This article explains how to build and target lookalike audiences.

Written by Marcus Johansson
Updated over a week ago

What is a lookalike?

A "lookalike" is a user that is similar to a given segment or conversion population. Lookalikes can be useful to prospect users that bear resemblance to a population of users that is of interest to target. For example, it might be a good tactic to target lookalikes in addition to the always-on retargeting of site visitors.

How are lookalikes scored?

Lookalike similarity is computed with a machine learning model that is based on previous observations of the segment population. The scoring model is computed once each day, based on observations from the past week. Processed bid opportunities are scored against the ML model to get a match between 0 and 100%.

The similarity theshold can be set manually. The default similarity threshold is currently 70%. A higher threshold means a smaller but more similar lookalike population.

Are there any limitations to lookalikes?

Currently segment or conversion tracker user population size needs to be at least 100 weekly users in order to have a lookalikes model built.

How do I use segment lookalikes?

Segment or Conversion lookalike targeting is available as target types in audiences.

NOTE: Audience size estimations will not be accurate for newly created audiences using lookalikes. Since audience size estimation is currently based on sampling, statistics will accumulate over time.

What's the difference between conversion lookalikes and running campaigns with optimization on the cost per conversion KPI?

The model used in campaign optimization based on KPI is built by connecting attributes surrounding the ad space with subsequent conversions. This is an ideal model when deciding when and what to bid to buy an ad impression for your campaign, provided that you have accumulated data.

The lookalike model disregards the circumstances around the ad, but can in return be fed with the attributes of the users for all conversion data tracked, and not just the conversions produced by previous buying.

What about GDPR?

The machine learning model for lookalike targeting is based on hashed and anonymized user data such as time, geo, device and advertiser category, recorded at the time a segment tag fires. As usual since GDPR came into effect, end user consent needs to be retrieved for use of segment tags.

Lookalikes are only available for segments that are set up and populated by your own BidTheatre account - no sharing of lookalike data occurs.

Did this answer your question?