Artificial intelligence models are advancing at a spectacular pace. They analyse documents, summarise thousands of lines of text, answer complex questions and can now trigger actions within company systems.
Given these capabilities, it is tempting to think that selecting the best model is the main decision in an AI project.
Yet a much older issue consistently returns to the centre of the discussion: data quality.
In January 2026, the World Economic Forum noted that fewer than one in five organisations consider themselves truly mature in terms of data readiness. More than half of the executives surveyed cite data quality and availability among the main obstacles to accelerating AI adoption.
In other words, we are building increasingly intelligent systems on foundations that can still be fragile.
AI does not erase data problems
Imagine an SME with several years of sales history.
In its CRM, the same customer appears under three names:
Dupont SA.
DUPONT.
Dupont Company.
An address has been updated in the invoicing software but not in the CRM. The latest sales exchange is stored in an employee’s inbox. A contract document sits in a shared workspace under a filename that makes it difficult to find.
An experienced person can sometimes reconstruct the situation through their knowledge of the company.
An AI system, however, works with the information it can access.
If the data is contradictory, incomplete or outdated, it does not spontaneously create the truth. Instead, it may produce an apparently coherent answer based on inconsistent information.
The problem is not the model.
The problem is the data.
Automating an error also means making it faster
When AI is used occasionally, incorrect data can lead to an incorrect answer.
When an AI agent is integrated into a process, the consequences can be more significant.
The agent may retrieve incorrect information, make a decision, update a tool, send a message or trigger an action.
The greater the autonomy, the more critical data quality becomes.
This can be an uncomfortable reality: AI can amplify an organisation’s strengths, but also its weaknesses.
A well-structured customer database becomes easier to use.
A disorganised customer database can turn an AI project into a succession of exceptions, corrections and manual validations.
Data is no longer just a technical matter
For a long time, Data projects were often seen as matters reserved for IT teams: databases, warehouses, pipelines, ETL and governance.
AI is gradually changing that perception.
When an executive wants to automate customer request processing, create an internal assistant or analyse sales activity, the question of data immediately arises.
Do we have a single source of truth?
Who is responsible for this information?
How often is it updated?
Can the AI access it?
Is any of the data sensitive?
How can its quality be verified?
These questions no longer concern only the IT department. They directly involve business teams and management.
Start small, but start with clean foundations
An SME does not need to launch a vast Data governance programme immediately.
The approach can be much more pragmatic.
For a specific use case, it is possible to identify the data that is genuinely required, assess its quality, define its reference source and resolve the most critical inconsistencies.
The aim is not to make all company data perfect before getting started.
It is to ensure that the data used by the process being improved is reliable enough.
At ENJAILLE, we believe that Data and AI can no longer be considered separately.
The AI model matters. The architecture matters. The choice between a cloud-based or local solution matters.
But no technology can replace reliable, accessible and properly managed information.
Before asking what AI can do with your data, another question therefore deserves to be asked:
can you genuinely trust your data?
Main source: World Economic Forum, “Why data readiness is now a strategic imperative for businesses”, 19 January 2026.