In a world overrun with LLMs, SQREEM wants credit for not having one.
On Wednesday, the behavioral intelligence company appointed Stephen Yap as its new CEO to help make that case. Yap joins from Perion, where he spent roughly a year and a half as CRO after a nearly two-decade-long tenure at Google.
Yap’s stint at Perion came at a time when the company was looking to improve its efficiency and combine a number of tech tools into a single platform. “We were able to do that,” he said – and so it was time to move on.
The number of AI companies that have “spawned” in the last six months or so is “remarkable,” Yap told AdExchanger. He was eager to hop into the thick of it.
Part of what attracted him to SQREEM is that it offers “something that the industry’s never seen before,” as he put it. Not the first time we’ve heard that one.
But, to his credit, this claim has some legitimacy to it. After all, how many AI startups these days market themselves as not using machine learning?
And that’s the tea
Rather than relying on machine learning, SQREEM’s large behavioral model (LBM) uses “state-space models,” which are a type of mathematical AI model that look at certain variables and track how systems change over time. It’s also an entirely distinct type of model from a large language model.
(LBMs can contain machine learning or large language models, but neither is an innate part of the model’s architecture.)
SQREEM’s LBM auto-generates audience personas by using state-space modeling to surface correlations between various behaviors.
The platform isn’t looking for “what is written or what people say,” Founder René Raiss told AdExchanger. “We’re interested in what they do.”
For instance, said Raiss, a shockingly wide range of people consume green tea. It appeals to folks interested in skin care, as well as those who are looking for coffee alternatives.
Plus, he added, “it’s got a vegan ‘Eat, Pray, Love’ component. The yoga crowd loves it.”
If you follow the behavioral thread far enough, you’ll eventually see that people who buy travel insurance purchase green tea 22% more often than the average person. The model can capture the progression of a search journey from green tea to antioxidants to skin care to sunscreen to vacation to travel insurance.
By moving only a few “degrees of separation” from the original behavior, said Raiss (in this case, looking up green tea), the model can determine “somebody’s life story.” (Perhaps that’s overstating matters, unless your entire life revolves around green tea, but the point stands.)
VeggieTales (and VeggieTruths)
But where does all this data come from?
SQREEM aggregates real-time data through connections into social media platforms, as well as by looking at publicly available open-web data and open-web traffic.
None of the data can be traced back to an individual, said Raiss, but the platform can track public perception in real time, such as engagement patterns and behavioral shifts before and after a particular event. It also runs sentiment analysis on social media comments to measure what people are actually saying, not just how many of them are saying it.
For instance, during the peak of the Cyclospora outbreak this summer (or, as Yap so ominously referred to it, “the lettuce issue”) one of SQREEM’s grocer clients noticed that its profits were dropping – and not just related to lettuce.
SQREEM queried its LBM to understand who was engaging in conversations around Cyclospora and what the main topics of those conversations were. Brands can connect their own internal agent or dashboard to SQREEM, Yap said, “so we’re part of their infrastructure.”
In response to the Cyclospora query, the model surfaced three audiences who engage with the topic most frequently, Yap said: elderly people, parents with newborns and people with preexisting health issues. Most of the conversations they had were questions about how exactly the parasite spreads.
One of the most interesting findings, Yap said, was that people were unsure whether the parasite could spread from lettuce to, say, broccoli if they brushed against each other in the produce aisle.
Since most of the concerns people had were information based, said Yap, the grocer realized that rather than steering clear of Cyclospora conversations, it should be “embracing” them. The grocer pivoted its strategy to communicate better with its audience about what was going on with lettuce, cyclospora and washing, Yap said.
At the end of the day, he added, a marketer’s main goal is to understand exactly “who their customer is” – not just their identifying traits but what sort of questions they ask and what qualities they value in a product.
That’s the appeal of an LBM, Yap said: its ability to understand consumer behavior in a way that indicates to marketers exactly what messaging will resonate – within reason.
For example, said Raiss, the model can detect “numeric entanglements” that point toward very high probability outcomes, and it’s usually right. But what it can’t do is predict pure random chaos. If “you’re about to become president, get eaten by a shark and [be] hit by lightning on the same day,” he joked, even the best LBM isn’t going to see it coming.
Some outcomes are so improbable no model could predict them.
