The problem with AI: It’s not you, it’s the data

Corporations are leaving billions on the table because they can’t get their data acts together. If they are to succeed at achieving value through data-driven initiatives such as artificial intelligence, they need to better align and support the backend data that is feeding these systems.

That’s the gist of the latest research, based on a survey of 2,500 executives and published by Infosys Knowledge Institute, which estimates that companies could collectively generate more than $460 billion in incremental profit if only people could manage their data resources a little better. 

This consists of improving data practices, trusting more in advanced AI, and integrating AI more tightly with business operations. Business value is still elusive. 

Also: 8 ways to reduce AI burnout

The survey identified three obstacles to effective AI implementations: Lack of a cohesive, centralized data strategy, weak data verification, and lack of proper infrastructure. Most companies don’t have a consistent data management strategy. 

Respondents want to manage data centrally, but this is not what most do right now. Analysis of the survey results “shows that centralized data management links to better profit and revenue growth. 26% of respondents currently have a centralized approach; 49% would like to have adopted this approach by next year.

“Data is not the new oil,” the study’s authors, Chad Watt and Jeff Kavanaugh, both with the Infosys Institute,  emphasize. “Businesses can no longer afford to think of their data as oil, extracted with great effort and valuable only when refined.”

Data today is more like currency: “It gains value when it circulates. Companies that import data and share their own data more extensively achieve better financial results and show greater progress toward ideating AI at enterprise scale — a critical goal for three out of four companies in the survey,” says Watt and Kavanaugh.

Also: The people building artificial intelligence are the ones who need AI the most

The success of currency is dependent on trust, and this also applies to data. “Advanced AI requires trust,” the authors state. “Trust in your own and others’ data management, and trust in AI models. Pristine data and perfectly programmed AI models mean nothing if humans do not trust and use what data and AI produce.”  

Companies that shared data, in and out of their organization, are more likely to have higher revenue and use AI better, the survey shows. “Refreshing data closer to real time also correlates with increased profits and revenue.” 

Another anti-oil analogy the study’s authors framed is that data is more like nuclear power than fossil fuel. “Data is enriched with potential, in need of special handling, and dangerous if you lose control. Twenty-first century data has a long half-life. When to use if, where to use, and how to control it are as critical as where to put it.”  

Most businesses are new to AI, the survey shows. More than 8 in 10 companies, 81%, have only deployed their first true AI system in the past four years, and 50%, in the last two. In addition, 63% of AI models function only at basic capability and are driven by humans. They often fall short on data verification, data practices, and data strategies. Only 26% of practitioners are highly satisfied with their data and AI tools. “Despite the siren song of AI, something is clearly missing,” the survey’s authors state.

Also: AI’s true goal may no longer be intelligence

The survey’s authors identified high-performing companies, which tend to have a strong focus in three areas:

• They transform data management to data sharing. “Companies that embrace the data-sharing economy generate greater value from their data,” says Watt and Kavanaugh. “Data increases in value when treated like currency and circulated through hub-and-spoke data management models. Companies that refresh data with low latency generate more profit, revenue, and subjective measures of value.”
• They have made the move from data compliance to data trust. “Companies highly satisfied with their AI (currently only 21%) have consistently trustworthy, ethical, and responsible data practices. These prerequisites tackle challenges of data verification and bias, build trust, and enable practitioners to use deep learning and other advanced algorithms.”
• They engage everyone in the AI process. “Extend the AI team beyond data scientists. Businesses that apply data science to practical requirements create value. Business leaders matter as much as data scientists. Good AI teams typically involve multiple disciplines. “Data verification is the greatest challenge to moving forward, along with AI infrastructure an compute resources.