The Lookalike Folly

As companies become customer-first, their valuable audiences become more specific. Perhaps your most valuable customers have signed up for a promotional credit card and have recommended a friend. Maybe they have bought 2+ vehicles from your brand and always purchase all the bells and whistles. Or, as a result of internal analytics, perhaps they fit into your most valuable buyer persona.

This valuable headway is thrown awry by many lookalike modeling methodologies; a long-accepted standard tactic for marketers who seek scale. In many cases, marketers are told they need to start with a seed group of 50,000 - 100,000 users to build a reliable lookalike audience. Your ideal consumer is turquoise. To get relatively close, audience providers are asking for a lot of blue and green. But what if your canvas isn’t that large?

Such large seed audiences are needed because demographic, age, and other sorts of weak probabilistic data provide less-reliable signals for effective lookalike modeling.

This is not practical. Marketers who work with Semasio don’t have to settle in this regard.

From as few as 300 existing customers Semasio can generate a Semantic Twin; our unique take on a Lookalike Model. How is it done? Well, Semasio sees more than 90% of the internet population. When a user our platform knows visits a page, we send a crawler. Through Natural Language Processing (NLP) we derive the most significant terms and phrases. Then, we append them to that users individual & anonymized user profile in real time. This surmounts to 40,000 valuable data points per user, per month. It is thanks to data quality that Semasio needs a positive sample of only 300+ unique users.

So, when you upload a positive sample of 300+ unique users, Semasio stacks that positive sample of users on-top of each other to find what differentiates their semantic profile from the rest of the population (the negative sample). With flexible controls to determine the balance between high affinity (stronger match to the semantic profile of the positive sample) and high scale (higher reach and predicted impression count), actionable audiences of users can be fine-tuned to suit any campaign’s unique needs. Segments are updated automatically in real time, just as our profiles are. We truly empower brands to find and test their “impossible audience”. For many leading global brands, this approach has led to a simultaneous reduction in CPA and increase in scale.

Further, it’s simple. Advertisers can pass over Cookie IDs via excel or one of our many direct integrations. If advertisers aren’t tracking their users, Semasio can place a static or dynamic pixel, depending on the goal. Then, we simply push the segment to your preferred DSP. Self-service users of our platform do this every day, completely on their own! Importantly, match rates are exceptional: greater than 90% for many of the leading DSPs.

Rest assured: Semasio’s clients own their data unequivocally. Any data brought to our platform is used only for that marketer and their use case. Semasio is GDPR, NAI, and DAA compliant.

Does a Semantic Twin empower strategic planning, too? Yes. Semasio gives marketers a dynamic, three-dimensional picture of an existing pool of customers: their affinities, the universe size of an audience pool based on their relative threshold of affinity, and on which URLs their valuable users congregate.

Whatever your valuable audience might be, don’t simply throw them into the mix of a generalized set of high value customers. In a customer-first world, empower the individuality of your consumers and the power of your data via improved prospecting strategies.

Are lookalike audiences not the best fit for your industry or vertical? Read about how Semasio creates segments of users from key terms, phrases, and URLs here.

Author: Mikael Holcombe-Scali

Mikael Holcombe-Scali is Business Development Manager at Semasio, based in New York City.
Find me on: