Back to the future with geo-contextual solutions

Gagan Shetty, Data scientist I, MiQ & Abhinanda Roy, Associate product manager, MiQ

The story so far

The programmatic advertising industry has traditionally relied on cookies to track user data, target relevant users and generate revenue from clicks and conversions. For example, your searches for the best deal on a necklace for your sister’s birthday place you into various categories that include you and millions of others. These categories have historically been used for targeting purposes but the situation has changed.

New data protection regulations, pressure from advocacy groups, and consumer privacy concerns prompted Big Tech to take another look at cookies. Apple’s Safari browser was first, blocking all third-party cookies for users by default in 2020. And Google isn’t far behind, with third-party cookies set to be deprecated in Chrome in 2024.

As a programmatic media partner, MiQ needed to look at credible proxies, the sons and daughters of cookies, for privacy-compliant targeting. We asked ourselves a few questions: How do we maintain campaign performance without the use of cookies? What can we repurpose and future-proof from existing solutions? And which data providers can we work with to achieve this?

Here’s how we reached our solution:

Start with what’s already there

Geo-contextual solutions were nothing new at MiQ. And with programmatic advertising reliant on knowing (1) where and (2) who relevant users are, we knew they would be a good starting point.

Moreover, a number of datasets appeared that could fill in the gaps left by cookies in geo-contextual targeting. These datasets segmented users by trait, with an astounding array of categories across age, education, lifestyle, activities, and more.

This data is collected without the use of third-party cookies. It comes instead from sources like telephone directories, censuses, and consumer surveys.

This means a whole new level of data, a whole new set of context, and most importantly a whole new level of targeting for end users.

And the best part? The data is at a ZIP code level. This means, for example, that we’d know the exact codes to target when reaching users with an interest in sports, within the 35–45 age and $20–50K income groups. With ZIP codes being one of the most precise geographical units available, these datasets gave us the accuracy we were looking for and became our replacement for pinpointed cookie-based user data.

From data into a solution

The raw data had 500+ segments, so we wanted to make it faster and easier for the users to find what they were looking for.

We decided to use a big thread to tie all of the segments together, resulting in a more precise view. To do this, we created segment groups. These are a second category to map the individual segments into broader categories e.g. the segment group ‘Age’ has segments including ‘age 19–24’, ‘age 25–29’, ‘age 30–34’, and so on.

With segment groups in place, the next step was to create a tool as follows:

1. Input: User selects a segment group, before choosing a segment within that group.
2. Output: List of relevant ZIP codes based on user selections

Refining the results

At first, our solution selected relevant ZIP codes based only on segment count (how often they appeared in the chosen segments), without looking at other contextual information. However, we wanted to improve our accuracy further, so we combined the segment count with household and individual counts in a weighted sum model using a multi-criteria decision analysis (MCDA) method.

For example, if ‘Age’ is the segment, X is the household count and Y is the individual count and (w1,w2,w3) are their respective weights, the final score would be calculated using:

Using this method, each segment would generate a separate list of ZIP codes.

Next, we considered how users would typically select multiple segments in one search. This was an issue as certain ZIP codes would appear in more than one segment, yet targeting the same ZIP code multiple times wouldn’t make sense. So we updated our methodology to send the user a final list of ZIP codes, only after combining the selected segments together and removing instances where a ZIP code has appeared more than once.

Here’s how the final workflow looked:

Effectiveness and adoption

• Geo-contextual audience data has outperformed cookie-based audiences in multiple head-to-head tests. The improvement is around 14–18% across different channels.
• Every quarter at MiQ, an average of 50 campaigns adopt these geo-contextual solutions across the US and UK, delivering an average of 300 million impressions targeted through these segments.

What’s next for geo-contextual

• Global market expansion
• Algorithm enhancement for better accuracy
• Automated activation across all major DSPs.

Conclusion

Given the balancing act required to protect user privacy and maintain the precision of targeting channels, we’re pleased with how our solution caters to both of these needs.

Plus, working with privacy-safe data and having solutions and capabilities built for custom requests means we’ve been able to deliver on performance despite the changing cookie landscape.

With an interest in writing and narrating stories, Abhinanda is an associate product manager for MiQ based out of the Bangalore office. Working alongside her is Gagan — one of MiQ’s data scientists — who has a keen interest in sports.