How to scale AI across the whole company

Earlier this month, a KPMG survey of 100 CEOs found that 39% expect to scale an AI pilot companywide this year. That number marks a big change from November, when just 14% of business leaders said their company was ready to use AI across the organization, according to a Cisco survey. 

However, there are naturally a series of technical challenges that come with the company-wide adoption of one of the most cutting-edge technologies of the world. And once that gets ironed out, companies still have to train a possibly anxious workforce to use an AI tool some of them fear could take their job. 

“It’s not the technology which is the problem, it’s the scaling,” said Amy Challen, Shell’s global head of AI, at the Fortune Brainstorm AI conference in London

Companies of all sizes—from startups with a few early employees to Fortune 500 giants with workforces that span the globe—are grappling with how to use AI. The companies best poised to succeed in the AI world won’t just be the new innovators that invented the technology but the existing firms that succeed in scaling it across their entire business, according to a McKinsey report from January. 

The risk of being left behind is too unpalatable for business leaders, and one mantra was clear as founders and tech executives discussed the topic onstage in London: companies have to pick the right uses for AI. 

A shotgun approach that tries to use AI for everything won’t work. In order to execute on that targeted approach, AI pilots can’t be manned by IT teams alone. “The key for me is agile scaling where you always have a joint team,” Challen said. “It’s always a technical team, but also a business team. Never do data science without the business.” 

At Shell, AI is already being used for what Challen refers to as predictive maintenance, which lets the company forecast where and when its oil machinery might need repairs. 

For companies that know they’ll use AI across virtually all aspects of their business, the challenge is getting their employees used to the feeling of being shadowed by AI, according to Sachin Dev Dugal, founder of the startup Builder.ai.

For example, Builder.ai employees might have an AI assistant manage their calendar for them, while those in the sales department have one give them advice during calls with customers. That in particular took some getting used to because the idea of being listened to by an AI system was “very jarring,” Dugal said. 

“You need to get people comfortable with the fact that there’s an AI on the call,” he said during the panel. 

The panelists agreed companies shouldn’t force AI down their employees’ throats, especially if they’re finding it hard to use. If companies are finding that a new AI tool—whatever its specific use case might be—isn’t being used across the company, they should ask themselves why,  Challen said. 

“If it’s not being used, it’s your problem—it’s not the user’s problem,” she said.

Hesitant employees can be a significant hurdle for getting AI pilots implemented across the company. In general, workers are open to trying out AI, but doing so can lead to mixed results. Some find that it’s undoubtedly a productivity booster, which tends to give them confidence going forward. By the same token, they’re also well aware that AI seems able to do part of their own jobs, which can give employees pause about using the technology. A PwC survey of 54,000 workers found that 13% of respondents had that fear. In fact, at least one study found employees who used AI more ended up being more fearful that it would replace them in the workforce. 

For now though, the most common refrain about AI in the workplace is that it will be a copilot to workers’ everyday responsibilities. Major tech firms like Microsoft, Google, and Salesforce have already started marketing many of their AI tools as productivity enhancers, framing them as the next evolution of the word processor or the spreadsheet. This time around though workers find that AI doesn’t just help them with a task but does the whole thing for them. And that presents its own learning curve. 

“It’s more of an education to get people emotionally comfortable with the idea that they’re one step removed from how they were running a particular task,” Dugal said.