AI Costs on the rise: how Italian SMEs manage tokens

The context: from token euphoria to the reality of costs

In the 2023-2024 biennium, the adoption of generative artificial intelligence followed a logic of maximum acceleration. Companies experimented, integrated, and scaled. Furthermore, language model providers lowered prices to gain market share. The result was an environment where the cost per token seemed like a secondary issue.

However, the situation has changed radically. By 2025, many organizations discovered that their AI bills had grown exponentially. According to a recent analysis by TechCrunch , the industry experienced a real breaking point. The phrase circulating among technical teams is emblematic: “The whole conversation shifted from tokenmaxxing and ‘go fast’ to ‘we need guardrails, how do we control this?'” .

So, 2026 is the year when the economic sustainability of AI became a strategic priority. Not just for large companies. Also for Italian SMEs that have integrated LLM-based tools into their operational processes.

The numbers that matter: how much does AI really cost for an SME

Estimating the cost of AI for a small or medium-sized business isn't straightforward. In fact, the expense is spread across multiple items: platform subscriptions, API costs per token consumed, development hours for integration, and prompt maintenance. Therefore, the actual budget is often higher than planned.

Some market estimates provide a useful overview. According to research by Gartner , by 2027 over 40% of organizations adopting generative AI will exceed their initially allocated budgets. Consequently, proactive cost management becomes a critical skill.

For an Italian SME with 20-100 employees, monthly AI-related costs can range from a few hundred to several thousand euros. This depends on the volume of requests, the complexity of the prompts, and the choice of model. In particular, the latest generation models like GPT-4o or Claude 3.5 have significantly higher per-token costs compared to previous models.

Beyond this, there are hidden costs. For example, context tokens: every time you send a long conversation to a model, you also pay for all the previous tokens. This mechanism can multiply spending in unintuitive ways.

Strategic reading: why the problem is structural

The problem of AI costs is not temporary. Likewise, it is not solved simply by waiting for prices to drop. In fact, competition among providers has already compressed margins. Further price reductions will be gradual and will not compensate for the growth in usage volume.

Therefore, the issue is structural. Companies that integrate AI into their workflows tend to increase consumption over time. This phenomenon is known as usage creep : each new use case adds tokens, each automation generates new requests. Consequently, without clear governance, spending grows organically and becomes difficult to control.

An analysis by Harvard Business Review emphasizes that the most effective organizations in AI adoption are not those that spend the most. On the contrary, they are those that define priority use cases in advance and measure the return of each application. Therefore, strategic discipline is worth more than budget availability.

For Italian SMEs, this principle is even more relevant. Margins are often tighter. Furthermore, internal technical resources to monitor and optimize AI spending are limited. In summary, a methodical approach is needed that not all companies have yet developed.

The work in progress: governance and operational guardrails

The industry's response to the cost problem is multi-layered. First and foremost, many companies are implementing granular monitoring systems. Each API request is tracked, classified by use case, and associated with a cost center. This makes it possible to identify areas of waste.

Next, we work on the prompts. The prompt engineering it's not just about output quality. It's also an economic optimization lever. More precise and concise prompts generate better responses while consuming fewer tokens. Therefore, investing in this skill yields a double benefit.

Furthermore, many organizations are reconsidering their model choices. The most powerful model isn't always the right one. For simple tasks like classification, data extraction, or generating short texts, lighter and less expensive models produce equivalent results. Therefore, a strategy of model routing — assigning each task to the most suitable model — can reduce costs by 30-50% without impacting quality.

We at SHM Studio we work with client SMEs precisely on this type of optimization. The goal is not to reduce AI usage. It's to maximize the value generated for every euro spent.

Operational implications for B2B retail and professional services

The sectors most exposed to the issue of AI costs in Italy are B2B retail and professional services. Both have adopted AI tools to automate communications, generate content, and support customer service. However, they've often done so without structured governance.

For B2B retail, the most common use cases include generating product descriptions, personalizing offers, and assisting with negotiations. Each of these processes continuously consumes tokens. Consequently, companies with large catalogs and intense sales cycles can incur significant costs.

For professional services — law firms, accountants, communication agencies — AI is used for drafting documents, summarizing information, and generating reports. In this case, the main risk is the context window overload : long documents repeatedly sent to the model multiply costs exponentially.

Therefore, the operational implications are clear. An audit of active use cases is needed. A mapping of real costs is needed. Finally, a prioritization strategy is needed to distinguish high-ROI uses from marginal ones. Companies that embark on this path today will have a significant competitive advantage in the next 18-24 months.

Tools and approaches to optimize AI spending in 2026

There are already tools on the market dedicated to monitoring and optimizing AI costs. Among the most used are solutions by LLM observability like LangSmith, Helicone, and Portkey. These tools let you track every API call, measure latency, and calculate the cost per generated output.

However, adopting these tools requires technical skills that many SMEs lack internally. Therefore, the role of an expert digital partner becomes crucial. An agency like SHM Studio can support the company in the audit phase, in selecting tools, and in defining sustainable usage policies.

In addition to this, there are architectural approaches that structurally reduce costs. For example, the caching of responses for recurring queries avoids calling the model every time. Similarly, the fine-tuning of smaller models on company-specific data can produce results comparable to large models at a fraction of the cost.

For those managing activities of Digital marketing or SEO , optimizing AI costs also translates into a review of contained production workflows. In particular, defining standardized prompt templates and limiting the length of contexts sent to the model are immediate, high-impact interventions.

The SHM Studio perspective: ROI first of all

The debate on AI costs risks polarizing into two extreme positions. On one hand, those who argue that AI use must be cut to contain spending. On the other, those who believe that any cost is justified by innovation. However, both positions are wrong.

The correct perspective is that of ROI. Every AI application must be evaluated based on the value it generates. Therefore, the question is not ‘how much do we spend on AI?’ but ‘how much is every euro spent on AI worth?’

To answer this question, SMEs need to develop specific metrics. For example, for content generation: cost per item produced with AI vs. cost with traditional methods, and compared quality. For automated customer service: cost per ticket resolved and first-contact resolution rate. For advertising campaign management : cost per qualified lead generated with AI support.

Finally, it's important to consider opportunity costs. A company that doesn't optimize AI spending today risks losing competitiveness compared to rivals who do. Therefore, optimization isn't an option. It's a strategic necessity.

Those who wish to delve deeper into these topics can explore the resources available in the SHM Studio blog or contact us directly via the contact page for a personalized consultation.