Agentic AI: How to Choose a Simple, Measurable, Low-Risk First Project

After discovering the potential of agentic AI and understanding how it can help organizations increase their capacity, one question almost always comes up: where should you start?

For many business leaders, the question is no longer whether artificial intelligence will have an impact on their industry. They are already seeing its influence on operations, customer service, sales, and administrative functions. The real challenge is determining how to integrate agentic AI into the business and choose a first use case that will deliver real value.

This is often where organizations make their first mistake.

Agentic AI in Business: Avoid the Trap of an Overly Ambitious Project

When a new technology generates significant interest, it can be tempting to transform several processes at once. Some businesses are already imagining an AI agent capable of managing all customer requests, coordinating multiple systems, or automating an entire department.

While that vision may eventually become a reality, it is rarely the best place to start.

The most relevant first projects are generally much simpler. They address a known pain point, a repetitive task, or a process that has been slowing operations down for some time.

During his presentation at Salon Connexion, Eric Murray emphasized a fundamental principle: the first gains should be quick, measurable, and relatively easy to achieve. These small wins make it possible to gradually expand the use of AI and build the confidence needed to go further.

Start with a Business Process, Not the Technology

One of the key statements from the presentation sums up this philosophy well:

“We start with a process, not the technology.”

This distinction may seem simple, but it completely changes how an AI project is approached.

Too often, organizations begin by comparing platforms, artificial intelligence models, or the features available on the market. But technology does not create value on its own. Value is created when technology is applied to a real business problem.

Before discussing tools, organizations should look at their operations. Which processes are slowing teams down? Where do delays occur most frequently? Which activities are repeated several times a day? Where are employees spending time searching for, transferring, or validating information?

This is generally where the best process automation opportunities and the most relevant use cases for agentic AI can be found.

How Do You Choose a First Agentic AI Use Case?

The best first projects are not necessarily the most impressive. The initial processing of requests submitted through a web form, preliminary invoice analysis, extracting information from documents, or preparing certain administrative follow-ups can all be excellent starting points.

A strong first use case should ideally meet a few criteria:

• the process occurs frequently;
• it takes enough time for the improvement to be measurable;
• its rules are relatively well defined;
• potential errors can be detected and corrected;
• results can be compared with a clear baseline.

These projects can generate tangible gains while giving teams an opportunity to better understand what AI agents can actually bring to their operations.

Why Start with a Low-Risk AI Project?

The level of autonomy given to an agent is another important consideration when choosing a first project.

There is no need to begin with an agent that independently makes important decisions or automatically executes every step of a process. A first AI agent can analyze a request, classify information, gather data, prepare a recommendation, or suggest an action while keeping human validation in place when necessary.

This approach makes it possible to test the solution in a controlled environment and assess the quality of its results before gradually increasing its level of autonomy.

It also gives teams an opportunity to understand how the agent responds to less predictable situations and identify the exceptions that need to be addressed.

The first project therefore becomes both an opportunity to generate value and a way to learn how to integrate AI into business processes progressively and responsibly.

How Do You Measure the Results of an AI Project?

The enthusiasm surrounding artificial intelligence can sometimes overshadow a fundamental principle: without a point of comparison, it is difficult to demonstrate real gains.

Before deploying a solution, it is important to document the current situation. How long does the task take? How many people are involved? What volume is processed each week? What delays or errors are currently being observed?

Eric Murray also emphasized the importance of measuring the situation before and after deployment.

Does a process that previously required ten hours of work per week now take four? Has a two-day processing time been reduced to a few hours? Can a team handle more requests without increasing its workload?

These results help determine whether the project is actually creating value and make it easier to decide whether to go further with automation or agentic AI.

Involve the People Who Really Know the Process

A process that seems simple on paper often includes exceptions that only the employees who perform it every day are aware of.

An invoice from a particular supplier may need to be handled differently. Certain types of customer requests may always require validation. Missing information may trigger an additional step.

These details matter when part of a process is being entrusted to an AI agent.

Frontline teams can identify these exceptions as well as the real pain points within the process. Their involvement therefore helps improve the solution while making adoption easier.

Rather than automating a process as it is documented, the goal is to automate the process as it actually works.

How Do You Get Started with Agentic AI?

Agentic AI should not be approached as one large technology project with a defined beginning and end.

Early deployments are as much about learning as they are about generating gains. They help teams understand the possibilities and limitations of the technology, adjust their ways of working, and gradually identify additional automation opportunities.

A simple, well-chosen first project can therefore become the starting point for a much broader initiative.

You do not need an AI strategy that covers your entire organization before getting started. You need a relevant first use case, a way to measure its impact, and a framework that is safe enough to learn from. The rest can be built from there.

At Kezber, we help organizations identify their first agentic AI use cases, assess potential gains, and progressively implement solutions tailored to their business reality.

Wondering where to start with agentic AI? Let’s identify a first project that is practical, measurable, and suited to your organization.