PUBLISHED DATE: 2026-07-20 03:46:00

Enterprise AI: From Token Spend to Business Value | Publicis Sapient


Sapient Bodhi ranked #1 for deep research.
Explore Bodhi
3M lines of COBOL → clear specs in 8 weeks.
Get the details
Meet Sapient Sustain: context-aware AI for complex IT operations.
Explore Sustain

AI’s Free-Spending Era Is Over

The question is no longer how much you spend on AI, it’s whether it produces real business value.
July 20, 2026
  1. See where AI creates value
  2. Match the model to the work
  3. Give AI the context it needs

From tokenmaxxing to valuemaxxing

Generate AI Summary
Loading AI-generated summary...

For the last two years, enterprise AI strategy has followed a simple rule: buy access to the best model you can get and let teams experiment. Consumption became a proxy for progress, and the more tokens a team burned, the more serious its AI effort appeared. The industry even coined a name for it: tokenmaxxing.

That era is ending, and not because of high token costs alone. The enterprise AI conversation is maturing. Executives who had once encouraged unlimited experimentation are now trying to rein in costs. Leaders who once asked, “which model should we use?” are asking a harder question: why isn’t AI delivering the business value we expected?

In Publicis Sapient’s global survey of 1,550 enterprise AI decision-makers, more than 73 percent say that AI is now used regularly or in most processes, yet only 10 percent say it is core to how the business operates.

Tokenmaxxing’s biggest risks don’t show up on an invoice. Palantir’s Alex Karp told CNBC that “something has gone completely wrong” in how AI is sold, arguing that enterprises are paying for tokens while getting little value back and giving up control of their data along the way. Run sensitive data and proprietary IP through someone else’s frontier model without your own application layer around it, and that data and knowledge can end up in a system you don’t control. Tokenmaxxing gets the attention, but the costlier problem is losing your competitive edge to a model someone else owns.

The market is wrestling with cost, governance, ownership and control all at once. Businesses are paying premium prices for AI without the outcomes to justify them, and that reality has forced some enterprises to pump their brakes. In May 2026, Microsoft reportedly rolled back its Claude subscriptions, while Amazon directed employees to use AI only to solve real problems, rather than applying it to an ever-growing list of use cases.

The answer isn’t necessarily to spend less; it’s to get more from what you spend, with AI built specifically for your business. Our research found that 42 percent of leaders say their organization isn’t structured to capture the value AI can create. Asked what limits the technology’s success, they are twice as likely to blame the way their organization runs (22 percent) as the technology itself (11 percent). Capturing value comes down to a few disciplines: know where AI is earning its keep, match each task to the right model and stop paying twice for knowledge your business already has.

1. See where AI creates value

Before an organization can improve AI’s return, it needs to understand where the return is coming from (and where it isn’t). Without that insight, AI’s cost and impact tend to surface only when the invoice arrives, long after the decisions that drove them.

The first step is visibility. Leaders need to see who is using AI, how many tokens they’re consuming, where those tokens are going and what each of those workflows produces for the business. For example, this process may help a company discover that 40 percent of its token usage comes from a single workflow that employees rarely use, while a high-value workflow consumes only a fraction of the budget.

Sapient Bodhi, our agentic AI platform built for enterprises, equips teams with observability dashboards that help them:
With this visibility, leaders can do more than trim waste. They can set policy, direct investment to the workflows that move business metrics and govern AI spending before it becomes a runaway line item.

2. Match the model to the work

Not every AI task requires the most powerful model. The problem is that many organizations default to frontier models, even for routine work. Simple requests, such as classifying a document or answering a common question, get sent to the same expensive model used for complex reasoning tasks.

A more effective approach is to intelligently route work to the right model. Every request passes through an LLM gateway that evaluates the task and sends it to the appropriate model. A simple task may be handled by a small language model, while a more demanding workflow may require advanced reasoning capabilities.

This is the job of Bodhi's LLM Gateway. Every AI request runs through a single control point, which automatically determines the right model for the job, ensuring that routine requests go to lower-cost models. This approach reserves frontier models for the most complex work that requires stronger reasoning capabilities.

According to Rakesh Ravuri, Publicis Sapient’s CTO, if a user simply says, “hello,” there is no reason to invoke an expensive frontier model. Instead, the platform should:
That flexibility matters now more than it did a year ago. Open-weight models are now credible alternatives to frontier models for real agentic work, and the release of GLM-5.2 may be the clearest signal yet. When a capable open- or fine-tuned model can do the job at a fraction of a frontier model's cost, sending that work to the frontier model is a wasted expense. Enterprises should be free to choose the right model for the right task, with that choice governed centrally rather than coded into every integration.

3. Give AI the context it needs

Duplicated reasoning can become a significant source of token consumption. When an organization runs a multi-agent environment, those agents often perform overlapping discovery work as they execute their unique tasks. One agent searches for policies and business rules that already exist somewhere inside the organization. Then another agent repeats the process. Then another. The enterprise already possesses the knowledge, but the system pays to rediscover it every time work begins.

When leaders name what most limits their ability to scale AI, the top two answers are integration across systems (38 percent) and data fragmentation (36 percent). This means disconnected systems and scattered data that force AI to work without a shared view of the business.

Consistent, shared business context can fix this issue. When agents can draw from a shared understanding of enterprise systems, operating rules and prior decisions, they don’t duplicate one another’s work. Tokens are spent on judgment, coordination and execution rather than on reconstructing information the organization already knows.

Our enterprise context graph—which Sapient Bodhi, Sapient Slingshot and Sapient Sustain, our three AI platforms, share—attacks this side of the cost problem. It creates a persistent, business-aligned map of how the enterprise works: its systems, rules, workflows, relationships and prior decisions. Agents work from this shared context instead of rebuilding it from raw documents, prompts and fragmented systems every time.

Context is also what makes AI’s output relevant and worth trusting. An agent that understands the operating model produces answers the business can act on, and that’s the difference between AI that merely runs and AI that delivers value.

From tokenmaxxing to valuemaxxing

The economics of AI are changing quickly, and enterprise leaders’ priorities are changing with them. Leaders want to know whether AI is producing business value, whether costs stay predictable as adoption scales and whether the business keeps control of its data, IP and competitive edge.

Nearly two-thirds expect major scaling in the next 12 to 24 months, but just 15 percent say they are equipped for it today. The real work lies ahead. That requires visibility into usage, governance over model selection and a persistent layer of business context that keeps agents grounded in the enterprise reality.

Together, these controls turn AI token management from a budgeting exercise into an operational discipline. Platforms such as Sapient Bodhi are increasingly being designed around these principles, combining model governance, intelligent routing and persistent business context to help organizations get real business value from their AI investments.