How Genesis staff are trying out generative AI within a pilot programme
When ChatGPT first burst into the mainstream just over a year ago, Genesis staff were as keen as anyone to try it out at work.
“We just cut it off. We had to look at it through a risk-based lens. It just came out of the blue,” says Genesis future energy general manager Steph Creasy.
But she also saw that as a temporary response.
“We couldn’t just ban it. We’d miss the opportunity to leverage the technology. When Microsoft offered us the opportunity to join the Copilot early access programme, we saw our chance to explore what generative AI could do, while safely managing the risk,” Creasy says.
After investing north of US$10 billion ($16.8b), Microsoft is now ChatGPT maker Open AI’s largest backer — and its Copilot AI piggybacks on ChatGPT’s smarts.
The free version of Copilot can be used for the likes of summarising a document or website or creating an image.
A pro version, which costs $37 per user per month, will work with
Microsoft apps like Outlook, Word, Excel, PowerPoint and Teams.
And, crucially, it offers commercial data protection. While it can be useful to, say, ask a generative AI to help edit the draft of your 10-page presentation to five slides, your data can also go into the public mix, or be used for AI training.
With its Pro product, Microsoft says: “User and business data in Copilot is protected and will not leak outside the organisation. Chat data is not saved, and Microsoft has no eyes-on access to it. Chat data is also not used to train our underlying models.”
You also get priority access to Chat GPT 4 (the version fed with up-to-date data) during busy times.
Months before Copilot’s official release, in November last year, Genesis became one of the first five organisations in New Zealand — and one of only a handful in the world — to trial the product. Genesis did not get it for free as a pilot customer, but did get priority support and other perks, Creasy says.
An hour a week in time savings
An initial, limited trial involved around 30 Genesis staff. “We wanted to flush out any challenges or ethical risks,” Creasy says. It’s now been expanded to some 300 employees (of a total complement of about 1300). Creasy says 70 per cent of staff on the pilot are saving at least one hour a week from using Copilot, while some power users are shaving off up to five hours. She says Genesis has used Copilot for:
• Making transcripts of meetings, then summarising them — freeing up people to focus on engaging with each other rather than note-taking.
• Reviewing drafts. The AI has been like “an always-on, creative colleague” for back-and-forth soundboarding, Creasy says.
• Setting the right tone for communications. For English-as-a-second language staff, Copilot can flag if the language is too formal for a coffee meeting. And English-as-a-firstlanguage staff can get a nudge about the tone of their emails “when it’s been a long and busy day”.
• Sifting through a long email conversation to summarise key themes and action points.
• A “diversity audit” of the content and imagery on a Genesis website.
Genesis has also created three inhouse AI tools, one for analysing call centre transcripts, summarising individual calls and looking for common themes; another to create a knowledge base for the firm’s powergeneration sites; and a third, still largely under wraps, that’s dealing with “multi-agent data”.
Staff have used Copilot to speed up some reasonably gnarly tasks.
Many of the broad themes aren’t new. Ditto business intelligence of “BI” systems, designed to sift through corporate data and highlight trends. And Creasy says Genesis — like most large companies — has been using machine learning (a type of AI) for years. But up to this point, it’s been the realm of data specialists. The gen AI revolution of the past 18 months has democratised the technology.
Her case in point: In the past year, all energy companies in New Zealand have shifted away from the previous early payment discount system, to comply with government directives.
Genesis has now changed its system to reduce costs for those with multiple bills or who are paying via direct debit. Finance analyst Andrea Crasta — a finance grad who had only
been with the firm a few months — was given the job of ensuring all these discounts were accruing correctly to customers — something that involved “interrogating” more than 100 different data tables.
“It broke everything down line by line, and I was able to manage everything myself in less than a day. It’s incredible — now I can write my own queries.”
Can you trust the results?
One possible fish-hook is the trustworthiness of generative AI, which is occasionally prone to “hallucinations” or making things up as it sometimes reaches the wrong conclusion.
The Herald recently profiled a Bell Gully alumnus who raised $8m to build an AI that creates legal contracts.
Asked about hallucinations, he said the AI’s work should be verified, just as a senior colleague would check a junior’s work before sending it to a client.
“I’d argue that humans make mistakes too,” Microsoft NZ country manager Vanessa Sorenson says. She adds there are tighter parameters with tools like Copilot. “It’s their data within their business. [Copilot] has a full, auditable trail about where it got the information from — and that to me is far better than a human grabbing something that could be wrong.”
Creasy adds “The use of these tools has forced us to go a bit faster on some of the tidy-up and structuring and classification of some of those data sets that we’ve got, because we need to be comfortable that we’ve got the right governance over the source data” to ensure it’s only accessed by Copilot users with the appropriate level of permission.
What’s next for Genesis and AI?
Creasy says the power company could use it for forecasting.
“I see the next phase of our strategy [asking] what are those really big value opportunities and how do we apply AI to those?” she says.
“Every organisation needs to think through which of those pieces that they’re going to build for themselves — because that’s a source of competitive advantage — versus where they’re going to take the tools off-theshelf like from Microsoft or Salesforce or whoever it might be.”