Everyone in programmatic knows that the best way to drive outcomes is to use some kind of identifier … right?
But a CTV campaign Shutterfly ran at the end of last year to promote its holiday gift packages and photo deals challenges that perception.
Shutterfly’s goal was to uncover new pools of potential customers to tap into. Rain the Growth Agency, its agency partner, suggested devoting a portion of its budget to test a custom algorithm that relies on search data, rather than deterministic or ID-based audience signals, to find geographic regions with the most prospects.
Although Rain had never tested a search-based custom algorithm before, it made sense as a means for identifying markets where there’s interest in a product category but low awareness of a specific brand, said David Nyurenberg, Rain’s director of video product development and innovation.
In other words, geos where lots of people are searching for digital photo services in general, but where brand awareness of Shutterfly lagged behind competitors.
Searching for incremental audiences
Rain sourced the search data from TV measurement company EDO, which can tie search insights to households within geographic markets using IP address and device ID matching.
Chalice AI created a custom algorithm to ingest this data from EDO and determine which geos Shutterfly should target and when.
“We layered the algo on top of a bundle of different CTV PMPs – all premium stuff like Hulu, Peacock, Spectrum, Paramount, Samsung and Roku,” Nyurenberg said. “And then the algo just flexibly optimized across all those PMPs within the markets that appeared to need the media in a given moment.”
Rain tested the search-based custom algorithm against other targeting and optimization tactics, including keyword targeting, retargeting and lookalike modeling, as well as third-party audience segments provided by a contextual vendor and an automatic content recognition (ACR) vendor. Rain declined to name these vendors on the record.
The custom algorithm outperformed in terms of driving site visits, account setups and photo package purchases from new customers, based on measurement by iSpot.tv.
From October to December, between 17% and 19% of the conversions attributable to the custom algorithm came from new customers. Most of the other tactics hovered in the 10% to 15% range. Retargeting was the least effective method for reaching new customers, accounting for just 8% of the audience.
In Nyurenberg’s view, these results prove that addressable targeting strategies aren’t the best way for advertisers reach incremental audiences. And that’s because most addressable tactics – particularly retargeting – rely on deterministic audience signals that can only be collected once a customer is already deep into the purchase cycle.
“Intent-based signals and third-party audiences show these people are likely on the conversion path already,” Nyurenberg said, “so how much of a difference is your media going to make?”
In fact, they’d probably have made a purchase even without having seen any targeted ads, he added.
But the search data EDO fed into the custom algorithm corresponded with people at the beginning of their purchase cycle, closer to consideration than conversion. The search data also demonstrated these audiences weren’t already familiar with the Shutterfly brand, so there was real value in targeting them with CTV ads to raise brand awareness, Nyurenberg said.
The custom algorithm also drove better new customer ROAS and a better cost per new customer than other addressable tactics, and that’s to be expected, Nyurenberg said.
The search algorithm “had the highest new customer percentage and it also had the lowest CPM, which makes sense,” he said. “You’re not layering on addressable data, which is very competitive, so you’re going to get the most cost-efficient spend.”
Ongoing optimization
Throughout Shutterfly’s three-month campaign, the custom algorithm would perform weekly optimizations so that once brand awareness and engagement had been sufficiently raised in a given market, it could reorient ad spend to new markets.
As a result, new customer ROAS and cost per new customer got steadily more efficient as the campaign progressed. New customer ROAS rose from $0.31 in October to $1.49 by December, and the cost per new customer fell from $243.26 in October to $57.31 by December.
Shutterfly also ended up targeting certain fruitful regions it wouldn’t have otherwise, Nyurenberg said.
“Typically, programmatic algos will allocate spend to the same high-avail cities, like New York, Los Angeles, Miami,” he said. But the Chalice algorithm disproportionately allocated spend to smaller metro areas, such as Honolulu, Reno, Santa Barbara, Utica, Palm Springs, Lexington and Bangor.
“This is not the typical DMA mix that you would expect to see in a programmatic campaign,” Nyurenberg said. “The algo really zeroed in on the ones that met the search criteria.”
The search-based custom algorithm approach worked so well that Shutterfly kept it active after the holiday campaign season ended, Nyurenberg added.
And Rain has now started pitching the strategy to other clients looking to drive new customer acquisition. For Nyurenberg, it was a successful case study to demonstrate the viability of targeting tactics that aren’t based on addressable audiences.
“This campaign proved that modeled tactics outperform ID-based targeting,” he said. “It feeds into the ongoing narrative that targeting, future forward, is going to be largely AI-modeled, not ID-based.”
