AI delivers innovation for KPMG

It started with the question, ‘What does AI mean for us?’. A year in, KPMG’s KymChat has become a valued tool in the business.

KymChat is a proprietary KPMG solution that uses Microsoft ChatGPT technology in a private, secure manner. The firm launched it with the aim of promoting innovation, boosting efficiency and creating a better experience for staff and customers, says Cowan Pettigrew, KPMG New Zealand’s Chief Digital Officer (CDO).

There was the awareness that staff may well be using AI at work – Microsoft has estimated 25% of people already do – and it was considered safer to manage that experimentation in a controlled way, Pettigrew says. 

“I would suggest a ‘blocking strategy is increasingly problematic. Very soon, AI powered solutions will become ubiquitous.  If you want to enable your business to thrive and grow in a digital-enabled manner, you must take the time to understand what AI technology can do,” Pettigrew says.

“KymChat was an opportunity to explore AI in a safe, secure, and transparent way."

Initially, the firm saw staff enjoying KymChat, but that quickly evolved into practical uses, says Alistair Evans, Digital Director at KPMG.

“One thing that surprised me is just how quickly it's gone from, ‘oh look, it can write poetry’, to people without technical skills applying AI in sophisticated ways and coming up with these amazing use cases,” Evans says.

“A year ago, they wouldn't have thought about it. They just didn't appreciate what it could do. With exposure and the ability to use these tools hands-on, our people have become very sophisticated in their AI use and application.”

There were a few amusing trends during the year - a similarity in the style of employees’ goals for the year may have been down to the use of KymChat, Pettigrew speculates.

Staff also learned how KymChat could enhance productivity when managing a high volume of emails by automating certain tasks such as filtering, prioritising and responding to emails.

“We saw an explosion in longer, detailed, emails that, clearly, some of our younger people didn’t have time to write,” Cowan says.

“But it quickly became more focused. I got an email from a partner that said KymChat has enabled her to, generally, finish on time and get home to her kids. I can’t give you the exact time-saving data behind it, but that is a real-world outcome from somebody who turned around and said, ‘holy moly, I’m getting home on time. I’m getting to see my kids at dinner time’.”

A big part of the learning for KPMG’s staff was how to “prompt” AI to deliver the results required. This is a key skill for achieving the potential benefits of the technology and one that is learned through practice. One lesson was don’t simply accept the first draft of AI material – keep refining it through further prompts until it looks right.

“There are individuals who have mastered prompt engineering faster than others around them. They were able to lift their productivity much faster than the rest of the team,” says Evans.

The AI project was managed with data security as a main focus requiring in-depth analysis of what data was available to KymChat, and how it was being used, Pettigrew says.

“We had to understand the entire data loop. So, from a risk privacy perspective, we track what goes in and what comes out - by user, by time, and by date stamp. At any stage, we can see where we’ve used AI, or aggregates of AI here, here and here.”

KPMG also ensures a human presence in the KymChat loop, although not at every stage as that would undermine productivity gains.

“Something that’s been quite important in our journey is helping people understand how it works and what it does. It’s like any trusted information – from another person, from a news website, from anywhere,” Evans says.

“With KymChat, the sources are referenced so you can see where it’s actually got your information from. You’re enabled.”

Governance was a key concern as the project was developed and there was a lot of communication between management and the board, Pettigrew says.

“The interaction between management and the board, as this tool was developed and rolled out, was constant. So, the board had confidence that we knew what we were talking about, and in the use cases that were being developed. It didn't take long before it got to the point where it moved into production and the oversight of our executive committee. We were developing for safety, and we built governance controls into the process itself.”

Pettigrew concluded: “It’s really important to remember that the role of AI is to enhance our productivity. AI doesn’t help our clients figure out their issues. Our people are the experts.”