In a revealing mark of the times, a new multinational survey demonstrates that despite major issues with data quality, misuse of data scientist time, and financially damaging underperformance of AI models, an overwhelming majority of companies place a high degree of trust in their AI outputs.

The survey, orchestrated by Fivetran, a global expert in data movement, and conducted by the independent market research company Vanson Bourne, examined responses from 550 participants across Ireland, the UK, France, Germany, and the US, all hailing from organisations with a workforce of 500 employees or more. A disconcerting finding was that UK and Irish organisations seem to struggle more than their international counterparts with basic technical hurdles to AI adoption. Notably, 73% reported challenges with the seemingly basic task of accessing all the necessary data to run their AI programs.

The survey reveals that these organisations annually lose approximately 6% of their global revenues, amounting to an average of $406 million. This staggering toll is attributed to the operational inefficiencies due to inaccurate or low-quality data which is used to build AI models. These defective models then engender ill-informed business decisions.

Survey results disclosed that nearly 90% of organisations are utilising AI and machine learning (ML) methods to construct models that enable autonomous decision-making. In addition, an emphatic 97% of participants indicated plans to invest in generative AI within the upcoming one to two years. However, big concerns regarding data governance, security and accuracy persist. For example, data hallucinations and inaccuracies were reported 42% of the time by organisations leveraging large language models (LLMs).

Taylor Brown, COO and co-founder of Fivetran, stated that, "The rapid uptake of generative AI reflects widespread optimism and confidence within organisations, but under the surface, basic data issues are still prevalent, which are holding organisations back from realising their full potential.” To surmount these obstacles, he suggests organisations need to enhance their data integration and governance frameworks to produce more reliable AI outputs and mitigate financial risk.

Interestingly, the survey found that various job roles perceive distinct "AI realities". For instance, only 22% of technical executives, who build and operate AI models, consider their organisations' AI maturity to be 'advanced', compared to the 30% of non-technical workers who hold the same belief. In terms of data hallucinations - found in 42% of generative AI cases - these lead to uninformed decisions, diminish trust in LLMs, and consume considerable staff time in identifying and correcting the faulty data.

Recognising these significant hurdles, the majority (67%) of respondents plan to implement new technology to strengthen basic data movement, governance and security functions. This readiness to improve, and the wide-scale adoption of AI technology, suggests reason for optimism in overcoming the challenges highlighted by the survey.