AI budgets in companies: token governance and ROI

From tokenmaxxing to token rationing: what's happening

In June 2026, TechCrunch documented a rapidly spreading phenomenon. Companies are running out of their AI budgets much sooner than expected. The reason? Employees are using generative models for low-value tasks: rephrasing emails, summarizing short documents, generating a few lines of text. Every API call consumes tokens. Every token has a cost. Multiplied by hundreds of users, the result is an out-of-control expense.

This scenario has a specific name: tokenmaxxing . In practice, the massive and often unaware use of language model capabilities. However, the euphoric phase seems to be over. We are entering the opposite era: the token rationing , that is, the conscious rationalization of AI consumption.

Therefore, the question that marketing and digital managers should ask themselves is no longer «which AI tool to adopt». The question is: “How do we govern AI consumption to maximize ROI?”

The numbers reshaping CFO priorities

The AI-as-a-service models market has grown exponentially. According to Gartner , by 2027 over 70% of mid-market companies will have integrated at least one AI tool into their core processes. However, the same research highlights that less than 30% of these companies have a formal framework for monitoring usage costs.

Furthermore, token-based pricing models create variability that's hard to budget for. A company using GPT-4o or Claude 3.5 for content marketing might see monthly costs fluctuate by 40-60% depending on request volume. As a result, CFOs are starting to ask marketing and digital teams to account for AI consumption with the same precision they use for advertising spend.

In Italy, the context is even more complex. SMEs often adopt AI tools through SaaS subscriptions that include usage quotas. When these quotas are exhausted, unexpected additional costs kick in. Thus, governance isn't just a topic for large enterprises: it also concerns those managing teams of 10-50 people.

Strategic reading: three dynamics to watch

Analyzing the phenomenon from a consulting perspective, three structural dynamics emerge that deserve attention.

First of all, democratization generates dispersion. When an AI tool is accessible to all employees, consumption is distributed chaotically. There is no prioritization of high-value use cases. Each user uses the tool according to their own logic, often for tasks that wouldn't require AI.

Secondly, the lack of output metrics nullifies ROI. Many companies measure token consumption (input), but they don't measure the value output generated. For example, how many work hours were saved? How many content pieces produced generated organic traffic? Without these metrics, the AI budget remains an opaque cost item.

Finally, competitive pressure pushes for hasty adoption. Many SMEs integrated AI tools in 2025 without a governance strategy. Today they are dealing with the consequences. However, this doesn't mean AI is a mistake: it means a more mature approach is needed.

The still open construction site: AI governance in Italian SMEs

Building an AI governance model doesn't necessarily require complex structures. We at SHM Studio we work with companies of different sizes and we observe that the most effective frameworks share some common elements.

Mapping use cases by value level. Not all AI uses are equivalent. Generating a campaign of SEO copywriting optimized has measurable value. Rewriting an internal email has marginal value. Therefore, the first step is to classify use cases into three categories: high value (strategic), medium value (operational), low value (to be limited or automated with cheaper models).

Allocating budgets by team or function. Similarly to how budgets are managed for google ads campaigns or for LinkedIn campaigns , AI consumption can be allocated by department. This creates accountability and encourages conscious use.

Choosing the right model for each task. Using a latest-generation model for every request is like using an SUV to go 500 meters. Lighter, more economical models exist that are suitable for simple tasks. Therefore, effective governance also includes a policy for selecting the model based on the task's complexity.

Practical implications for marketing and digital teams

For marketing managers, the topic of AI governance translates into concrete decisions. Here are the priority areas for intervention.

• Current consumption audit: before rationalizing, it's necessary to understand where the tokens are going. Many providers offer detailed usage dashboards. Using them is the starting point.
• AI integration in strategies SEO and Digital marketing :high-ROI use cases include optimized content generation, semantic SERP analysis, and campaign personalization. These uses justify a significant token investment.
• Team training: employees must understand that every prompt has a cost. It's not about limiting creativity, but about directing it towards measurable goals.
• Continuous monitoring: AI consumption is not static. It grows with adoption. Therefore, the budget must be reviewed at least quarterly, not annually.

Furthermore, it's worth considering the integration of tools for AI governance in their technological stacks. Some platforms allow setting consumption thresholds, automatic alerts, and differentiated access policies by role.

Towards 2027: whoever governs AI will govern competitive advantage

The topic of AI governance is not going away. On the contrary, it will grow in importance as models become more powerful and per-token costs are spread over ever-increasing volumes. According to Harvard Business Review , organizations that invest in AI governance today are building a competitive capability that is difficult to replicate in the short term.

For Italian SMEs, this means acting now. Don't wait for the budget to run out to start thinking about how to allocate it. Don't wait for a CFO to ask for reporting to build output metrics. Therefore, AI governance is not a constraint: it's an enabler.

Teams that can balance freedom of use and cost control will get the most out of AI. Those who simply cut off access to tools risk losing the real benefits of the technology. The balance point is found in strategy, not in blocking.

To learn more about how to structure a sustainable AI strategy for your company context, the team at SHM Studio is available for dedicated consulting. You can contact us through the contact page or explore our digital services . Further insights are available in our Blog .