Skip navigation
3D computer rendering of body.png

AI Lag Presents Opportunity to Some Industries, Survey Finds

O’Reilly survey examines revenue-bearing AI projects in production, and how adoption patterns might change in the coming year.

The latest in a series of three AI-focused industry surveys from O’Reilly focused on revenue-bearing artificial intelligence projects in production for organizations, asking how the AI-adoption patterns at these organizations might change in the coming year. O’Reilly says it found that AI use by companies is quickly expanding into deep learning, reinforcement learning, knowledge graphs, and human in the loop, even as half of the companies surveyed said that a lack of data and skilled people slowed down AI adoption.

But don’t be distracted by the information on staffing, said Ganes Kesari, co-founder and Head of Analytics & Innovation at Gramener. “The biggest skill gap reported by companies in evaluation stage is that of ML modelers and data scientists,” Kesari said. Having AI-skilled professionals is great, Kesari said, but not if they can’t figure out how to make those skills work for your particular enterprise.

The Survey Findings

O’Reilly’s new survey AI Adoption in the Enterprise, released earlier this year, drills more deeply into its results from two earlier surveys: The State of Machine Learning Adoption in the Enterprise (July 2018) and Evolving Data Infrastructure (January 2019).

One area of focus was an examination of the preferred AI applications across different industry sectors, including education, the public sector and retail. O’Reilly found that retail, health care and technology are leading the way in AI adoption, and the public sector/government, education and manufacturing are among those lagging behind.

The lag in some sectors could also present an opportunity, O’Reilly argued, pointing to two talks at their recent AI conference that focused on how AI applications can provide a technological safety net for small businesses. Applications focused on finding anomalies have clear applications for education (in analyzing data from standardized testing, for example) and the public sector (in identifying possible fraud or security breaches).

O’Reilly also found that the AI tools with which IT professionals should familiarize themselves are likely to be different depending on their industry. Spark-NLP and H2O are popular in finance, the survey indicated, while Google Cloud ML Engine has made inroads in health care. TensorFlow is still the industry leader in AI applications, but PyTorch and Keras are closing that gap.

Challenges Ahead

The growing variety of AI tools, and their potential uses, is actually part of why AI adoption in the enterprise has been slow, said Bill Galusha, director of product marketing at ABBYY.

“There are dozens of AI products but they aren’t easy to use,” Galusha said. “No one wants to hire a team of data scientists to build and integrate data models--or depend on having that ‘unicorn’ employee who possesses both technical and business depth to recognize AI use cases.”

If enterprises could find easier ways to deploy AI, the technology would become more widely used, he said.

It’s important to note that of the organizations surveyed by O’Reilly, four-fifths were already using artificial intelligence. “The survey isn't reflective of the current industry maturity levels in analytics,” Kesari cautioned. “With over 81% of respondents from organizations with a mature practice or with POCs, the base is skewed toward an analytics-savvy audience.”

But, based on what those organizations already know, according to O’Reilly’s survey, there are learnings to pay attention to. These include working to identify use cases for AI for your specific company and its culture; looking ahead to challenges posed by a lack of data and/or skilled staffers; ensuring your AI systems aren’t just optimized to business metrics but are also reliable and safe; not underestimating your infrastructure needs, now and for the future; and thinking beyond deep learning.

Ultimately, Galusha said, within the enterprise, the decision maker (buyer) and user for AI has changed. The business user is now identifying and driving use cases, he said, not IT.

“For AI to be mainstream within the enterprise, the line-of-business user needs out-of-the-box cognitive skills they can quickly deploy across multiple projects,” Galusha said. “The enterprise needs AI solutions that empower more employees, not just two people who are trained in it.”


Hide comments


  • Allowed HTML tags: <em> <strong> <blockquote> <br> <p>

Plain text

  • No HTML tags allowed.
  • Web page addresses and e-mail addresses turn into links automatically.
  • Lines and paragraphs break automatically.