Why data scientists hold the keys to unlocking generative AI’s business value

Informatica’s recently released CDO Insights 2025 survey found that 67% of surveyed data leaders have not been able to successfully transition even half of their generative AI pilots to production. 

There are many reasons why generative AI projects never make it out of the pilot stage, but one of the biggest reasons is data. As companies launch AI pilots, they quickly run into data problems. Often, they can’t get a handle on their unstructured data, or they don’t have a way for data to flow between siloed tech systems.  

While generative AI has made artificial intelligence more accessible to more businesses and industries than ever before, data science as a discipline has never been more crucial. Without data scientist support, most generative AI pilots will fail and companies looking to differentiate with AI will fall far short of their ROI expectations.  

In this article, I’ll explain why data scientists are one of the most important keys to unlocking generative AI’s business value. 

What is a data scientist?

A true data scientist is someone who bridges the domains of business analysis, data wrangling, and AI & ML (machine learning) modeling to provide advanced and automated solutions to business problems.

Data scientists mitigate AI’s underlying complexities

As generative AI tools become more user friendly, companies may think they’re “plug and play.” In many cases, generative AI’s accessibility is both an advantage and a disadvantage because users may not understand the underlying complexities. They often overlook the following issues: 

Issue: Quality data acquisition. Most businesses have plenty of data. However, they lack quality data that’s in a usable format and relevant to the business use case. The old adage, “garbage in, garbage out,” is especially true when it comes to generative AI, and bad data can doom a project from the start. 

Solution: Data scientists can define data specifications, integrate data from different sources, and curate the data to make it ready for AI use cases. 

Issue: Modeling and optimization. To scale an AI initiative, teams first need to experiment on a selection of different AI/ML algorithms and architectures and validate the performance results. This type of experimentation validation is typically outside most implementation teams’ scope of expertise.

Solution: Data scientists can recommend the most appropriate configurations based on various constraints such as accuracy, inference speed, cost, privacy, compliance, and more.

Issue: Use-case oriented interpretability. Without an understanding of how AI works, AI outcomes can be a "black box" in which it’s difficult to understand how the model inference was made. 

Solution: Data scientists help make AI’s outputs more interpretable and ensure the models' decisions are aligned with business goals, helping stakeholders trust and adopt AI solutions.

What do AI complexities look like in a common business use case? 

Right now, many companies are exploring generative AI knowledge assistants to use with both internal employees and with customers. The first problem they run into is that they don’t know if their existing knowledge content is ready for AI solution development. Second, they usually don’t have empirical data on the most in-demand knowledge themes. Lastly, their data sits in multiple (often siloed) places.

These are three big hurdles for an AI implementation of this type, so it’s a perfect opportunity to leverage a data scientist. A data scientist can assess the quality of existing knowledge content with tools like content intelligence. A data scientist can also analyze customer interaction data to capture the most in-demand knowledge themes with tools like conversation intelligence. The scientist can then help minimize the gap between knowledge supply and knowledge demand, and connect the data across multiple systems. 

Data scientists keep AI running – and improving

AI is not something businesses can “set and forget.” AI and ML solutions always get better or worse. They never stay the same. 

If left alone, the AI solution will degrade, due to changing world conditions or model drift. For those outside the realm of data and analytics, model drift (or model decay) is the decline in accuracy of a model caused by changes in data distribution or relationships between variables. This decline in AI performance is another reason why AI pilots are often abandoned. The solution simply stops working as intended. 

On the other hand, with retuning, refining, retraining, and additional data source incorporation – activities performed by data scientists – the AI solution improves over time. Data scientists are trained to interpret fluctuations in model performance and can identify reasons for the changes. They can then determine the next best action, which may be model calibration, model feature removal, model refresh, model retrain, or something else. 

The case for data science in AI value realization

More and more, businesses are recognizing the importance of data science. According to CX Today’s AI in Customer Experience survey, “41 percent of CX teams [are] planning to hire data scientists and AI experts to secure maximum benefit from the technology.” 

The report goes on to state, “… CX leaders must appoint a dedicated, agile team behind their AI transformations, continually monitoring, optimizing, and turning the screw. Without that, AI won’t achieve its potential.”

In short, businesses must invest in data science to maximize AI ROI. Without data science, AI pilots will fail to reach their goals and innovation will become stagnant — the exact opposite of what AI is intended to do.