Not long ago, I sat with an apartment owner who had taken over 144 units. There were 15 vacant, 12 on move-out notice, and four evictions for the end of the month. If the notices and evictions were to go through, the occupancy rate would be at 78 percent.

During our conversation, the owner asked me to help him identify a common factor among people who were moving out.

That could very well be done. In fact, I had done something similar for another client.

Little did I know that the project would be dead even before I started.

As soon as I had access to the systems, I realized the lack of data. When he took over the property, historical data, previous residents, and leases, which are critical for model training, were not transferred over.

This project became a casualty, joining the ranks of more than 80 percent of data science projects that fail.

Unfortunately, the failure rate for a data science project is high. The lack of data is not the only culprit; numerous factors contribute to these failures.

Before writing this article, I conducted research on the topic. What I found was that most people experienced what I have, but on different levels.

Based on my experience and research, I have compiled a summarized list of the reasons why most data science projects fail.

1. Stakeholder Disconnect

This was the single most common case for me. In every project I’ve led or participated in, there has always been a disconnect between the stakeholders and the technical team.

In one of the projects I worked on, the stakeholders felt that the data science team was taking too long to analyze and clean the data. Even after several meetings to explain our process and present the findings, the efforts were in vain. The project was cut short, and the data science team was dismissed.

In another case, a project lead instructed me to skip the EDA process and proceed directly to creating the algorithm. When I explained why EDA was essential, he dismissed me, saying the data was well-curated through the ETL process.

In another project, we lost the support of the higher-ups because the results did not meet their expectations. And what I mean by that is the numbers were telling them a story they didn’t want to hear.

2. Not Having the Right Data

Data scientists work with the available data. We do not create data, as a rule. Under the right circumstances, we can propagate data using its statistical attribution. In business, that is not something that could add value to the analysis.

What does not having the right data entail? As I mentioned in the opening, one owner lost access to years of historical data during a property acquisition. Without that foundation, there was nothing to build on.

The lack of correct data could also mean that you do not own it. During one project, stakeholders requested a detailed analysis of the competitors. Including vacancy, delinquency rates, and their monthly rental collection. This would be an impossible task as we didn’t have access to the competition data.

3. The Analysis Reveals What Everyone Already Knew

The coffee shop owner wants to know what the most sold product is in his shop. When the result turns out to be coffee, the data science project is shut down and considered a failure.

It is not that data science is useless for small businesses or always a failure. It has its place and application.

In working with many small businesses, I’ve learned to ask the right questions and understand the needs to provide the right solution. Data science is not data analysis.

In other cases, even when properly applied, it might return an answer that everyone already knew. The key is not stopping it, but building upon it.

When I worked on a customer churn analysis, it was already evident that the highest cancellation rate occurred in a specific region. Taking a step further, I was able to identify exactly why. A company was offering an enhanced service for less than what my client was charging.

4. No Clear Business Question

For the last two years or so, I’ve been hearing a lot of “we want to use AI.” Or another popular one, “we want to do something with our data.”

Since the AI boom, businesses have been scrambling to get in on it. AI development is expensive. And precision is a must.

Without a specific business problem or question in mind, a data science project will likely be disappointing.

I’ve been in numerous meetings where stakeholders requested AI models simply to claim they were powered by AI. But the results could have been obtained in a much easier and cheaper way.

5. No Path from Insight to Action

The analysis is solid. The model performs well. The presentation goes smoothly. Everyone nods in agreement.

Then the project dies in someone’s inbox.

This is the failure mode that frustrates me the most because the technical work succeeded. What failed was the bridge between insight and implementation.

I once delivered a maintenance optimization model to a property management company. The model accurately predicted which units would need HVAC repairs in the next 90 days. The operations team loved it. But no one had figured out how to integrate those predictions into their work order system. The maintenance supervisors kept doing things the way they always had. Six months later, the model was forgotten.

The gap between “here’s what the data says” and “here’s what we do differently on Monday morning” is where many projects go to die. A model without an implementation plan is just an expensive experiment.

How to Avoid These Failures

After years of projects that succeeded and failed, I’ve learned to ask certain questions before writing a single line of code.

About the data:

  • Does the data we need actually exist?

  • How far back does it go? Is that enough for what we’re trying to do?

  • Who owns it, and can we actually get access?

About the stakeholders:

  • Who is the decision-maker, and are they genuinely bought in?

  • What does success look like to them—not to us?

  • What finding would change their mind? What finding would they reject even if it were true?

About the outcome:

  • If this analysis works perfectly, what decision changes?

  • Who implements that change, and do they know this project exists?

  • Is there a system or process ready to act on the insights?

If I had asked these questions before the 144-unit project, I would have discovered the data gap in the first conversation—not after I was already committed.

Final Thoughts

Data science projects fail for many reasons, but rarely because the math was wrong. They fail because of misaligned expectations, missing data, unclear objectives, or no plan to act on results.

The technical work is often the easy part. The hard part is everything that surrounds it.

If you’re considering a data science initiative for your business, start with the questions above. They won’t guarantee success, but they’ll help you fail faster and cheaper—or better yet, avoid the preventable failures altogether.

A data science project could be a long game. If you don’t have the data, systems can be set up to capture the necessary data going forward. In a year, we’ll have what we need. Sometimes the best outcome of a failed project is knowing what to build next.