AI Projects: Don't Start with the Model, Start with the Value

The True First Step of an AI Project

When a company decides to embark on an artificial intelligence project, a common mistake is to start with the choice of the model. This approach, while understandable, is often premature. In reality, the starting point should be the identification of the business value to be created, the analysis of available data, the establishment of a framework of trust, and the assessment of the ability to integrate with the existing information system.

For the past two years, many organizations have focused on questions such as: "Which model should we use?" or "Should we opt for GPT, Claude, Mistral, or another?" While these questions are legitimate, they should be asked after defining the business problem to be solved and the data to be leveraged.

The Illusion of Spectacular AI Models

Generative AI models, capable of writing, translating, or analyzing, captivate with their performance. They give the impression that the essence of the project lies in their selection. However, in a company, real value does not come solely from the model. It depends on its ability to be connected to a specific business context, reliable data, governance rules, existing processes, and concrete use cases. A high-performing model that is poorly integrated can yield little value, while a basic model, well-contextualized, can transform an activity.

This is a crucial idea: AI is not just a generation technology. It is an integration technology.

Defining Business Use Before Anything Else

Before choosing a model, it is crucial to define the intended business use. Whether it is to reduce the processing time of a file, improve the quality of customer support, accelerate document production, assist developers in the software lifecycle, help sales teams prepare for their meetings, detect operational anomalies more quickly, or automate part of a repetitive business process, the objective must be clear and measurable. An AI project should not start with: "We want to use AI." It should start with: "We have a specific, measurable, recurring, costly, or strategic problem, and AI might help us solve it." The nuance is important. In the first case, the company is seeking a testing ground for a technology. In the second, it is looking for a lever to improve a business outcome.

The Importance of Data

Without quality data, AI remains a mere demonstrator. In many projects, the obstacle is not the model. It is the quality, accessibility, freshness, governance, or structuring of the data. Where is the data? Is it in files, databases, SaaS tools, emails, tickets, wikis, ERPs, CRMs? Is it reliable? Is it up to date? Is it understandable? Is it usable by AI without exposing sensitive information?

This is often where the AI project becomes an architecture project. For an AI assistant to provide a useful response, it must understand the context. And this context does not come solely from its general training. It comes from internal data, business documents, organizational rules, histories, reference materials, past decisions, and company-specific constraints.

The question is therefore not just: "Which model to choose?" but also "How to provide the model with the right context, at the right time, with the right level of security?"

Building Trust

Generative AI introduces a new reality into information systems: probability. A traditional software executes a deterministic rule. An AI model produces a probable response, influenced by its training, its prompt, its context, and its parameters.

This does not mean it is unusable in a business context. But it does mean it must be framed. What level of error is acceptable? Which responses must be verified by a human? What actions can be automated? Which decisions must remain under human control? How to trace responses? How to audit sources? How to prevent the leakage of sensitive data? How to manage biases, hallucinations, or non-compliant responses?

Trust is not decreed. It is built through architecture, governance, testing, observability, and safeguards. A serious AI project is therefore not just about connecting a model to an interface. It is about creating a reliable system around the model.

Integration and Industrialization

An AI prototype can be impressive in just a few days. A reliable AI product requires real industrialization capability. This is often where many projects fail. They work in the lab but not in production. They impress in demonstrations but do not integrate into real processes. They perform well on a few examples but become unstable at scale. They attract innovation teams but remain difficult to adopt by business units.

To move from prototype to product, very concrete issues must be addressed: identity, access rights, confidentiality, supervision, monitoring, cost management, response quality, performance, latency, security, compliance, and the management of the lifecycle of prompts, models, and data.

In other words, AI must become a component of the information system. Not an isolated experiment.

A Complete Value Chain

A successful AI project rarely relies on a single element. It relies on a complete chain: business use cases, data, context, model, security, integration, user experience, value measurement. The model is an important piece. But it is just one piece. Value emerges when the whole operates as a coherent system.

This is why the most mature organizations do not just ask which model to use. They build a platform, a method, and governance capable of replicating AI use cases at scale. They shift from a demonstration logic to a product logic. From an experimentation logic to an industrialization logic. From a technological fascination logic to a business impact logic.

Choosing the Model Last

The choice of the model should be a consequence of needs, not a starting assumption. If the use case requires complex reasoning, the choice will be different. If it demands low latency, the choice will be different. If it deals with sensitive data, the choice will be different. If it requires a private deployment, the choice will be different. If it needs to be highly specialized, the choice will be different. If it must optimize costs, the choice will be different.

The right model therefore depends on the context. There is no universally best model. There is a model suited to a use case, a risk, a budget, an architecture, and a business strategy.

Conclusion: Start with Value

Starting an AI project by choosing a model is like starting a digital transformation project by choosing a technical framework. It is sometimes necessary. But that is not where success lies. The real question is not: "What model are we going to use?" The real question is: "What capability do we want to create for the business?"

A capability to make better decisions, to produce better, to serve customers better, to reduce complexity, to accelerate teams, to ensure reliable operations, to transform the employee experience, to create new services.

AI is not a model project. It is a project of value, data, architecture, trust, and adoption. And that is precisely why the first question should never be: "Which model to choose?" but rather: "What problem do we want to solve, with what level of trust, and how are we going to move from experimentation to real impact?"