Fujitsu Australia shares customer AI use cases

A global study by Gartner reveals that more than half of organisations are piloting generative AI or putting it into production.

Among the myriad companies in Australia exploring the use of AI with clients is Fujitsu, whose Asia-Pacific acting head of data and AI Chris Benson spoke about the risk of sharing personal information with generative AI services like ChatGPT at the recent Databricks World Tour in Sydney.

Benson described how the company is helping a government client use AI.

"We've worked for a government agency where we've needed to detect road barriers that have got defects in them or roads that have got defects in them," Benson said.

"It's about labour intensive processes, generally, when we're helping [customers], and the government [agency], that's image processing and identification so that you don't have somebody sitting there having to look through millions of images to try and identify faults."

Fujitsu is also helping its mining and manufacturing clients use AI for predictive maintenance.

"We typically do predictive maintenance, so that's just standard AI, or what are call traditional AI. That's looking at machinery and seeing where potential failures are within that machinery or in a system of machinery," Benson said.

"If you can begin to help with that predictive maintenance, it really helps uplift the production lifecycle in mining and manufacturing. We've got a number of clients that we do that for."

Opportunities to help customers scale AI

Benson described the type of work customers are requesting around implementing AI.

"If they're new and only beginning to build out their the data platforms, they'll probably be looking to do very simple kinds of AI use cases; traditional AI recommendation engines looking for specific occurrences of data that may pop up here and there, like fraud detection," Benson said.

"Image recognition is quite popular at the moment as well." 

"Then it moves up to companies that have got [data] platforms; they've been dabbling in AI and they don't really know how to scale it."

"So we look at things like MLOps and LLMOps...and helping them to take those models and really scale them to enterprise grade and get them working to a production type environment."

"Because quite often what we're seeing is the person has just got it on their laptop, or they've got it on a small server somewhere, and they don't know how to actually take that model and move it into a production where it gets properly monitored and maintained and versioned."

Benson provided an example of a customer Fujitsu is helping to scale AI.

"We're helping a major car retailer around some of the large language models there," he said.

"They started off with a POC internally...and they've seen that it's great on a laptop, but they now want to take this further and see how they can productionise it, not just have it run off three or four inputs, but run off 20 years of inputs."

"That's looking at taking something that's been a great POC, but they don't know how to take it and scale it. And we're helping them to take that and scale it out and build it out better for them."

Pilots key to understanding value of AI

Benson described how Fujitsu is helping customers set up pilots to understand the value that generative AI can provide.

"A lot of times the customer will have some sort of in house team that will be looking at some of these [generative AI] models already," he said.

"The main reason for that is the subject matter expertise in the area, not around model building, but in the area that you're trying to do the predictions on, is so in depth that you need the subject matter experts to really help you out."

"So we often go in there, we can help the teams identify the right questions, we can help teams work out what is the right way?"

"What are the right models to be using?"

"Is it supervised or unsupervised learning? Is this traditional AI?

"Or is this a large language model problem?"

"A lot of what we do around that initial use case is proving that the use case is going to be useful...because sometimes [customers] don't know what the uplift will be and what the predictability of the model will be." 

"Only once you've done a bit of a pilot and a test case around it, can you prove 'this has shown an increase of so many dollars or you're going to have a reduction in so much customer turnover, or an uplift in production somewhere.’"

"Only after you've gone through those processes do they really begin to understand how AI can help them."