Myth #11: Bad Data Is the Main Problem

Ask why AI projects struggle and “the data isn’t good enough” is probably the most common answer. And there is a good reason for that. Missing data, inconsistent definitions, poor master data, or information scattered across systems can make an AI use case difficult or sometimes impossible.

But I think we put too much of the blame on data.

In many projects, the bigger problem is that the organization has not really committed to the transformation. There is no senior business leader who owns the outcome, the project competes with twenty other priorities, and everyone wants AI as long as it does not require changing the way the business actually works.

And AI does require change.

If AI is going to take over part of a planning process, manage transportation exceptions, support procurement decisions or automate customer interactions, you cannot simply insert it into the existing process and assume everything around it stays the same. Some activities disappear. Others move to different people. Approval steps that existed because a person was doing the work may no longer make sense. People who previously spent their day collecting information or following up on routine exceptions may only need to handle the cases AI cannot resolve.

This is where many projects get stuck.

Nobody wants to remove an old report or control even though the new process makes it redundant. The pilot continues, but the underlying process stays exactly as it was.

None of these are data-quality problems.

They also cannot be resolved by the project team alone. Someone with authority over the business needs to decide which process will be used, which exceptions are genuinely necessary, which steps can disappear and who will be responsible for the new way of working. If every disagreement has to go through another steering committee or remain unresolved because nobody wants to challenge the existing process, the project will stall regardless of how good the model is.

This is why executive sponsorship on a slide is not enough. The business leader needs to be involved when these issues arise and actually resolve them. They also need to give the project enough priority that operations, IT and other teams provide the people and time required to implement it.

Data problems are real, but many of them can be fixed when the organization decides they are important enough to fix. Definitions can be aligned. Missing information can be captured. Master data can be cleaned. Systems can be integrated. None of this is easy or cheap, but it can be done.

And data itself needs ownership. If supplier names are inconsistent, inventory records are incomplete or the same KPI is calculated differently across locations, cleaning the data for a pilot may get the project moving. But unless someone is responsible for keeping that data accurate, the problem comes back.

This is also why a successful pilot does not necessarily translate into a successful rollout.

A pilot can work around many of these problems. The project team can manually clean the data, bypass an integration, choose one cooperative site and resolve exceptions manually. That is perfectly reasonable when the objective is to find out whether the technology works.

At scale, the company has to decide what the process is. It has to decide what AI does, what people still do, what happens when AI cannot resolve something, which local variations are actually necessary and which are simply there because “we have always done it this way.” Someone also has to own the data and the result after the implementation team leaves.

This is why “our data isn't ready” can sometimes hide a much broader problem. Sometimes the data really isn't ready. Sometimes nobody has agreed on the process the data is supposed to support. Sometimes the business wants the benefits of AI without changing roles, processes, controls or priorities.

Another data-cleansing project will not fix those problems.

Before spending months trying to make the data perfect, decide what the process should actually look like with AI in it.

1. What will AI do?
2. What will people still do?
3. Which steps are no longer needed?
4. Which local differences really need to remain?
5. Who can make those decisions when teams disagree?

Then determine what data that process needs and who will be responsible for keeping it usable.

The lesson

Bad data can stop an AI project. But clean data does not compensate for a process nobody is willing to change, a project nobody has authority to prioritize, or an AI implementation that simply adds technology on top of the way the company already works.

Get your data and processes AI-ready →