Trends in the evaluation and implementation of enterprise AI adoption

New O’Reilly research explores the techniques, tools, and practices enterprise organizations are using to better understand how artificial intelligence (AI) has evolved over the past year. While this year’s survey generated nearly three times as many responses as last year, indicating overall industry growth, there are still challenges ahead.

Demand for AI expertise exceeding supply

To gauge the overall maturity of AI, the survey sought to uncover challenges respondents faced when evaluating solutions. While in last year’s survey respondents cited company culture (22%) as the major bottleneck to enterprise AI adoption, lack of skilled people and difficulty hiring topped the list this year, noted by 19% of respondents.

This shift is significant, as it implies a greater overall acceptance of AI, but it also reveals the very real and persistent AI talent gap.

While it’s not surprising that demand for AI expertise has exceeded supply, it’s important to understand which specific skills and professional titles are most critical to AI adoption. Companies feel the skills shortage most acutely in the areas of ML modeling and data science (52%), understanding business use cases (49%), and data engineering (42%).

The survey also found that the percentage of companies with AI products in production over the last year (25%) is flat when compared with 2020 (26%) and 2019 (27%), which may be reflective of the AI skills gap.

Key findings of enterprise AI adoption

• The second-most significant barrier to AI adoption is quality data (18%). Organizations are beginning to realize the importance of good quality data—a sign that the field is maturing.
• The percentage of respondents reporting mature practices (26%)—that is, the ones that had revenue-bearing AI products in production—has stayed roughly the same over the last few years.
• Among respondents with mature practices, scikit-learn (65%) and TensorFlow (65%) were the most used AI tools. This varies slightly for respondents evaluating or considering AI: scikit-learn (48%) and TensorFlow (62%).
• Supervised learning (82%) and deep learning (67%) were the most popular techniques used by respondents at all stages of adoption.
• When asked what kinds of data mature respondents were using, 83% cited structured data (logfiles, time series data, geospatial data), followed by text data (71%). Answers were similar among general respondents.
• As for evaluating risks, mature organizations checked for unexpected outcomes or predictions, interpretability and transparency, and model degradation. Although privacy and fairness, bias, and ethics ranked above 50%, they were only midrange concerns.
• The retail sector (40%) has the highest percentage of mature practices. Education (10%) has the lowest percentage but the highest number of respondents who are considering AI.

“Enterprise AI has grown; the sheer number of survey respondents will tell you that, but deployment of AI applications into production has remained roughly constant, and with it, overall maturity in the field,” said Mike Loukides, VP of content strategy at O’Reilly and the report’s author.

“It’s no surprise that the demand for AI expertise has exceeded the supply—that’s been predicted for years—but it’s important to realize that it’s now become the biggest bar to wider adoption.”