The aim is for Enterprise AI Assistant to serve as the single knowledge base for an organisation, enabling it to deploy an ecosystem of AI agents to help with tasks in various functions across its business.

Eating their own dog food

A pilot has been running with a large Government customer, plus two firms in the private sector.

Datacom has also applied its product to its own systems, including the creation of a payroll knowledge assistant to use with its Datapay product.

“That’s proving to be really successful,” Compagnone said.

“It’s helped our payroll professionals to answer really complex payroll queries and it’s accurate to the point it can pass the payroll exam to 93% accuracy.”

What about the other 7%?

“Well, this is the thing. As with any large language model, it doesn’t necessarily complete a task at all times, but it will get you close enough,” Compagnone said.

“It’s augmenting their [payroll staff’s] capabilities rather than replacing it. So, for example, a really complex payroll inquiry about some sort of maternity cover might have previously taken hours of trying to navigate the legislation to give the right response. And then you have to write an email.”

Now, the legislation check and a starter email can be delivered by the AI Assistant almost simultaneously. “You still need to check it and ensure that it’s right, but it will reference all of the legislation.” Summarising complex regulations, with sources cited for easy cross-checking by a human, is a key use for the Enterprise AI Assistant across various areas, she says.

What we do in the shadows

Another barrier to AI updates has been ambiguity about where data resides, and whether it can be used to train AIs, and the levels of attendant privacy risk.

“Almost every customer that I’ve talked to is very suspicious about that,” Compagnone said.

“It’s sometimes blocking some from even trialling anything themselves, and it’s definitely stopping them from making these tools available to their teams.”

Enterprise AI Assistant can help ring-fence queries to data that resides inside your organisation.

Datacom was also working toward sovereign data capability for those who need to ensure everything stays within our borders and directly within New Zealand legislative oversight.

Flexibility was another watchword.

“Other AI assistants are very fixed. You have to use a specific large language model and a specific platform,” Compagnone said.

“Ours is more like a pick-and-mix, where you can use any large language model you want to. It’s LLM-agnostic.

“And it can sit on your preferred platform, which might be AWS, Azure or Datacom.” (Datacom, easily the largest homegrown IT services outfit with its more than 6000 staff, operates four of its own data centres in New Zealand, and two across the Tasman in partnership with AirTrunk.)

“You can also choose whether it’s managed in-house by us at Datacom on a subscription service model,” he said.

READ MORE: My Net Worth: Datacom AI director Lou Compagnone

Chris Keall is an Auckland-based member of the Herald’s business team. He joined the Herald in 2018 and is the technology editor and a senior business writer.