Finance
How Enterprise AI Pricing Outgrew its Token Phase
Usage-based pricing ties AI spend to outcomes, and fungibility keeps the model durable as priorities change.
Julie Colwell
Principal Strategist
²ÝÝ®ÊÓƵ
As AI capabilities expand, vendors are moving away from flat subscriptions toward , outcome-based, and hybrid pricing structures. For CFOs, this variability can feel like a step backward in budget certainty.
But this isn't the first time a technology shift has forced a transition to new cost models. The move from on-premise to cloud computing initially made budgeting harder, until usage-based cloud spend became standard. AI pricing is on a similar trajectory. As tools for tracking and predicting spend catch up, it will become easier not only to forecast, but to measure value.
Report
AI Agents Are Here¡ªBut Don¡¯t Call Them Boss
The Rise and Fall of Tokenmaxxing
We've come a long way since AI pricing started. "Tokenmaxxing," treating token consumption as a proxy for AI adoption, had a rapid rise. Meta reportedly ran an internal leaderboard nicknamed that ranked roughly 85,000 employees by token usage, with the top user said to have burned through 281 billion tokens in a single month.
The fall came just as fast. Within days of each other, after blowing through its entire annual AI budget in the first few months of the year, and Microsoft told engineers internally that tokenmaxxing is no longer what they are optimizing for, setting per-employee token budgets and defaulting to a cheaper model. Meanwhile, IBM has argued ¡ªoptimizing for business outcomes instead of usage volume¡ªis the more durable framework.
¡°Valuemaxxing¡±¡ªoptimizing for business outcomes instead of usage volume¡ªis the more durable framework.
Most finance leaders have watched some version of that arc play out inside their own organizations. It speaks to the challenge of finding a long-term pricing model when AI¡¯s actual business value is evolving.
Increasingly, the answer finance teams are landing on is usage-based, or consumption-based, pricing: paying for what's consumed rather than a flat fee disconnected from results.
What Gets Metered Now
With usage or consumption-based pricing, AI agents are , not when a query runs or a token gets consumed. In this model, completed tasks are easily quantifiable. For example:
- Screening a r¨¦sum¨¦ against a requisition
- Sourcing candidates from an existing pool
- Redlining a contract against a playbook
This pricing model produces a materially different budgeting conversation than counting tokens. A finance leader can forecast next year's spend off last year's requisition or contract volume, because the unit being priced is a business action, not an artifact of model architecture.
Why Fungibility Matters
As vendors ship specialized agents for different functions, the easy path is to sell a separate metered currency for each¡ªone pool for recruiting agents, another for finance agents, another for legal. Siloed credits look precise but, in practice, recreate habits of legacy software licensing:
- Budget locked to a specific use case months before anyone knows which use case will actually deliver value
- Unused allotments sitting idle because the org shifted priorities and the spend didn't move with it
Enterprise AI strategies don't hold still long enough for that kind of rigidity to work. A finance organization that commits budget to a contract-review agent in January may find by June that the higher-value use case is in workforce planning instead. If the spend committed to one can't be redirected to the other without a new procurement cycle, the pricing model is fighting the reality of how AI adoption unfolds inside a company.
Fungible spend¡ªone pool that applies to whatever agent or capability is delivering the most value at a given moment¡ªis what makes usage pricing work in practice.? With this approach, CFOs are buying capacity and deciding later where it goes, better mapping to how AI adoption behaves inside a large organization.
With fungible spend, CFOs are buying capacity and deciding later where it goes.
Who's Accountable When The Pool Is Shared
While fungible spend better aligns to AI adoption, it begs an important question: If any team can draw from the same pool, who's accountable for what gets spent? Constraining the pool defeats the purpose. But, pairing fungibility with real-time visibility enables CFOs to see spend shift across the organization instead of discovering it at the end of a budget cycle.
What a Durable Model Looks Like
The pricing models that last through a technology's early years share a pattern:
- They track something real
- They're built assuming they'll need to change
- They give CFOs the visibility to manage variability rather than just absorb it
Token-based pricing struggled because usage and value unraveled as models improved. A rigid, siloed usage-credit model would struggle too¡ªjust more slowly¡ªby locking spend to assumptions about which AI use case matters most. As the market keeps changing, those use cases can and should change with it.
Seat-based pricing hasn¡¯t left the building. It is, of course, built on the assumption that a human was doing the work. And plenty of organizations, including our own customers, still lean on it for the predictability it offers. But as AI agents are trained to act, not just think,, tracking value at the level of an action or outcome is becoming table stakes. Gone are the days of measuring a tool¡¯s value by the number of logins.
Tracking value at the level of an action or outcome is becoming table stakes.
When the unit being priced is an action or an outcome, not a token or a seat, Finance isn't just paying a bill. It's paying for tangible value AI is returning to the business.
Report
AI Agents Are Here¡ªBut Don¡¯t Call Them Boss
A gap between effort and impact is the real ROI problem. Discover the solution.