A state of the art lookalike modeling approach
The uniqueness of Resonate lookalikes models can be found in the strength and uniqueness of the Resonate data and the power of our state-of-the-art AI technology found in rAI, Resonate AI. While others are developing models on stale, infrequently updated data, Resonate lookalike models are powered by billions of daily signals, providing the most accurate and up-to-date understanding of the consumers changing behaviors, interests, and intent.
Through rAI, Resonate gives our clients access to cutting edge deep learning and AI techniques to find their exact audience. While others use legacy, easy-to-run, and low-fidelity methods to find consumers similar to a seed audience, Resonate develops a model in real-time, leveraging up-to-the-minute data, and scoring on the fly to produce the highest quality output.
If your need for reach is paramount, Resonate allows you to loosen the scoring criteria to find the best fit of performance and reach for your needs.
And the best part? Our lookalike modeling is fully self-service in the Resonate Ignite platform. We even offer the advanced capability to create a lookalike model using your onboarded CRM data combined with Resonate Elements to further enhance your reach and targeting. Create multiple models in seconds, and activate immediately.
How it works
You can activate on an audience saved with your 1st party data via retargeting or lookalike modeling. We recommending first using a retargeting delivery method for activating your 1st party data. If you don't see enough scale, then you can come back and do a lookalike delivery.
- Select Retargeting when you want to deliver to people in your 1st party data set.
- Lookalike Modeling is a better choice if you're looking to target people similar to your selected audience. This model considers the aggregate behavioral footprint of users that match your audience and ranks them in order of importance. If you select LAM, you have to define your match rate. The lower the rate is, the less different the LAM audience is going to be from your audience. Consequently, the lower the rate is, the smaller the audience size is going to be. 30% is the highest possible match rate for credible results. Setting the slider at 5% means we will set aside the highest 5% of closest matches. Setting the slider at 30% is less stringent, we will set aside the highest 30% of closest matches. Since the LAM pool is capped around 230 million devices as well as ranking varies from 1-30%, it means up to around 70 million devices can be activated against.
Activating via lookalike modeling helps you scale your 1st party audiences by finding more people who look like your best customers. This helps combat challenges with low sample size when activating 1st party data.
You can also exclude your 1st party audience from your audience definition to suppress your customers from your targeting. This ensures you're not wasting ad dollars by spending on your existing customers who have already purchased.
Resonate lookalike modeling is an instance-based approach, which calculates a distance-like measure that balances the frequency of visits to particular domains with their overall abundance. This measure is used to rank the similarity between a target set of records and a candidate set.
The method starts with a target set of records and computes the frequency of visit for each domain for each record in the target set. Then, for each candidate record, the frequency of visit for each domain is calculated, and we compare the total differences across these frequency-weight visit records. The distance-like measure is then computed by balancing the frequency of visits to particular domains with their overall abundance.
The distance measure can be used to determine the similarity between the target and candidate sets, and the target and candidate sets can be ranked based on the similarity score. This allows the user to identify the candidate records that are most similar to the target set.
LAM is a useful tool in targeting settings, where standard Key Performance Indicators (KPIs) apply, as well as in scenarios where sample sizes are small or where defining a null set is intractable. For example, it can be used to determine whether a person did not click on an advertisement due to the creative being ill-suited or because it was not even presented.
The performance of LAM can be assessed in multiple ways, including targeting, recall of the initial segments, and the ability to track performance metrics. In practice, LAM operates between a tension of theoretical expectations and small sample sizes, and recall of the training set is the only performance metric that can be tracked.
In conclusion, LAM is a useful and effective method for creating audience segments and ranking the similarity between target and candidate sets.