To get the most out of deploying AI, identify the best use cases first

Sometimes, deploying AI is like an “As seen on TV” product. Take light-up shoelaces (bear with us.) They’d certainly be useful if you frequently...lose your shoes in the dark. But if you manage to keep tabs on your trainers? You’re stuck with something you paid for that you won’t use.

The same goes for AI: Before deploying it, identify a use case. That’s the biggest piece of advice Morning Brew got from experts when we asked them what IT pros need to do to ensure their cloud platforms are ready for the data storage and analytics that AI entails. Without a thought-through use case (and AI-ready data), you risk committing AI—and its associated energy and costs—to questionable projects that might not benefit larger goals.

“Workload optimization goes down to even the granular task, job, application, and user level,” Linda Yao, VP and general manager of Hybrid Cloud & AI Solutions at Lenovo, told Morning Brew. “That actually allows the organizations to ensure that they are using that precious compute [power] and using that precious technology on things that have a chance of succeeding rather than churning through those tokens.”

On time

Once IT pros have a use case, the next step is evaluating the data that AI will work with. Mike Ni, VP and principal analyst at Constellation Research, told us that when it comes to deploying AI, “you don’t need the biggest cloud platform. You need one that can connect to your data, analytics, and AI tools.”

To accomplish that, Ni said, IT pros need to locate the data and ensure it’s clean and accessible. On top of that, AI tools need to be able to reach data at the necessary speed.

“If I need [data] because I’m in a call center and I want [AI] to tell me something about the customer who’s calling,” Ni said, “then the accessibility has to be in a semi-fast mode to get that data back.”

There’s another concern when it comes to connecting AI tools and relevant data: The freshness of the data itself. If data isn’t up to date, it might not be relevant.

“Sometimes the data gets old. It could be correct, but it was updated a month ago,” Ni said. Or, “this data is correct, but it’s not connected properly to that customer.”

Trust issues

To make the most of AI tools, IT pros need to be able to trust the data they’re using. That, Ni said, falls along two different axes.

On one level, data needs to be trustworthy. If it’s not, that’s not something an AI tool will be able to distinguish, because it runs off the documentation it’s given.

“A human would have noticed that data is wrong, and therefore not actually pull that into a report or use it to make a decision. A machine has no idea,” Ni said. “An AI is just going to use it because you said that was the data to use.”

The trustworthiness of data also pertains to whether or not it’s secure, Ni said. If data contains sensitive or personally identifiable information, IT pros need to manage access to it and make sure it can maintain its own governance.

Overall, “AI needs to get a job, not a playground,” Ni said. And that job requires measurable outcomes—all of which may be possible on existing systems.

“When you start the process and you actually have a measurable outcome, you start iterating,” Ni said, “and that’s where you start really seeing some really interesting unlock.”