AI arrived with a bang to redefine business operations, but even though the tech is ready to go next-level, many companies are hamstrung by a lack of the high-quality data necessary to move at the speed of AI, as well as a shortage of AI skills among the workforce. But are those obstacles slowing down big-money AI investments? New research from data movement firm Fivetran suggests they are not, and many businesses are even losing money hand over foot because the AI models they already have are underperforming due to these deficiencies.
The firm’s new report, AI in 2024—Hopes and Hurdles, shows 81 percent of organizations trust their AI/ML outputs despite admitting to fundamental data inefficiencies. The companies are losing on average 6 percent of their global annual revenues, about $406 million, due to underperforming AI models, which are built using inaccurate or low-quality data, resulting in misinformed business decisions. based on respondents from organizations with an average global annual revenue of $5.6 billion.
Most organizations have faced barriers limiting their use of AI:
The study—based on an online survey conducted by market research firm Vanson Bourne that polled 550 respondents across the U.S., U.K., Ireland, France and Germany from businesses with 500 or more employees—found that nearly nine in ten organizations are using AI/ML methodologies to build models for autonomous decision-making, and 97 percent are investing in generative AI in the next 1-2 years. At the same time, organizations express challenges of data inaccuracies and hallucinations, and concerns around data governance and security. U.S. organizations leveraging large language models (LLMs) report data inaccuracies and hallucinations 50 percent of the time.
As organizations enter the advanced stage, data quality becomes a key barrier:
“The rapid uptake of generative AI reflects widespread optimism and confidence within organizations, but under the surface, basic data issues are still prevalent, which are holding organizations back from realizing their full potential,” said Taylor Brown, co-founder and COO at Fivetran, in a news release. “Organizations need to strengthen their data integration and governance foundations to create more reliable AI outputs and mitigate financial risk.”
Data issues are a big hurdle even for those ahead of the game:
Different “AI realities” exist across various job roles
Approximately one in four (24 percent) organizations reported that they have reached an advanced stage of AI adoption, where they utilize AI to its full advantage with little to no human intervention. However, there are significant disagreements between respondents who work more closely with the data and those more removed from its technical detail.
Technical executives—who build and operate AI models—are less convinced of their organizations’ AI maturity, with only 22 percent describing it as “advanced,” compared to 30 percent of non-technical workers. When it comes to generative AI, non-technical workers’ high level of confidence is coupled with more trust, too, with 63 percent fully trusting it, compared to 42 percent of technical executives.
Most organizations have begun to adopt Generative AI to some extent:
There is a further dissonance between data experts at various levels of seniority within an organization. While those working in more junior positions see outdated IT infrastructures as the top barrier to building AI models (49 percent), their more senior colleagues say the problem is primarily employees with the right skills focusing on other projects (51 percent). It is true that data workers are forced to direct their resources towards manual data processes such as cleaning data and fixing broken data pipelines. In fact, organizations admit that their data scientists spend the majority (67 percent) of their time preparing data, rather than building AI models.
Bad data practices still prevalent
The root of the wasted data talent potential and underperforming AI programs are the same: inaccessible, unreliable and incorrect data. The magnitude of the issue is shown by the fact that most organizations struggle to access all the data needed to run AI programs (69 percent) and cleanse the data into a usable format (68 percent).
New generative AI use cases have introduced further complications, with 42 percent of respondents experiencing data hallucinations. These can lead to ill-informed decisions, reduce trust in LLMs or the willingness of staff to use the tool, and consume staff time in locating and correcting the data. With 60 percent of senior management using generative AI—and their responsibility to make strategic decisions—any issues with the quality and trustworthiness of data will be further amplified.
Data issues may be resolved soon, but closing the skills gap will also be necessary:
Data governance is a key focus area for AI use
Fears of generative AI use also remain, with “maintaining data governance” and “financial risk due to the sensitivity of data” tying for the top spot of concerns among organizations (37 percent). Solid data governance foundations will be particularly important for organizations that plan to either build their own generative AI models or use a combination of existing external and internally-developed models. However, as the majority (67 percent) of respondents plan to deploy new technology to strengthen basic data movement, governance and security functions, there is reason for optimism.
