Anthropic gave leaders a blunt signal about planning for LLM charges when it introduced blended pricing for enterprise users of Claude recently. Enterprises with over 150 seatholders will now pay a blanket per-seat charge ($20 per seat at the moment), plus a variable charge based on API usage and which Claude model they’re using.

When you think about pricing strategy, this isn’t a surprise. Forget about the escalating cost of doing business for LLMs like Anthropic and OpenAI driving price increases. This is indeed leading to pressure to raise prices, but customer willingness to pay is the ultimate determinant of price, and until now, you’ve been benefiting from dirt-cheap LLM charges. For example, Claude at $200/month (Anthropic’s previous premium-tier fixed pricing) equates to $1.25 per hour. Meanwhile, you’re paying your cheapest labor pool about $40 per hour. And if you’re using Claude for more skilled work, the gap between Claude’s pricing and your cost of labor is even wider. So as enterprises mature in their ability to get value from LLMs, they should anticipate substantial price increases from LLMs looking to capture a greater share of the value their models are providing.

So I Need To Find Even More Money?

Leaders’ grasp on the (growing) total cost of AI is evolving faster than their ability to tell the value story, and the increasing cost of LLMs is exacerbating this dilemma. This is leaving firms in funding purgatory, as CIOs can’t free up enough budget to both implement and scale the business’s AI-powered workflows. Meanwhile, CEOs and boards insist on more and faster AI-powered business transformation. The result is frustration and a slow pace of scaling AI deployments.

What funding model allows AI to scale fast enough for the C-level while not blowing the CIO’s budget? According to an executive at a multibillion-dollar manufacturer, the business must “own AI success while IT supports it.” Therefore, the model should touch the business’s budget as well as the CIO’s.

Here’s one approach:

  • Include the direct operating expense in the CIO’s budget.
  • Charge out the direct operating costs, including inferencing (tokens), content, and AI operating staff, to the business.
  • Keep the build and training costs in IT.

Forrester’s A Necessary Primer On Workforce Planning report establishes four categories of workforce resources: build (upskill), buy (hire full-time employees), borrow (contingent resources), and bot (AI). IT owns the bot pool of resources, and the business is the recipient of the bots’ services. So the interests, risks, and budgets of both parties are aligned in the chargeback highlighted above.

For organizations already using a chargeback structure for IT spend like many federal, state, and local government agencies, this isn’t much of a deviation other than it being an “above the line” chargeback, i.e., it is included in the accountability view of the CIO and business budgets. One enhancement over time could be to derive a rate-per-hour metric for the agent to check the charged spend against market rates for outsourced resources.

So What Should Tech Execs Do?

If you decide that a chargeback of operating AI spend is the right model for you, here are the steps that IT, finance, procurement, and the business need to take to get started:

  • Know your total cost of AI ownership. This includes not just developing visibility in your LLM billings to accurately attribute usage but also understanding non-LLM charges like cloud egress and storage charges for your proprietary data, the cost of internal development resources, etc. The shift toward variable LLM pricing makes knowing AI total cost of ownership even more critical, as it signals a shift in the value discussion from resource costs to process delivery cost.
  • Agree on the chargeback methodology and revisit semiannually to address shifts in costing and/or usage. The business can use this mechanism to hold IT accountable for delivering agents efficiently as costs increase.
  • Allocate funding appropriately based on direct benefit to the operation.
  • Establish a standard budgeting template for building, deploying, and scaling agentic workflows.
  • Implement processes for: 1) assumptions, entry, and sign-off for budget and forecast and 2) confirming and recording chargeback amounts for actuals.

Want to talk through these steps in more detail? Book a guidance session with me or send me an email at gzorella@forrester.com.