How can businesses use AI to advance operations?

Artificial intelligence may not replace your job, but someone using AI could.

John Boersma is an AI and machine learning strategy consultant and is the principal at JB Management Consulting. He presented what he called a practical guide to AI for leaders at Wednesday’s Economic Forum meeting.

He said there are three types of AI: generative AI, predictive AI and robotics. While generative AI is often what’s shown in the news, he said, business leaders can really use predictive AI to their advantage.

He said customer churn modeling is one of the top uses of predictive AI, used by businesses to analyze customer retention.

As an example of the modeling, Boersma pulled up a model predicting employee loss at a hypothetical business. The model took into account different employees’ years on the job, last evaluations, superiors and commute times to figure out if the employees were likely to quit the job.

Boersma said subject-matter expertise is essential to getting this kind of modeling right because otherwise, the model would include the wrong data points.

“I'm emphasizing the importance of deeply understanding the problem that you're working on, and that means including the people that really understand the business,” he said.

He said it takes a team to produce an effective AI system, not just one expert.

“I think there's been a tradition in this business to go look for somebody who's got all these skills — and they do exist, but they're expensive and rare and hard to manage sometimes,” he said. “So don't think about hiring a data scientist that's going to solve all your problems.”

He also recommended against waiting for perfection, as the technology will have kinks and results will improve over time.

By looking into the data points and randomizing them, people can dial down how exactly AI is reaching its conclusions, Boersma said.

“Sometimes you'll hear people say, ‘It's a black box. You can't look inside. You can't tell what the system is doing. We're uncomfortable because we don't know how the decisions are being made,’” Boersma said. “I want to challenge that belief. I think that is fundamentally a misunderstanding.”

As for generative AI, he said, people need to be aware that the technology can be wrong. He said in his experience, it’s right 90% of the time.

As an example, he asked AI to generate a statement before the presentation: Write a three-sentence description of the founding of the Economic Form of Albuquerque. AI responded with a generally accurate description but had an incorrect founding date, as one audience member pointed out.

That’s why AI is currently being used in the business industry as a tool, Boersma said.

He said he often uses it himself for scripts or computer syntax. He still checks the AI results, he added, but it makes working much more productive.

“You should not be concerned that AI is going to replace your job or there could be some competition from AI,” Boersma said. “You should be concerned that a person using AI is going to be competing with you.”

Still, large companies that could afford to roll out AI systems for customer use aren’t really doing it yet, Boersma said. He pointed to corporations that still use old-fashioned chatbots based on scripts over AI that could be more conversational and interactive.

Why?

Boersma said customer-facing AI just isn’t mature enough to take off yet, and companies think it’s too dangerous to put out there.

“Sometimes, I say to clients, ‘You have a choice whether you want your chatbox stupid or insane,’” he said, getting laughs from the audience.

He said he thinks the technology could currently be ready for employee-facing functions with training and processes around it, and maybe generative AI will progress further in a year.

“Watch what your market leaders are doing. Watch what your competitors are doing,” he said. “Maybe build capabilities and be ready to roll something out. But for the reasons we've seen, this technology is not ready to be customer-facing.”

He also warned about the biases of AI. He said the technology uses historical data to make predictions, and historical data contains “whatever biases, prejudices, preconceptions, flawed decision-making processes — whatever was in that — in the past.”

“If you just use that data blindly to make predictions in the future, the AI system can then replicate those biases, those unfairnesses, those blind spots in judgment or even amplify that,” he said.

Different audience members after the presentation asked about how people could go about using AI in different fields like health care and education, and Boersma said the technology needs to be monitored and applied carefully to specific situations.

For example, he said, in education, teachers could use AI as a tool rather than resisting the change and only looking at it as something students use to cheat with. He pointed to Wikipedia, which instructors initially opposed but eventually put policies around for student usage.