NEW YORK — Business leaders see AI as a driver of business innovation, the type of technology that can lead to business evolution — and new revenue streams — if adopted or built fast enough.

As a response, more businesses are working to deploy AI at scale, seeing it less as experimental and more as a broadly accessible resource in their IT toolkit. The ultimate aspiration is that technologies such as AI can improve day-to-day operations, reduce costs, and enhance customer service and satisfaction, according to a Snow Software report.

"Artificial intelligence is starting to be pervasive," said Nitin Mittal, principal and US AI strategic growth offering consulting leader at Deloitte, speaking Wednesday at the AI Summit in New York. The technology is "pervasive as it relates to being embedded in every business process, in devices that are proliferating, in applications that are being deployed."

Eighteen months ago, prior to COVID-19, three-quarters of companies were experimenting with AI, according to Deloitte data. Today, 54% of organizations have now moved to the next phase, which is scaling AI their practices.

As AI matures, it is outgrowing its reputation as "a distinctive technology" for enterprises. But it also means adapting enterprise processes toward sustainable AI practices. 

Here are Deloitte's three key pillars toward scalable AI practices: 

• Cultivating data

Garbage in, garbage out — as the frequent AI refrain goes. The data pipeline feeding models can make or break enterprise initiatives. 

"I don't think this is an unfamiliar problem with any organization," said Dominic Rasini, senior manager, strategy and analytics at Deloitte, speaking Thursday at the event. "As a matter of fact, I would say that this problem has been around for quite some time."

The maturity of the technology is putting a sharper focus on the quality of the data, Rasini said. New capabilities and use cases are available to organizations that can use data as the lifeblood to power them. 

Data quality is a priority for businesses in the years ahead. Nearly half of companies are planning to improve data quality and processing by 2022, according to a MIT Technology Review Insights report.

• Internal operating models

Equipped with clean, actionable data to fuel their AI aspirations, the next step for enterprise leaders is to successfully turn it into insights. ML Ops, a set of practices to reliably and efficiently run machine learning models, can help companies meet their AI goals, according to Rasini.

Companies that shift their IT operations to a DevOps management and "start to embed an ML Ops approach, are two times more likely to succeed in their AI agenda," said Rasini. But MLOps alone won't cut it.

"A very important facet of that is to understand that this is all part of a broader ecosystem of capabilities, not a bunch of silos trying to work independently," he said.

• Talent supply and demand

Once businesses decide how to build AI capabilities and which data should power it, leaders often grapple with the question of whether to buy or build.

"All organizations will have their own levers that need to be pulled in order to be able to make these decisions that are right for them," said Rasini. 

Whether companies should build products internally or externally can depend on factors specific to their goals and type of industry, he said. A life sciences company, for example, might address the question differently than a retailer focused on mobile commerce.