Generative artificial intelligence (AI) is expected to see accelerated adoption in the next two years, with an overwhelming majority (97%) of organizations planning to invest in the technology. These company executives are placing a significant amount of trust and capital into AI, but they don’t have the proper data foundation in place to improve their business outcomes, according to new research from Fivetran.

The survey conducted by market research firm Vanson Bourne reached 550 respondents in the United States and Europe. While more than 80% of organizations say they trust the outputs of their AI models, more than 40% have experienced data inaccuracies, hallucinations and data biases in their AI outputs.

In the U.S., enterprises using large language models (LLMs) reported data inaccuracies and hallucinations from those models 50% of the time. Data hallucinations lead to ill-informed decisions, damage trust in LLMs or employee willingness to use genAI, and waste employee time tracking down accurate data.

“As companies rush to adopt generative AI [genAI], they continue to face major issues with inaccessible and unreliable data,” Fivetran Field CTO Mark Van de Wiel told SDxCentral. The research found 69% of organizations struggle to access all the data needed to run AI programs, and 68% struggle to cleanse their data into a usable format.

“Under the surface, basic data issues are still prevalent, which are holding organizations back from realizing their full potential,” Fivetran COO Taylor Brown said. With a majority (60%) of senior management using genAI to make strategic decisions, those data quality and trustworthiness issues will continue to complicate success.

The cost of AI data issues These data challenges translate to significant financial losses for businesses. Underperforming AI models built on inaccurate or low quality data result in an average of $406 million in lost revenue annually, based on data from respondents at organizations with more than $25 million annual revenue and an average of $5.6 billion.

Just 4% of respondents either did not measure or did not know the financial impact of underperforming AI programs and were not included in the average financial loss figure.

It will likely take years for organizations to fully capitalize on their investments in AI, considering most enterprises “are in the very nascent stages of adoption – where a lot can go wrong,” Van de Wiel said.

“The old expression ‘garbage in, garbage out’ has never been more appropriate than in the world of AI/ML,” he added. “Accurate, timely, secure and governed data is the cornerstone of successful AI/ML and genAI applications. Most companies are still struggling with fragmented, disparate data silos and can’t unlock that valuable data to be able to train AI/ML models,” he said.