Programmatic: How to Optimize Your "Lookalike" Segments Using the Fragmentation Strategy? 🪨⛏️

Ah, lookalikes (LAL)... if you work in digital advertising, you’ve definitely heard of them: it's that magical technique that lets you expand your reach in the blink of an eye 🫰✨

⚙️ How does it work?

You start with a "seed" audience — for example, all visitors to a specific page — and look for statistically similar profiles.

Technically speaking, most LAL technologies use the same method: they calculate the affinity score between the seed segment and other available segments.

Affinity Score = Probability of finding the seed audience in Segment \ Probability of finding the seed audience in the total population

Then, a new LAL segment is created, grouping together all segments that have a high affinity score.

Benefits:

📈 Scale reach while maintaining a statistically high-quality targeting.

Limitations:

⚖️ The trade-off: You have to balance volume and quality. Since there is no standard industry definition of what a "reasonable" LAL is, the logic used behind this trade-off can often be opaque.

🌱 Seed quality: The quality of the LAL depends heavily on the quality of the seed audience.

🔢 Data depth: The quality of the LAL also depends on the number of segments used for the projection (e.g., if a publisher or advertiser only has 10 segments in their DMP, their ability to build a good LAL is pretty limited).
Overcoming Limitations: The Fragmentation Strategy 🪨⛏️

When it comes to that last limitation, there is a strategy to boost efficiency: fragmentation (or segment splitting).

Let’s go back to our example of a publisher with 10 audience segments. What if they decided to cross-reference these 10 segments with one another?
Instead of just 10 segments, they would suddenly have around a hundred sub-segments to run their projection on.

And technically, the publisher doesn't even need to manually create those 100 sub-segments:
🧮 They can just run virtual calculations within their DMP.
👬 Then, they can generate a single LAL segment using a comprehensive SQL formula that isolates only the high-affinity sub-segments.
With fragmentation, you can achieve a much better balance between reach and quality ⚖️

Link to original post : https://www.linkedin.com/feed/update/urn:li:activity:7486284040116035585