How APAC CIOs are driving AI projects

Key challenges

While many organisations grapple with data issues in their AI initiatives, this was not the case for SAP and MyRepublic. Saha said the lack of compute power was a bigger bottleneck, particularly when running highly specialised algorithms and models that could take days to generate an outcome.

It was the same case for MyRepublic, which had to prioritise the compute capacities it needed to drive sales and marketing campaigns powered by AI models, for example. Yeo said that to address the issue, MyRepublic is using cloud-based serverless computing services to ease the compute crunch.

Nigel Lim, a regional IT manager at a Japanese multinational conglomerate, said his firm is still in the early stages of implementing AI, as it is running disparate systems across different businesses that make it hard to pull data together.

“To get full value from AI, you need to deploy it at scale, simply because the upfront investment is significant”

While the challenge is being addressed, Lim said he is exploring the use of textual analysis to mine helpdesk information in a bid to improve the productivity of support staff. “If it goes well, we can clean up the data and release it for use by users and chatbots,” he added.

Yeo’s thinking around chatbots differs. Instead of powering chatbots with a knowledge base, his team is working with Google to build a chatbot powered by user stories.

“It can help users troubleshoot network issues by accessing our network infrastructure to see if there is a fibre cut, or if there is a circuit problem in our GPON [Gigabit Passive Optical Networks] equipment, all the way to looking at your accounts to see if you’ve been suspended because you haven’t paid your bills,” said Yeo.

“If there is no problem with all of that, it will assist you, step by step, to troubleshoot your issues. The plan is to drive all support calls to live chat. People increasingly want digital ways of interaction – no one likes to call and we can see that in our user reviews in Australia.”

At SAP, which receives a million IT support requests from employees each year, 10-15% of those requests are now handled by chatbots, said Saha, who emphasised the importance of achieving scale in AI deployments.

“To get full value from AI, you need to deploy it at scale, simply because the upfront investment is significant,” he said. “If you are only doing it in one subsidiary or country, it is unlikely that you will get a good return on investment.”

That said, Saha warned that AI projects do not always deliver guaranteed results based on a certain metric and within a specific period of time.

“But one thing that’s different between AI and other projects is that the returns from AI will get better as the machine gets smarter and more optimised over time,” he said. “Just look at Amazon’s Alexa, which can now handle exponentially more scenarios than when it was first released.”

The need to understand how AI algorithms achieve specific outcomes or make certain decisions has been a topic of interest in academia and industry for a while now, and Saha said SAP has embarked on explainable AI projects to look into the factors that influence AI decisions.

“One area where this is very useful is medical diagnoses,” he said. “Today, there are AI capabilities that can detect cancer better than human doctors can. But we don’t know what the AI sees in an image that tells you it is likely to be cancer or not.”

MyRepublic’s Yeo also struggled to understand how one of the company’s AI algorithms concluded correctly that a staff member had switched religions by going through three months’ worth of web-browsing history. “We couldn’t tell how the algorithm came up with that conclusion,” he said.