Progressive Insurance assessments limits of AI-generated commercials – and learns when to tug again

This wasn’t a press unlock masquerading as a case learn about. It used to be a glimpse right into a are living experiment – one who produced 96 audio advert variants in simply two weeks, greater quote begins by way of 31% and spread out fully new questions on how a lot personalization is an excessive amount of.

“We were the first to apply GenAI to write the scripts, create the voices and the music beds – all of it,” says England, who leads technique, analysis and analytics at Progressive. “We normally produce a handful of audio ads over seven weeks. This time, we had 96 made and approved in 14 days.”

Claritas, Progressive and the Consortium for AI Personalization

Claritas, a advertising intelligence corporate with roots going again just about 5 a long time, performed a central function within the procedure. Known for its industry-standard id graph and segmentation equipment, Claritas has an increasing number of moved into AI-driven personalization. It got patented AI tech to generate and optimise inventive messaging in genuine time, permitting manufacturers to pair the appropriate message with the appropriate particular person – or character – dynamically.

Briggs, who in the past helped pioneer virtual advert size, cellular effectiveness and social media analytics, got here out of retirement to sign up for Claritas. “I was supposed to be in Tahoe,” he informed us. “But I read the patent, saw the potential and realized this was the biggest shift in marketing I’d ever seen.”

Together with MMA Global – a business affiliation representing CMOs and advertising innovators – Claritas and Progressive ran the mission throughout the Consortium for AI Personalization, a analysis initiative the place manufacturers check real-world packages of AI and percentage what they be informed.

Why audio? Why now?

Progressive selected audio now not as it used to be simple, however as it used to be possible. “At the time, generating realistic video using AI wasn’t quite there,” says England. “But synthetic voices and music were already good enough to put in front of customers.”

The first marketing campaign introduced on SiriusXM, then expanded to Spotify, iHeart and different main streaming platforms. The device allowed Progressive to change script, voice and song background in genuine time. The consequence: six scripts × 4 voices × 4 tracks = 96 combos, every tuned to resonate with other segments.

“We even talked to synthetic personas – built from decades of Claritas data – to ask what kind of music they preferred,” says Briggs. “They told us country music was popular among likely converters. That wasn’t one of the original beds. So in the next round, Progressive added it.”

The personalization paradox

The effects have been obviously sure. But now not absolute best.

“What we found is that 96 versions were maybe too many,” says Briggs. “There’s an optimal ratio – what we call the learning-to-earning balance. If you only have two people converting from 1,000 impressions, you don’t have enough data for the AI to learn from 96 variants.”

This is the place Briggs provides a dose of realism continuously lacking from AI discussions: the so-called 5 nines of accuracy. “Even if your model is 99.999% accurate, that’s still one wrong result in every 100,000 decisions,” he explains. “Now imagine you’re delivering 40m ad impressions. That’s 400 errors and, in regulated industries like insurance, that’s 400 too many.”

Hence, the significance of a human within the loop. Progressive’s inventive director, Chris Monaco, reviewed and filtered 126 GenAI ideas to settle at the ultimate 96. Every variant ended with the similar, legally required disclaimer, voiced identically. “We didn’t cut corners,” says England. “Compliance was baked in.”

Efficient, sure – however efficient too

The numbers talk for themselves. AI-enabled campaigns delivered a 197% elevate over the keep watch over staff. Quote begins jumped 31%. GenAI-enhanced audio commercials tripled publicity and stored 98% of listeners engaged via to the tip.

And most likely maximum usefully, Progressive found out a brand new framework for speedy experimentation – a approach to check hypotheses about tone, message, be offering or supply with out ready months for insights.

“It compressed the whole cycle,” England says.

Looking forward: video, e mail, brokers

Claritas is already increasing this fashion to different media, together with generative video and e mail. Briggs sees a near-term long run by which 20% of manufacturers run absolutely AI-orchestrated, multichannel campaigns – customized in genuine time, anchored to first-party knowledge and optimized regularly.

But even he recognizes there are limits. “Until the models are flawless, humans still need to review what goes out.”

England has the same opinion. “AI doesn’t replace marketing,” she says. “It just changes it. And if you’re a marketer, that means learning how to use it – now.”

Advice for manufacturers nonetheless staring at from the sidelines?

“Don’t be intimidated,” says Briggs. “But do get involved. The gap between those who experiment now and those who wait is going to be very difficult to close later.”

England’s recommendation is similarly blunt: “Don’t think AI is someone else’s job. It’s everyone’s job. And it’s moving fast.”