AI bills can blow out by 1000 percent: Gartner

Organizations adopting AI need to learn how to manage the emotional and monetary costs the tech creates, while also worrying about capturing productivity benefits, according to analyst firm Gartner.

The firm offered that opinion today in Australia where it kicked off the first of its flagship Symposium events for the year with a keynote from distinguished VP analysts Mary Mesaglio and Kristian Steenstrup.

"It is really easy to waste money on generative AI," Mesaglio observed, before asking attendees if they could remember their first cloud computing bills.

"The reaction was probably shock, then confusion about what you were charged for," she suggested. The same is likely to happen for AI bills, she warned, noting that "500 to 1000 percent errors of AI cost estimates are possible."

Vendor price hikes are one reason for such blowouts. Simply not paying attention to the cost of using cloudy resources will also bite AI experimenters.

Inappropriate use of AI is another problem: Mesaglio said orgs need to learn when search will suffice instead of using AI for simple queries. AI can also become a victim of its own success - users enjoy it and start writing longer and more complex queries that cost more under token-based pricing schemes.

The cost of queries or AI workloads can also change depending on how orgs structure their AI's access to data. Allowing AI to work on unstructured data may find more info and produce better results, but is an expensive approach. Centralizing data and having it tended by IT departments has its own challenges.

Successful AI can also create other problems.

Mesaglio described one as "productivity leakage." Citing Gartner research that workers using AI can save 43 minutes a day, she asked if employees would use that time to work harder?

"I would use the time to get a [café] latte," she said, suggesting many other would do the same and create "productivity leakage" that sees ten to thirty percent of AI's benefits drain away.

The two analysts also pointed out that the productivity benefits of AI do not apply equally, by using the example of two call center workers: one a new hire and the other with five years' experience. The new hire, the analysts suggested, will quickly enjoy a big boost to productivity from adopting AI because it surfaces info they need to solve customer queries. The old hand's experience means they won't benefit from the insights AI offers - because they already know.

The more experienced worker, the pair suggested, may later come to resent AI if it means an AI-assisted worker with less experience quickly performs at the same level.

That negative reaction, the pair suggested, will likely be found across organizations because AI by its nature will generate emotional responses to technology.

Organizations will therefore need to consider who has responsibility for considering how AI impacts people.

The two analysts suggested that as orgs consider AI, they should consider if they need a "steady" or "accelerated" pace of adoption.

Organizations in industries that aren't being reformed by AI probably have ten or fewer proof of concept projects and can afford the steady approach. Other orgs that want to lead with AI - either to seek advantage or to keep pace - need the "accelerated" stance.

The good news, the analysts opined, is that all orgs are running their own race: vendors are speeding ahead to try to capture the market, but users can choose to ignore their sprint and move at the pace that best suits their needs. ®