Mouna Benazzou: “The true cost of AI usage goes beyond billed tokens”

Nov 16, 2026

The creation of a dedicated Ministry for Artificial Intelligence in Morocco marks a new milestone in the Kingdom’s ambition to position itself as a key player in this strategic technology. Beyond institutional momentum, however, the real challenge lies in generating economic value. At a time when businesses are questioning the return on their AI investments, Mouna Benazzou, Head of Capgemini Morocco, shares her perspective on this equation.

  1. Artificial intelligence is increasingly becoming a major driver of business transformation. However, its deployment often comes with costs that exceed initial forecasts. What are the main factors behind this cost inflation today, and which expenses do companies tend to underestimate?

The first observation is that many companies still underestimate the true cost of scaling AI. The latest Capgemini Research Institute study shows that 59% of executives report that AI-related spending was significantly higher than expected over the past 12 months. Among the companies concerned, overspending reached an average of 24%.

This is partly due to a perspective that remains too focused on the apparent cost of technology, models or tokens. In reality, the true cost also lies in data, integration, infrastructure and security, but above all in indirect costs that are still poorly understood, such as human oversight, reworking outputs and governance.

A significant share of expenditure also remains less visible in initial budgets. This includes human supervision, checks, corrections, reruns and the reworking of generated outputs. These costs can increase with agentic systems, which rely on multiple models, tools and execution steps to complete a single task. As workflows become more complex, accurately monitoring their costs and performance becomes increasingly important.

The report also highlights this challenge: only one-third of executives feel confident in their ability to assess AI’s indirect costs. This points to a genuine management challenge. Companies need to move beyond tracking technology consumption towards a total-cost approach, covering the entire lifecycle, from design to operations.

At Capgemini, we believe that cost management must be built into the earliest architecture and deployment decisions. It is essential to scaling AI use cases while maintaining their economic viability.

  1. Beyond technology investments, the real question is now about return on investment. How can companies concretely measure the economic value generated by AI and ensure that deployed projects deliver tangible business results?

The first step is to start with the desired business outcome, not the technology itself. A company must clearly define what it wants to improve: reducing process costs, accelerating processing times, minimising errors, improving service quality, enhancing the customer experience or developing new revenue streams.

It is then essential to define performance indicators from the outset and establish a baseline. Relevant metrics may include time saved, cost per operation, output quality, error reduction or improved turnaround times. The real challenge today is demonstrating that these improvements translate into tangible business value. Our study shows that 57% of executives still find it difficult to demonstrate the business value of AI.

A task being completed faster does not necessarily translate into economic gains if the output subsequently requires additional checks, or if the time saved is not reinvested in value-creating activities. This is the fundamental difference between theoretical productivity gains and value actually captured by the business.

AI can improve operational performance, but it can also enable companies to develop new services, accelerate innovation, strengthen decision-making and unlock new growth opportunities.

This measurement must also be continuous. A project that performs well during a pilot phase will not necessarily deliver the same results at scale. Companies must therefore assess outcomes, adjust their solutions and identify which use cases genuinely deserve to be scaled.

This is precisely how Capgemini supports its clients: by starting with their business challenges, leveraging the Group’s technological and industry expertise, and building transformation roadmaps tailored to their operational realities. Our role is not simply to deploy AI solutions, but to help businesses turn technology investments into tangible, measurable and sustainable results.

  1. In the use of artificial intelligence, token consumption is often associated with the cost of models. What exactly does this unit of measurement represent, and why is its cost insufficient to assess the profitability of AI investments?

A token is a basic unit of language processing used by artificial intelligence models. It can represent a word, part of a word or a punctuation mark. Every time a user submits a prompt to a model or the model generates a response, tokens are consumed. Their volume is therefore one of the main billing mechanisms for many AI services, particularly models accessed through APIs.

For businesses, this consumption represents an operational cost that must be factored into investment assessments. However, the price of a token says nothing about the value of the output generated. A request may consume a large number of tokens without producing a usable response. Conversely, a limited amount of usage may be enough to automate a task, accelerate a process or improve decision-making.

The challenge, therefore, is to avoid confusing technology consumption with value creation. The true cost of AI usage goes beyond billed tokens. It must also include infrastructure expenditure, integration costs, human oversight, potential corrections and security requirements.

This distinction becomes increasingly important as companies expand their AI use cases. The question is not simply how to reduce token consumption or choose the cheapest models, but whether the resources deployed are generating measurable economic gains.

  1. With the growing number of artificial intelligence models and their sometimes widely varying operating costs, some companies are now adopting multi-model strategies. How can this approach optimise investment, increase flexibility and better align solutions with business needs?

A multi-model strategy means moving away from reliance on a single model and instead using the most appropriate model for each task. This is a logical evolution, as models differ in their reasoning capabilities, accuracy, speed, latency, reliability and cost.

An architecture can therefore combine advanced models for complex tasks requiring in-depth reasoning, lighter models for repetitive or high-volume operations, and specialised models for specific business domains.

The choice depends on the expected level of performance, data sensitivity, security constraints and acceptable cost. One of the key technical mechanisms behind this approach is model routing. It directs a request to the most appropriate model based on predefined criteria, such as task complexity, expected quality, cost, response time or risk level.

Other mechanisms, such as caching, make it possible to reuse previously generated results and reduce redundant computations. Optimising the context provided to models also helps control token consumption.

The Capgemini Research Institute report confirms this trend: more than three in five executives expect their organisations to adopt a multi-model strategy over the coming year. This reflects a growing awareness that model selection has become as much an economic decision as a technological one.

However, a multi-model strategy is not simply about connecting several models. It requires an orchestration architecture, comparable evaluation mechanisms, performance observability and governance rules to monitor costs, output quality and security.

At Capgemini, we support this evolution by combining AI expertise, technology architecture and industry knowledge. Our objective is to design environments capable of balancing performance, cost, reliability and security according to each company’s actual needs.

The best architecture is therefore not necessarily the one using the most powerful model, but the one that deploys the right capabilities, at the right time and at the best overall cost.

We are entering a new phase of maturity in artificial intelligence. It is becoming increasingly clear that no single model can optimally address every business need.

This evolution is already tangible: more than three in five executives surveyed expect to adopt a multi-model strategy over the next 12 months.

Models have different characteristics in terms of reasoning capabilities, speed, reliability, cost and data processing. The challenge is therefore to dynamically select the most relevant model for each task, balancing cost, capability, latency, risk levels and business requirements.

For example, a company may use a lightweight, more cost-effective model for repetitive or high-volume operations, while reserving more advanced models for complex challenges requiring greater reasoning capabilities.

But the benefits of a multi-model strategy go beyond cost optimisation. It also reduces dependence on a single provider, increases flexibility and allows architectures to evolve in line with technological innovation. This nevertheless requires genuine orchestration capabilities: determining which model to use, when to use it, what level of oversight is required, and which performance, security and cost criteria should apply.

This is precisely where technological expertise and an understanding of business needs must come together. At Capgemini, we address these challenges by combining artificial intelligence expertise, technology architecture and industry knowledge to design environments tailored to companies’ operational realities.

The best model is not necessarily the most powerful or the latest one. It is the model that best meets a specific need while maintaining the right balance between performance, cost, reliability and quality.

  1. Integrating AI is no longer just a technological choice; it involves a deeper transformation of organisations, processes and skills. In your view, what are the main changes companies need to make to fully harness this technology?

The first challenge is to rethink work processes. Simply adding AI to existing ways of working is not enough. Companies need to reconsider task allocation, decision-making processes, responsibilities and control mechanisms to fully leverage its capabilities.

This is particularly true of agentic systems, which can execute multiple actions and interact with different tools. Their deployment requires clearly defined levels of autonomy, human validation checkpoints and escalation mechanisms in the event of errors.

The second challenge is skills. Mastering the tools must go hand in hand with business expertise, critical thinking, analytical capabilities and sound judgement. Employees must be able to assess generated outputs, identify their limitations and retain responsibility for decisions. Training must therefore go beyond simply learning how to use tools and support the evolution of roles and professions.

It is also essential to consider how productivity gains are reinvested. Time freed up through the automation of certain tasks can be used to strengthen customer relationships, innovation, service quality and higher-value activities. Without this reflection, technological gains risk remaining theoretical.

I also strongly believe in the complementarity of generations and areas of expertise. Business experience, customer knowledge and mastery of new technologies must reinforce one another. This diversity of skills is what makes it possible to develop relevant and responsible use cases.

Finally, this transformation requires shared governance. Executive management, business teams, technology teams, control functions and human resources must work together to define priorities, responsibilities and expected outcomes.

At Capgemini Morocco, this conviction translates into a clear priority: developing skills, supporting the evolution of professions and creating the conditions that enable our teams to fully contribute to the transformations we deliver with our clients.

AI does not replace human expertise; it extends its reach. Technology can accelerate transformation, but it is people who give it meaning and determine its impact.

  1. At a time when companies must simultaneously accelerate their digital transformation, manage budgets and preserve their competitiveness, how can they strike the right balance between technological ambition and financial discipline? Which strategic trade-offs are becoming essential?

Technological ambition and financial discipline are not contradictory. On the contrary, they should reinforce one another. The question is not whether to reduce investment in AI, but how to allocate it more effectively and ensure that it supports the company’s strategic priorities.

The first trade-off is to prioritise investments according to their potential impact, feasibility and contribution to business objectives. Companies must move beyond multiplying experiments and focus their resources on initiatives that address clearly identified challenges. This also means accepting that projects that fail to demonstrate their relevance may need to be redirected or discontinued.

The second trade-off concerns technological foundations. Data quality, architecture, security, interoperability and governance can require significant investment, but they determine the reliability of solutions and their ability to scale. Neglecting these foundations can generate much higher correction and maintenance costs further down the line.

The third involves balancing short-term benefits with the development of long-term capabilities. Some initiatives can deliver operational gains quickly, while others require investment in infrastructure, skills and technological expertise. Both dimensions are essential to building a sustainable competitive advantage.

The economics of AI is therefore becoming a board-level governance issue. It is no longer a decision that belongs exclusively to technology teams: business, technology and finance leaders must share the same indicators and jointly assess investment, risk and value creation.

This discipline also requires the ability to regularly reassess priorities, measure outcomes and reallocate resources when expected benefits are not achieved.

This is the perspective from which Capgemini positions itself as a transformation partner to executive leadership teams. By combining consulting, technology expertise, engineering and business knowledge, we help organisations make informed decisions, secure their transformation roadmaps and align strategic ambition with economic performance.

Competitiveness will not be determined by the companies that invest the most in AI, but by those that know how to invest wisely and turn those investments into sustainable results.

  1. For Morocco, which is accelerating its positioning in digital, data and artificial intelligence, what are the main challenges and opportunities associated with this new AI economy? How can Moroccan companies move beyond technology adoption and make AI a genuine driver of competitiveness and value creation?

Morocco has several assets that can help it play an increasingly important role in this new economy: highly skilled talent, a growing digital ecosystem, international openness and experience in delivering complex projects. The challenge now is to turn these strengths into differentiated expertise and stronger design capabilities.

For Moroccan companies, the opportunities span industrial performance, operational efficiency, customer relations, decision support and the development of new services. However, creating value means starting with the specific challenges of each industry and building the necessary foundations: data, infrastructure, security, governance and skills.

Companies must also develop the ability to design, integrate and govern AI solutions, rather than simply use them. This requires expertise in data, architecture, software engineering, cybersecurity, intelligent systems and change management. Partnerships between businesses, universities and institutions are essential to bring educational programmes closer to the actual needs of projects.

This is an ambition I want to champion as Head of Capgemini in Morocco: bringing together our complementary expertise in technology services and engineering, our proximity to clients and our international network to strengthen the Kingdom’s contribution to global transformation.

This complementarity makes it possible to connect the transformation of information systems, data and AI with advanced engineering and industrial innovation. It also creates opportunities to evolve Capgemini’s delivery centres, including those in Morocco, towards greater involvement in design, specialised expertise and the resolution of complex challenges for international clients.

Our ambition is to help position Morocco as a hub for skills and value creation, capable of designing solutions and contributing to international value chains. The challenge is no longer simply to deliver from Morocco, but to create value in Morocco for the world.