This AI technology could supercharge political campaigns

Written by Anna Whyte for The Post

Imagine, as a business owner or the leader of a political party, being able to interview a focus group, including with specific follow-up questions, and then have a summary of findings - all within the space of a few hours.

That's exactly what David Talbot, a director of Talbot Mills Research has done, and the AI technology - which only became possible less than two years ago - has proven so useful he’s in the process of commercialising it and selling it around the world.

It's not yet the gold standard, but it is the kind of technology that could see researchers and polling groups spending hours, rather than days or weeks, interviewing and then writing reports.

The development of artificial intelligence will be a game changer for the polling industry, says Talbot, with new companies opening the door to fast, accessible polling - only possible due to recent technology advances.

Polling is used frequently by businesses in the development of products or advertising campaigns, and political parties and organisations on all topics from particular policies, the popularity of politicians and party support.

Talbot Mills Research, a well known polling company in New Zealand, conducts Labour’s internal polling and also does political polling for its corporate clients. It works with businesses in New Zealand and Australia, and across the ditch worked with Anthony Albanese and Dan Andrews.

On qualitative polling - essentially asking focus groups for their thoughts on particular topics including asking them targeted questions - Talbot wondered a couple of years ago, “could we train an AI to interview people the same way I would interview people?”

With the advent of generative AI, like ChatGPT, Talbot said he realised that, particularly on the qualitative side of polling, “that technology might be able to really supercharge that side of research”.

“And so we trained an AI and we prototyped a little system, and it turns out it works mind-blowingly, amazingly.”

The result is a new company, qualifyr.ai, which “was only possible in the last 18 months because of the changes to the availability of generative AI infrastructure”.

Talbot said it was scaling up fast, in part driven by the Australian election and the need for campaigners to understand public sentiment.

For example, in Australia, it had been used to ask people how they feel about topics such as climate change, “and they're getting results back in hours, not weeks”.

The use of AI could also open doors to smaller groups - like NGOs, sole traders or small businesses - that may not have the money to spend on qualitative polling and quantitative polling, where it can indicate a mood from a sample size at a moment in time.

“Fast moving consumer goods is probably the most likely growth area for these kinds of products.”

He used the example of a new small business owner wondering if whether a particular logo would work well.

That person could upload their slogan and logo, some suggested questions which the system could help with, “And it would say, overall, how do you feel about this logo, or how does the logo make you feel? Can you see any problems with the slogan? Do you think the slogan is memorable?”

It can then write up a summary of 16 interviews and email a report.

“And it all happens in about five hours.”

The questions change depending on the project.

“So the questions we would ask on a political project would be very different than if we were working for a power company or airline.

“What we've programmed the AI to do is to start with a core set of questions, but it's now flexible enough, based on all our training, to investigate other options.” That means it could ask follow-up questions.

Talbot is also the founder of another start-up, six years-old, called Stickybeak which he describes as a self-service platform.

“It's much more targeted at helping people make a decision about something, whether it's a packaging decision or a creative execution or logo or slogan. Making that quantitative process much more agile.”