.

Here’s the distinction that matters: when your teams use AI features built into a subscribed platform, token costs are absorbed by the vendor and included in your subscription. You don’t see them. When your business builds custom workflows, automations, or agents that call AI models directly through an API, you pay per token on top of your subscriptions.

A token is roughly three-quarters of a word. Every prompt sent to an AI model and every response generated back is measured in tokens and billed accordingly. Rates vary significantly by model tier — lighter models run under $0.001 per 1,000 tokens; the most capable models run $0.015 or more per 1,000 tokens. Most business workflow applications land somewhere in the $0.003–$0.008 per 1,000 token range on a blended basis.

For a company running meaningful custom automation — AI-assisted content workflows, custom integrations between systems, intelligent routing and classification across departments — direct API costs typically run $500–$5,000 per month depending on workflow volume and design efficiency. Companies running poorly designed or unmonitored automations can spend significantly more without realizing it.

Layer 3: AI Agent Costs

Agents deserve their own cost category because they behave fundamentally differently from standard AI interactions, and the cost implications are proportionally different.

An AI agent doesn’t make a single model call when it completes a task. It makes many — reasoning through a problem in steps, checking its work, calling external tools, handling errors, and generating outputs. What looks like one completed task to the end user might involve 15, 30, or 50 model calls behind the scenes. For complex, multi-step agent workflows, token consumption per task can run 10,000–100,000 tokens or more.

At scale, agent costs are material and need to be tracked as their own budget line. A company running agents across customer service routing, fulfillment exception handling, marketing content generation, and financial data processing could realistically be spending $1,000–$10,000 per month in agent-related model costs alone, depending on task volume and workflow complexity.

The critical governance question for agents is not just what they cost, but whether you have visibility into what they’re doing and why. An agent making unnecessary calls, retrying failed steps repeatedly, or processing more context than a task requires will run up costs silently. Monitoring and cost alerting are not optional at this scale — they’re budget hygiene.

Agent costs are the most volatile line in an AI budget. They’re also where the highest-value work happens. The answer is not to avoid agents — it’s to govern them properly from day one.

Layer 4: Implementation, Management, and Governance

This is the cost layer that is most consistently underestimated — and the one that most directly determines whether the other three layers deliver any return at all.

AI tools and platforms don’t run themselves. Workflows need to be designed and built. Automations need to be tested, monitored, and maintained when underlying systems change. Agents need to be evaluated for performance and cost efficiency. Staff need training. Governance policies need to be established and enforced. Data quality needs to be maintained to keep AI outputs reliable.

For a mid-sized company building and running AI across multiple departments, this work is either being done by internal staff — which means it’s consuming salary that isn’t showing up in the AI budget line — or it’s being managed by an external implementation partner. Either way, it’s a real cost that belongs in the budget.

Rough ranges for this layer:

• Internal AI operations ownership (dedicated or partial FTE): $3,000–$10,000/month in fully loaded salary allocation, depending on scope and seniority.
• External implementation and ongoing management partner: $2,500–$15,000/month depending on the complexity of the stack, the number of active workflows, and the scope of ongoing optimization work.
• One-time implementation costs for new systems or major workflow builds: $5,000–$50,000+ depending on complexity. These are capital expenditures, not operating costs, but they belong in the full budget picture.

The temptation is to minimize this layer to make the AI budget look leaner. That’s a false economy. Underinvesting in implementation and management is the most reliable way to ensure that the platforms and subscriptions in layers one through three underperform or fail outright.

What It Actually Adds Up To

Here’s what a realistic fully loaded AI operations budget looks like for a 150-person company running AI actively across multiple departments. These are directional ranges — your actual numbers will depend on your specific platform choices, contract terms, workflow complexity, and how aggressively you’re automating.

For a company at the lower end of this range — using AI tools selectively, with modest automation and a lean management approach — $10,000–$15,000/month is a realistic all-in budget. For a company that is actively automating across departments, running agents, and investing in implementation quality, $30,000–$50,000/month is not unusual. Companies that have not yet audited their stack for redundancy and efficiency often discover they’re spending at the high end without getting high-end results.

Building a Budget You Can Defend

A defensible AI budget has four properties: it’s complete, it’s attributed, it’s monitored, and it’s reviewed regularly.

Complete

It accounts for all four cost layers — not just the subscriptions that show up on the credit card statement. If implementation labor is being absorbed into a department budget without attribution, that’s an AI cost that’s invisible in your AI budget. Bring it into the light.

Attributed

Every major cost line should be attributable to a workflow, a team, or a business outcome. “AI costs” is not a useful budget category. “Customer service AI routing — $1,200/month, reducing average handle time by 18%” is a budget line that can be evaluated and defended.

Monitored

API costs, agent call volumes, and subscription utilization should be tracked actively, not discovered at month end. Most platforms provide usage dashboards. Most API providers support spending alerts and caps. Use them. Build monitoring into your AI operations from the start, not as an afterthought when a surprise bill arrives.

Reviewed Regularly

AI platform pricing changes. Vendor contracts come up for renewal. Workflows that made sense six months ago may be inefficient today. A quarterly AI budget review — stack audit, utilization check, ROI attribution, cost optimization opportunities — should be a standing item on the finance calendar, not a reactive exercise.

The Question Underneath the Budget

Every CFO who has looked at a technology budget seriously knows that the number on the page is only half the conversation. The other half is: what are we getting for it?

AI is not exempt from that question. In fact, it’s more important to ask it here than almost anywhere else in the technology stack — because AI spend is growing fast, the value attribution is less mature than traditional software ROI, and the vendors selling AI tools are not incentivized to help you answer it honestly.

The businesses that get the most out of AI investment are not necessarily the ones spending the most. They’re the ones who designed their AI operations deliberately — with clear use cases, proper implementation, cost visibility, and a framework for measuring what the investment is actually returning.

That’s a harder thing to build than a subscription. But it’s the thing that makes the subscription worth having.

→ Book a WHIM Strategy Call to walk through your AI cost structure and build a budget you can defend

About WHIM Innovation

WHIM Innovation helps organizations harness the practical power of AI, automation, and custom software to work smarter and scale faster. We combine deep technical expertise with real-world business insight to build tools that simplify operations, enhance decision-making, and unlock new capacity across teams. From AI strategy and workflow design to custom monday.com apps and fully integrated solutions, we partner closely with clients to create systems that are efficient, intuitive, and built for long-term success.