AI Adoption Is Scaling Faster Than AI Cost Discipline

For the past few years, the enterprise AI conversation has focused on adoption: Which companies are using AI? How many employees are using it? And how quickly can organizations move from experimentation to deployment?

That conversation is changing.

As AI moves from copilots and chatbots to autonomous agents and multi-agent workflows, a less visible challenge is emerging: the ability to understand and manage the economics of AI at scale.

KPMG’s Q2 2026 U.S. AI Quarterly Pulse Survey found that 53% of organizations are using AI agents, while 18% are orchestrating multiple AI agents across workflows. Yet only 26% have full, real-time visibility into the cost of running AI, and 35% of leaders cite AI cost management and economic literacy as a barrier.

The message is simple: AI adoption is accelerating, but financial visibility is struggling to keep pace.

From copilots to AI agents

The first wave of enterprise AI was relatively straightforward.

A company deployed a chatbot or copilot, employees used it for specific tasks, and the value proposition was usually measured through productivity gains.

Agentic AI changes that model.

An AI agent can interpret a task, retrieve information, call software tools, interact with APIs, make multiple model calls and potentially hand work to another agent. Instead of a single interaction, a business process can become a chain of AI-driven decisions.

That makes AI more powerful but also makes its economics more complicated.

KPMG's data captures this transition. While the share of organizations using AI agents remained broadly stable at 53%, the share orchestrating multiple agents across workflows doubled from 9% to 18% in one quarter.

This is an important shift.

Companies are moving from asking:

“Where can we use AI?”

to:

“How can AI coordinate an entire workflow?”

And once AI becomes part of the workflow itself, understanding its cost becomes considerably more important.

The cost visibility gap

The numbers reveal an interesting disconnect.

53% of organizations are using AI agents.

But only 26% have full, real-time visibility into AI operating costs.

Meanwhile, organizations are already building basic governance infrastructure. KPMG found that 66% have AI monitoring dashboards and 61% have approval processes in place. Yet only 36% have direct token or usage controls.

In other words, many organizations can see that AI is being used. Fewer can see precisely how much that usage costs as it happens.

That distinction matters because AI introduces a different cost structure from traditional enterprise software.

AI has a different unit economics problem

Traditional software often has relatively predictable economics:

Users × subscription price = software cost

AI can be considerably more variable.

Costs can depend on:

• Input and output tokens

• Model selection

• Number of model calls

• Inference requirements

• API and tool usage

• Retrieval and data processing

• Agent iterations

• The number of agents involved in a workflow

As a result, two workflows that appear similar from the outside can have very different underlying economics.

A simple customer-service interaction might require one model call.

A more sophisticated agentic workflow could involve several model calls, database queries, external APIs and additional validation steps before producing an answer.

The relevant metric therefore isn't necessarily:

“What does one AI response cost?”

It is:

“What does it cost to complete the entire business outcome?”

That is a much more useful way to think about AI economics.

Tokens are only part of the equation

It is also tempting to reduce AI economics to token pricing.

That is increasingly inadequate.

McKinsey's recent analysis of agentic workflows highlights the importance of looking at total cost of ownership, including both fixed and variable costs. In some banking customer-service workflows, token costs can account for only 20–25% of variable agent costs, with the remainder coming from other components of the workflow.

This changes how organizations should evaluate AI investments.

A cheaper model does not automatically produce a cheaper workflow.

Likewise, a more capable model does not automatically produce better economics.

The right question is whether the additional capability produces enough incremental business value to justify the additional cost.

That brings AI much closer to a traditional business strategy problem:

What is the return on the incremental dollar spent?

From AI adoption to AI unit economics

As agentic systems scale, businesses will need to develop a more granular view of AI performance.

Instead of measuring only adoption, organizations could start tracking metrics such as:

Cost per task
How much does it cost to complete an AI-driven task?

Cost per successful outcome
How much does the business spend when the AI actually achieves the intended result?

Cost-to-value ratio
How much measurable business value is generated for every dollar spent?

AI cost as a percentage of revenue
Is AI-related expenditure growing faster than the revenue or productivity gains it creates?

These metrics shift the conversation from AI activity to AI economics.

The objective is not necessarily to minimize AI spending.

It is to understand where additional AI spending creates enough value to justify itself.

Visibility may be connected to ROI

There is an important signal in KPMG's global research as well.

Its Q2 2026 Global AI Pulse surveyed 2,145 senior leaders across 20 countries and jurisdictions. Organizations with full visibility into AI operating costs were five times more likely to report established ROI than those without full visibility 15% versus 3%.

This does not prove that cost visibility causes higher ROI. Other factors, such as governance, leadership accountability and organizational maturity, could also influence the result.

But the relationship is difficult to ignore.

You cannot effectively optimize an investment that you cannot measure.

And as AI becomes embedded into more workflows, the ability to connect usage, cost and business outcomes will become increasingly important.

The next phase of enterprise AI

The first phase of enterprise AI was about proving that the technology worked.

The second phase is about deploying it at scale.

The next phase may be about deciding where AI actually deserves to scale.

That requires a different set of questions.

Not simply:

Can AI perform this task?

But:

Can AI perform this task reliably, at an acceptable cost, and with enough business value to justify scaling it?

The companies that answer those questions well may gain an advantage that has little to do with simply having access to the latest model.

They will know which workflows to automate, which models to use, when to introduce human oversight, and where additional AI capability creates diminishing returns.

In that sense, the next AI advantage may not be more AI.

It may be better AI economics.

Usage → Cost → Outcome → ROI.

That is the framework that could determine whether enterprise AI moves from impressive experimentation to sustainable business value.

At Unpromptd, we do not just open doors. We help you own the room. --- If you are a global platform looking to enter or expand in Asia, reach out to us at contact@unpromptd.com. Let’s build the future together.