AI at Any Price?
![](https://e3mag.com/wp-content/uploads/2024/05/2405-7-Capterra.jpg)
![](https://e3mag.com/wp-content/uploads/2023/12/pmf_RGB_400x400_2-150x150.png)
Some 81 percent of companies trust their AI/ML results, and nearly nine out of ten companies use AI/ML methods to create models for autonomous decision making. 97 percent will invest in generative AI in the next year or two. On average, they plan to invest 12 percent of their annual global revenue in this area.
These are the findings of a new study conducted for Fivetran by market research specialists Vanson Bourne. It provides insights into how AI is actually being used in organizations—particularly in terms of challenges, ROI, and costs. Companies with 500 or more employees in the US, UK, Ireland, France, and Germany were surveyed on the topic of AI and the associated processes and structures.
The companies surveyed, who generally have a high level of confidence in their AI/ML results, also report that they have fundamental data inefficiencies. Companies using large language models (LLMs) report data inaccuracies and hallucinations 50 percent of the time. On average, they lose six percent of their global annual revenue, or 406 million USD (out of an average annual revenue of 5.6 billion USD).
Hallucinating AI
The cause is inadequate AI models built with inaccurate or poor-quality data. According to the survey, nearly nine out of 10 organizations are using AI/ML methods to create models for autonomous decision making. At the same time, companies have problems with data inaccuracies and hallucinations, as well as concerns about data governance and security.
German companies are still in the early stages of using AI (60 percent), while only 39 percent in the US and 36 percent in France classify themselves as "AI beginners". Correspondingly, companies there consider themselves advanced: 31 percent (US) and 28 percent (France) use AI that requires as little or no human intervention as possible. In Germany, the figure is only 14 percent.
Overall, nearly nine out of ten companies (89 percent) use AI/ML methods to create models that can make predictions and decisions automatically. 80 percent of companies in the US and 75 percent in France have been doing this for at least six months, while only 44 percent in Germany say the same. Trust in the results of AI is also low in Germany: while 30 percent of German companies fully trust the results of generative AI, 47 percent of US companies and 48 percent of French companies say the same.
About one in four companies (24 percent) said they have reached an advanced stage of AI use, where they are fully leveraging the benefits of AI and relying on little or no human intervention. However, there are significant differences of opinion among respondents: technical managers who develop and operate AI models are less convinced of their company's AI maturity. Only 22 percent of them describe it as "advanced," compared to 30 percent of non-technical employees. The situation is different for generative AI: 63 percent of non-technical employees have full confidence in it, compared to 42 percent of technical managers.
Experts and hierarchies
There is also a disconnect between data professionals at different levels of an organization: while junior-level employees see outdated IT infrastructure as the biggest obstacle to developing AI models (49 percent), senior-level colleagues see the main problem as people with the right skills being focused on other projects (51 percent). In fact, they are forced to devote their resources to manual data processes such as data cleansing and repairing broken data pipelines. Organizations confirm that their data scientists spend the majority (67 percent) of their time preparing data rather than building AI models.
Poor data practices
The root cause of the wasted potential of data professionals and the underperformance of AI programs is the same: inaccessible, unreliable, and incorrect data. The magnitude of the problem is illustrated by the fact that most organizations struggle to access all of the data needed to run AI programs (69 percent) and put it into a usable format (68 percent).
New approaches to generative AI have introduced further complications: 42 percent of respondents have experienced data hallucinations. These can lead to poor decisions because the information base is inadequate. They reduce trust in AI or the willingness of employees to use the tool. They also take a lot of time to find and correct. Given that 60 percent of senior executives use generative AI and are responsible for strategic decisions, problems with data quality and trustworthiness are exacerbated.
Data governance and AI
Concerns about the use of generative AI also persist, with "maintaining data governance" and "financial risks due to the sensitivity of the data" being the top concerns for organizations (37 percent). A strong data governance foundation is particularly important for organizations looking to deploy either fully or partially homegrown generative AI models.
However, there is reason for optimism, as the majority (67 percent) of respondents plan to use new technologies to strengthen basic data movement, governance and security capabilities. Taylor Brown, co-founder and COO of Fivetran and a client of the study, said: "The rapid adoption of generative AI reflects widespread optimism and confidence within organizations.
But beneath the surface, there are still fundamental data issues that are preventing companies from realizing its full potential. Organizations must strengthen their data integration and governance foundations to achieve more reliable AI outcomes and minimize financial risk.
To the partner entry:
![](https://e3mag.com/wp-content/uploads/2024/02/Fivetran_Logo_png_1000.png)