Handouts are not uncommon at industry events. But at Mutinex’s Marketers & Money, attendees might have been surprised when an off-white, almost cream piece of A4 was distributed featuring a patent number, two algebraic formulas and phrases like “Gaussian kernel”.
Considering B&T finds remembering our own phone number tough and struggles to multiply by 10, we thought we’d pay close attention to CEO and co-founder Henry Innis’ presentation unveiling its new Campaign Varying model.
“This is a very important initiative for us,” he said.
“Quick show of hands. How many people in the room believe your ROI should change when your creative changes? I certainly bloody do. How many people in the room believe ROI changes when the format e.g. videos versus link ads changes? I certainly do. And how many people believe ROI changes when the publisher changes? I certainly do.”
Innis explained that to this point, market mix modelling tools have only managed to interrogate effectiveness and model future changes to the channel level. Mutinex’s new Campaign Varying model will change that—and the firm believes it to be genuinely novel, too, hence the patent number.
“We believe [measuring to the channel level is] a fallacy. We don’t believe that links causally to how the modern media landscape works. For 25 years, we’ve been limited by this completely. We’re completely decoupling from the idea of just channel-level ROI at Mutinex and moving towards format, creative and publisher as first class citizens within the model because that is the only way to begin to connect market mix modelling to what truly matters—media planning,” continued Innis.
So how is Mutinex managing it? Its new model analyses the underlying attributes—creative, format, publisher, geography, audience segments, CPM and more—to give you a better understanding of how a channel performs, rather than simply supposing that channel performance changes over time.
This model, Innis explained, had been validated using open-source MMM-eval on real-world datasets across the banking, telco and retail sectors. He added it outperforms Google’s Meridian MMM and Mutinex’s previous Time-varying model across all key metrics. Mutinex found that this new model reduces errors by a third compared to Meridian, is significantly more stable when new data is added than Meridian and that it is far less susceptible to noise—meaning that reported ROI shifts are taken as a signal, not variance in the model. That last point is particularly important.
“Crucially, this model unlocks granular causal inference. By decomposing channel-level spend into its constituent attributes—Campaign, Format, and Publisher—marketers can optimise not just where they spend (Channel) but what they put there (Creative/Format). This moves MMM from a high-level allocation tool to a granular level for creative tactical efficiency,” read the handout.
“In the average market mix model, in a time-varying context, we look at how channels’ effectiveness changes over time. We learn around 200 or so parameters over four years per channel… If you think about how much your creative changes, how much your offers change and things like that, it can explode in complexity,” said Innis.
“When I explode the complexity of a model, what I should see is an explosion in the stability metrics. So the more I expand things, the less stability I get. However, with the Mutinex Campaign-Varying model, we get the exact same stability we would get, if not better, than any of the other models, except we’re now able to learn 4,480 different ROI characteristics across a four-year period for an average customer.
“Now if you think about plugging into the DSPs, that’s a 2,082 per cent increase in the granularity of your MMM. As we get to more granular insights and as we try to plug these tools into the planning systems, this is going to become mission critical.”
Taken with Mutinex’s MAITE platform—and its new access to WARC’s library of award-winning work—Mutinex believes it can give marketers answers to nearly all their campaign and media questions in one spot.
“The challenge for most companies is that the number of questions that they get asked, particularly as a CMO, and the speed at which they’re expected to answer them has become prohibitively fast,” Innis told B&T exclusively earlier this week.
“We’ve invested so much more in the experience and the agentic layer. We’ve invested so much more than just the pure play modelling. We understand that our job is not to provide measurement, it’s to reduce the cost of getting to an answer.”
“I’ve seen people able to write board reports with this. I’ve seen people analyse price volatility and trends with this. I’ve seen people able to brief executive teams—what would have taken a week—in the room. Fundamentally when we talk about charting a pathway to marketing superintelligence, this is what it looks like.”
It’s a bold pitch indeed.
