Program report · filed November 13, 2026

AI in SalesMeasured report

Tokens Are 10-25% of AI Spend: Forrester Says Cost Taxonomy Is the Missing Link

Tokens and API costs are only 10-25% of total AI spend, the Tokenomics Foundation argued in September. Forrester says AI ROI needs a cost taxonomy first.

By Nathan Brooks4 min read733 words

Program notes

  1. The Tokenomics Foundation argued in September 2024-era commentary that token and API costs are only 10-25% of total AI spend.
  2. The Tokenomics Foundation launched under five months ago.
  3. Forrester's example attributes a $2M Product A lift and $1M Product B lift to two shared agents with no clean cost split.
  4. TBM-style layered cost models require agreed allocation formulas and ongoing governance, not plug-and-play setup.
Tokenomics Needs A Taxonomy
PlateTokenomics Needs A Taxonomy — lukasbieri / pixabay

Tokens and API costs account for only 10% to 25% of total AI spend, according to an argument the Tokenomics Foundation made this September. That number, if it holds, reframes the entire budget conversation. The foundation launched under five months ago to build a shared discipline for managing token costs. It has since drifted toward something larger: a methodology for connecting the total cost of AI to business value.

The gap is structural, not cosmetic. You can measure AI costs without a taxonomy, but you cannot connect those costs to business value without one.

Why does AI ROI need a defensible cost basis?

The ROI formula is straightforward. Value sits in the numerator. Cost sits in the denominator. The numerator is where most organizations already struggle — they must decide whether an AI investment saves employee time, increases revenue, improves customer outcomes, or reduces risk. The Forrester AI Value Matrix offers a starting point for that determination.

The denominator is the harder half. Forrester's own analyst, writing under the byline of this argument, offers a concrete scenario:

  • One agent improves evaluation and routing of sales leads for Product A and Product B.
  • A second agent speeds up sales meeting prep, creating more customer-facing time for both products.
  • The combined effect is a $2 million sales lift for Product A and a $1 million lift for Product B.

Neither agent belongs cleanly to either product. So how should the organization attribute each agent's impact across the two products for its ROI calculations? The question has no obvious answer, which is precisely the problem.

The proposed model has three layers:

  1. Categorize AI costs — LLM costs, cloud services, on-prem capitalization, labor, and the rest.
  2. Attribute those costs to defined AI capabilities, so each agent carries a total cost built from its actual mix of LLM, cloud and on-prem spend.
  3. Allocate each agent's costs to business outcomes — in this case, splitting them between Product A and Product B, including hard-to-tag items like on-prem storage and non-timesheet labor.

Only then do the cost and benefit sides of the ROI calculation align.

What can AI leaders borrow from TBM?

Technology leaders have solved an adjacent problem before. Technology Business Management (TBM) does not connect a server or cloud instance directly to business value. It uses intermediate layers to translate technical resources and costs into applications, services, products, and ultimately the business units consuming them.

AI needs the same connective tissue. Whether it arrives as a TBM extension, a Tokenomics Foundation taxonomy, or a standard from some other body remains an open question, and the source does not claim to know.

The caution from TBM is worth more than the template, though. These models are powerful, but they are difficult to implement and maintain, because they are not plug and play. Each layer forces assumptions about how to split blocks of cost across stakeholders, and maintaining those assumptions consumes real resources. In the two-agent scenario, no formula for splitting costs between Product A and Product B is self-evident — it requires an agreed indicative metric, such as headcount of salespeople or units sold, plus ownership and governance to earn stakeholder confidence in the output.

What can organizations do before a standard exists?

The advice is to stop waiting. Build your own taxonomy now and adapt it as a broader standard emerges. The suggested sequence:

  • Start with the outcome. Define the benefit and the unit it will be measured in.
  • Map AI capabilities to outcomes. Identify contributing agents, define their capabilities, and set allocation drivers where capabilities are shared.
  • Build the total cost of each capability. Define spend categories and attribute them to agents. Observability tools may do this cleanly; labor will not, so stakeholders must agree on the assumptions driving attribution.
  • Govern the assumptions. Document allocation methodologies, assign owners, and secure stakeholder agreement.

Run one use case first to demonstrate the method and build buy-in, then expand across functions and up the org chart. And do not chase perfect attribution on day one — the goal is enough structure between technical consumption and business outcomes that leaders can understand, challenge, and trust the economics.

With token costs at a fraction of true AI spend, whoever owns the attribution model will own the AI budget conversation — and the Tokenomics Foundation has given itself the task of making that model a standard.

via forrester.com (Source)

Filed under

Share this article:

More from Nathan Brooks

Nathan Brooks

Show full bio

News editor covering companies, deals and strategy in Sales technique and methodology, sales enablement at Quota Brief.

141 articles

Also rated

  1. 1% of Users Drive 75% of AI Costs — and Most SaaS Pricing Can't Handle It
  2. Salesforce caps employee AI spend at $1,000 a month after token bills spike
  3. Benioff Rejects "SaaSpocalypse" Talk as Salesforce Positions AI as Software Buyer
  4. Forrester: AI Won't Save RevOps — It Will Redefine It
  5. AI Captures Growing Slice Of Sales And Marketing Startup Funding

« Previous articleNext article »