The return on investment (ROI) of AI is dismal, new reports show, meaning that C-suite patience may not be paying off as AI adoption only seems to be increasing.

This is according to the new 2024 State of AI report from Appen Limited, which surveyed 500 IT decision makers across the US, finding that AI adoption is growing.

However, the adoption of generative AI and machine learning are hindered by a lack of accurate and high-quality data. The report found a 10% increase year over year in bottlenecks related to data – its sourcing, cleaning, and labelling.

“Enthusiasm around GenAI and other AI-powered tech remains high, as indicated by the rapid growth, but users are quickly finding that the promise of these tools is matched by an equally daunting challenge,” said Si Chen, VP of strategy at Appen.

“The success of AI initiatives relies heavily on high-quality data, and this is becoming more difficult as AI use cases increase in complexity and become more specialized.

“This is reflected by the fact that high-quality annotations, meaning high consistency and accuracy, are the top features companies seek in a data annotation solution. Those building the AI tools and models of tomorrow value strategic data partnerships now more than ever.”

Adoption of GenAI is up 17% in 2024 versus the previous year, a healthy growing pace. At the same time, however, 86% of respondents retrain or update their ML models at least once every quarter, indicating a critical desire for fresh, relevant and high-quality data as accuracy declines.

Data accuracy, meanwhile, has decreased by 9% since 2021, making the quest for high-quality data a major challenge – as models are being iterated more frequently, data remains the most significant challenge, especially where accuracy and availability are concerned.

Recommended reading

• GenAI Top Spending Priority of CIOs
• 2024 AI and Cloud Funding to Hit $79BN in US, Europe, Israel, Accel Says
• Largest AI Data Centre in Europe to Be Built on Former Britishvolt Site

Further, enterprise deployment and ROI appear to be down in 2024. The mean percent of AI projects making it to deployment has dropped by 8% since 2021, while the mean number of deployed AI projects that have shown meaningful ROI has dropped 9.4%.

Despite the rapid growth of GenAI, managing bias and ensuring fairness in GenAI model training remains a key challenge. Custom data collection is the primary method for sourcing genAI training data.

The diversity of data is also the most valued feature in AI, followed by the efforts to reduce bias, and ensure scalability, with 97% saying these elements are vital to building AI models.

The research also found that humans are still more vital to AI than some might expect, as 80% of respondents highlighted the importance of human-in-the-loop ML, validating that human insight is key to refining AI systems.