AI is not a technical problem, it's an organizational problem

Why most AI projects hit their first obstacles well before the choice of model or technology.

The magic-solution fantasy

Since the arrival of ChatGPT and large language models, many companies have wanted to integrate artificial intelligence into their operations.

The requests often look the same:

“We want an AI to answer customer questions."

"We want an agent that can find our documents."

"We want to automate our business processes.”

In many minds, the problem seems essentially technical.

You just need to pick a powerful model, connect a few tools, and let the magic happen.

Yet after leading several AI projects for organizations of different sizes and sectors, one observation keeps coming back:

technology is rarely the real problem.

AI does not create knowledge

Artificial intelligence does not automatically transform a disorganized organization into an efficient one.

If information is scattered:

  • across email inboxes;
  • on users’ workstations;
  • in several cloud spaces;
  • in folders whose structure no one really knows;

then AI will inherit the same disorder.

The best model on the market cannot compensate for a lack of document governance.

AI does not create knowledge.

It exploits what already exists.

The invisible problem: data

When watching commercial demonstrations, it often feels as though everything depends on the model’s power.

In reality, most of the work lies elsewhere.

Before even talking about AI, you usually need to:

  • find existing information;
  • identify reliable sources;
  • remove duplicates;
  • structure documents;
  • organize reference data;
  • document processes.

This phase often represents the largest part of the project.

Yet it is rarely highlighted.

Because it is less spectacular than a chatbot demo.

What I actually encounter on the ground

When I work on an AI project, I rarely start by talking about models, agents, or RAG.

I usually begin with a few simple questions:

  • Where are the documents?
  • Who produces the information?
  • Who validates it?
  • Is there documentation?
  • How are folders organized?
  • What are the business processes?

And it is often at this point that difficulties appear.

Sometimes I ask for documents or a process map.

Then several weeks pass.

And the inevitable question arrives:

“So, where are we at?”

The answer is usually simple:

“I’m still waiting for the materials needed to move forward.”

In other cases, I am told:

“The documentation exists, but we no longer know where it is.”

Or:

“The person who managed that is no longer with the company.”

It also happens that I receive several hundred documents from different sources:

  • PDFs;
  • Word documents;
  • scans;
  • business exports;
  • duplicate files;
  • inconsistent folder trees.

All of it without clear organization, without validation, and sometimes without knowing which versions are still up to date.

The business side belongs to the organization, not the integrator

A common confusion keeps coming up in AI projects.

Some clients imagine that the integrator will naturally understand their business.

Yet every organization has its own vocabulary, its own rules, and its own exceptions.

Take a law firm as an example.

I can set up:

  • the infrastructure;
  • the document repositories;
  • the tools;
  • the AI agents;
  • the search engines;
  • the RAG systems.

However, I cannot guess on my own:

  • which files relate to labor law;
  • which files concern corporate law;
  • which documents are considered authoritative;
  • which procedures are actually used;
  • which exceptions exist in daily practice.

This expertise belongs to the professionals in the field.

It cannot be outsourced to a vendor nor delegated to an artificial intelligence.

AI depends directly on this knowledge.

The myth of technological magic

Many projects start with an implicit idea:

once the tool is installed, the artificial intelligence will understand the company.

It will find the information.

It will reconstruct the processes.

It will organize the knowledge.

It will correct inconsistencies.

Unfortunately, that is not how things work.

AI is capable of exploiting structured knowledge.

It is not capable of recreating on its own knowledge that has never been organized.

Dealing with reality

In practice, we often end up working with what is available.

Projects move forward despite:

  • incomplete documentation;
  • scattered knowledge;
  • imperfect reference data;
  • partially documented processes;
  • business experts who lack time.

And it is entirely possible to get useful results.

But we need to stay clear-eyed.

The result will rarely be as performant as that of an organization that has taken the time to review its knowledge, documents, and processes.

AI can compensate for some weaknesses.

It cannot turn a lack of organization into operational excellence.

Organization before technology

A company that truly wants to benefit from artificial intelligence should start by asking a few simple questions:

  • Where is our knowledge?
  • Who holds it?
  • Who maintains it?
  • How is it passed on?
  • Is it documented?
  • Is it accessible?

When these questions remain unanswered, AI often becomes an amplifier of the existing disorder.

Conversely, when an organization already masters its processes and documentation, the gains can be considerable.

Conclusion

Artificial intelligence is a remarkable technology.

But contrary to what the prevailing narrative sometimes suggests, it is not a magic solution.

The main challenge is usually not the choice of model, vendor, or infrastructure.

The real challenge is organizational.

AI often acts as a revealer.

It highlights what the company truly knows about itself, its processes, and its own knowledge.

Before asking:

“Which AI should we use?”

It might be useful to start with a much simpler question:

Do we really know how our own organization works?

← Back to blog