- One-quarter of companies report a 50% failure rate with AI projects, according to IDC's Artificial Intelligence Global Adoption Trends & Strategies report.
- AI models require trial and error periods preferably fueled by high quality and high volume data, Ritu Jyoti, program VP of artificial intelligence strategies at IDC, told CIO Dive. But tech industry's favorite "fail fast" strategy cannot apply to AI because companies rack up costs and diversions.
- More than 60% of organizations changed their business models to adjust to AI adoption, according to the report. Next-generation business models, like those belonging to Netflix and Amazon, are only possible because of AI, said Jyoti.
The AI solutions the Amazons and Ubers of the world are investing in are the ones that offer open source frameworks, edge computing, facial recognition, predictive maintenance and e-commerce searches.
A company that uses AI at its foundation "refers to companies that have a business model where the company would cease to exist if AI did not exist," said Jyoti.
Companies that thrive go toe-to-toe with existing competitors as well as next-generation competitors, giving them an advantage when developing end-to-end customer experience. "This is a serious competition based on a transformative business model," she said.
If a company's business model is outdated, unable to support AI projects, "you cannot compete," according to Jyoti. Aligning AI strategies with a new business model, in turn, opens companies up to being on the positive side of disruption.
Still, nearly one-third of businesses say AI applications lack practicality for their business. Conservative implementation and experimentation is often influenced by shallow talent resources and a shortage of AI building blocks.
However, companies don't necessarily need to reframe their entire business model around AI, according to Jyoti. Companies can remake models by department or at the line-of-business level.
"The change and cost impact will also vary from one company to another," she said, likely contributing to the emergence of AI classes and adoption priorities. Experimental AI applications have low adoption rates, like back office automation, synthetic training data and GANs, a type of neural network.