In the span of a few years, custom algorithms have gone from nonexistent to a necessity for many media buyers.
Custom algorithm company 59A, for instance, has clients who have used its algorithms for years and now use its tech for all of their digital activations.
One of those top clients is based in the US – which was part of the impetus behind 59A’s recent expansion. The company launched six years ago in the UK, but it has done business within the US for three years. 59A officially expanded into the US last week, appointing Jon Nash as its first US CEO.
A custom algorithm in media buying is basically a “bespoke piece of code” that determines the best way for an advertiser to plan, optimize and allocate campaign spend across platforms, said Nash.
Several companies, including Chalice and DoubleVerify-owned Scibids, have made a name for themselves in the custom algorithm space, positioning their tools as an alternative to walled gardens with more restrictive data.
If a brand is working in Meta, for instance, only having access to Meta’s platform data and the brand’s first-party data can be limiting, said Nash.
59A develops each brand algorithm by looking at both online and offline data points and determining who and where the target audience is likely to be. From there, it deploys the tailored algorithm across ad platforms, using the results to inform future algorithms based on what worked well.
The company breaks down the development process into three stages: thinking, doing and learning.
The thinking stage involves collecting data to help curate the algorithm. In addition to standard online sources, like platform data and a brand’s first-party data, 59A also pulls data from a variety of open-data sources without any sort of digital IDs. Those sources can be relevant to a broad audience (like the Chamber of Commerce) or industry-specific (like tracking road accidents). The specificity of these sources can give advertisers detailed insights, like an academic publisher using a state database of lawyers to better understand who to target with particular books.
The data doesn’t just inform advertisers about who they should be trying to reach; it’s also about where and when. For instance, said Nash, a pharmaceutical brand might need to take into consideration certain times of year when flu rates are higher or whether a new virus broke out.
The thinking stage functions similarly to a campaign-planning tool, but the data is ultimately used to build out the custom algorithm, Nash said.
But, perhaps surprisingly, all of the data mining and analysis necessary in this first step isn’t in the hands of AI. Instead, 59A’s business intelligence and strategy team heads the process, thinking up and analyzing the various data points.
The team uses a brand’s goals to determine different data points – ranging from weather to where interested consumers might live – to track and build a proposal. Each data point receives a tiered rating from one to five to determine the likelihood of someone in that category converting.
The last element of thinking is actually building the algorithm. (Yes, building the algorithm is part of the thinking, not the doing, step. No, it doesn’t make a ton of sense to us, either. 59A Founder and CEO Adam Ray justifies it by explaining that the thinking stage “is a period of ‘pre-optimization’ before any media dollars are spent.”)
To build the algorithm, the business intelligence team layers all the data sets on top of each other to see the audiences, times and regions most relevant to the brief, said Anthony Farley, 59A’s VP of scale – it’s sort of like a heat map.
That’s when the segment is given a tier, which is translated into a piece of code that allows the buying platform to determine the optimal bid price and budget allocation.
The doing stage involves implementing the algorithm across all the brand’s advertising platforms.
Typically, advertising platforms are fragmented, each using different data points for targeting, Nash said. A custom algorithm, he explained, unifies all the data across platforms to target the same, specific audience everywhere.
The final stage, learning, is similar to the thinking process, but now with campaign data to back it up. It’s “kind of rethinking,” said Farley. The team tweaks the algorithm based on how the predicted audiences performed in practice.
For a brand’s next campaign (or the next phase of the current one), 59A reevaluates each data point based on its recent performance. Tweaking the algorithm from campaign to campaign is “the real value” that the company offers, said Farley, along with the scale of data 59A considers when developing each algorithm.
Bringing “real-world data into the equation,” he said, provides advertisers with a much more granular understanding of their audience and how to reach them.
