Is Increased AI Use Actually Creating Value?
Updated: Sep 8
As enterprise AI adoption grows, procurement, finance, technology, and business teams face a harder question than simply understanding pricing:
How do you know whether increasing AI consumption is actually creating value?
Many AI services use consumption-based pricing, where customers pay according to usage through measures such as tokens, API calls, compute, credits, or agent activity. That makes usage visible. It does not make value visible.
An organization can consume more AI and create substantial value. It can also consume more AI without improving a single business outcome. This becomes even harder to evaluate when employees and workflows span multiple AI tools. That is where the AI Consumption Value Model (ACVM) comes in.
What Is Consumption-Based Pricing in AI?
Consumption-based pricing ties cost to how much an AI service is used rather than charging only a fixed monthly or annual amount. Common measures include:
Tokens processed
API calls or transactions
Compute time
Credits
Agent or automation activity
This offers flexibility, but it also creates variability. As adoption grows, organizations need to forecast consumption, monitor spend, and understand what is driving increases.
The problem is that higher consumption does not necessarily mean higher value.
Moving Beyond Cost per Token
Most AI cost-management conversations focus on questions such as:
How many tokens did we use?
Which team consumed the most credits?
Are we approaching our allowance?
How much did API spending increase?
Those are useful cost questions. They are not value questions. FinOps makes a similar distinction between resource-efficiency measures, such as cost per token, and business-unit measures, such as cost per transaction or customer. Technology Business Management likewise connects technology cost and consumption to business outcomes rather than evaluating spend in isolation. ACVM applies that thinking specifically to AI.
Introducing the AI Consumption Value Model
The model follows six connected layers:
Consumption → Activity → Unit of Work → Outcome → Value → Decision
1. Consumption: What Did We Use?
Track tokens, credits, API calls, agent runs, subscriptions, and other AI costs.
These measures explain what was consumed. They do not prove that the consumption was worthwhile.
2. Activity: What Was AI Helping Someone Do?
Connect consumption to activities such as researching, drafting, coding, reviewing, summarizing, troubleshooting, or processing documents.
This gives usage context.
3. Unit of Work: What Did the Organization Produce?
Identify a business-relevant unit such as:
Incident resolved
Proposal completed
Case reviewed
Invoice processed
Customer request answered
Instead of asking: What did 5 million tokens cost? ask, What did AI cost per successfully completed case? That is a much more useful unit.
4. Outcome: What Changed?
Compare the AI-supported workflow with the organization's own baseline.
Measures might include:
|
|
Using an organization's own baseline matters because research consistently shows that AI effects vary by task and worker.
For example, a study of 5,179 customer-support agents found an average productivity gain of 14%, but gains reached 34% for novice and lower-skilled workers while effects were much smaller for experienced workers. The lesson is not to apply 14% to every business case. It is that AI value depends on the work, user, and context.
5. Value: Why Does the Outcome Matter?
ACVM separates value into several forms:
Economic value: revenue, avoided expense, or cashable savings
Capacity value: productive time released for other work
Performance value: better quality, cycle time, reliability, experience, or risk outcomes
Strategic value: capabilities the organization could not previously perform effectively or economically
This distinction matters because one of the most common AI ROI shortcuts is to convert every hour saved directly into salary savings. That can be misleading.
Research involving 7,137 knowledge workers across 66 firms found that active AI users spent about two fewer hours per week on email and reduced after-hours work, but researchers did not find broader changes in the quantity or composition of tasks performed. So time saved should usually be treated as capacity created, not automatically as money saved.
A better path is:
Time Saved → Capacity Released → Capacity Redeployed → Outcome Produced → Value Realized
6. Decision: What Should We Do Next?
The purpose of measuring value is not to prove that AI is worthwhile. It is to decide what to do with the next dollar. For each AI-supported workflow, an organization should be able to choose whether to:
|
|
This is where marginal value becomes useful. Instead of asking only whether AI created value overall, ask:
What additional outcome did the next increase in AI spending produce?
If spending rises 20% while successful cases rise 35% and cost per case falls, increased consumption may be desirable. If spending rises while outcomes remain flat, the organization has a reason to investigate.
Measuring Value Across Multiple AI Tools
ACVM is designed to sit above individual vendors. A single workflow might involve ChatGPT, Claude, Rovo, Copilot, an internal model, and AI-enabled automation. Trying to attribute the final outcome to one platform can quickly become misleading. Instead, allocate relevant AI costs to the workflow or unit of work.
For example: AI cost per successfully resolved case can include all relevant AI consumption associated with that workflow. This makes vendor data useful without letting vendor-specific billing units define the organization's measure of value.
From Usage Reporting to Value Management
A traditional consumption report might say:
Useful, but incomplete. | ACVM adds the missing context:
|
Now leaders can evaluate whether higher consumption deserves continued investment.
Takeaways
Consumption-based pricing makes AI usage measurable. It does not make AI value measurable.
The AI Consumption Value Model connects the two: Consumption → Activity → Unit of Work → Outcome → Value → Decision. Instead of relying on generic productivity claims or treating token counts as evidence of success, organizations can establish their own baselines, define their own units of work, and evaluate whether AI consumption is improving outcomes that matter to them.
That changes the conversation from: “How do we keep AI consumption down?” to, “Where is additional AI consumption producing enough value to justify the next dollar?”
Sources
Brynjolfsson, E., Li, D., & Raymond, L. R. Generative AI at Work. National Bureau of Economic Research.
Dillon, E. W., Jaffe, S., Immorlica, N., & Stanton, C. T. Shifting Work Patterns with Generative AI. National Bureau of Economic Research.
FinOps Foundation. Unit Economics.
Technology Business Management Council. TBM Framework.
Comments