![](https://crypto4nerd.com/wp-content/uploads/2023/06/0kPHD4vhuPM_Nw_dF-1024x683.jpeg)
With the rapid growth of marketing technology and the always-increasing sales target by companies over decades, marketers have been exploring ways to achieve larger audiences and better conversions. Gigantic ad platforms which include Facebook and Google pounces on this opportunities by creating advanced features like look-alike audiences, a type of audience with similar profiles to your current customer base, and by optimizing its algorithms. Even with all the features and enhancements, relying on one ad platform might not be the best move as consumers suggest. According to McKinsey (2022):
60 to 70 percent of consumers do their research and shop both in stores and online, and they shop 1.7 times more than shoppers who use a single channel. They also spend more in monetary value.
Omnichannel marketing becomes a great solution for marketers to create scalable reach and build unified audience experience across channels as it can have a real business impact to the company. Moreover, it has been used everywhere by numerous brands and different industries, including Best Buy, Nike, and Sephora.
However, building omnichannel marketing is not as easy as it sounds. One of the major challenges in creating a seamless omnichannel experience is about connecting the dots — integrating audience data among different marketing channels (eConsultancy, 2019). The problem is strengthened by the fact that the gigantic ad platforms create a walled garden for their own “playgrounds”. It poses a considerable problem for digital marketers who want to take a holistic approach as the ad platforms are super siloed. Those platforms don’t do well in communicating with each other, meaning that cool features like look-alike audiences or other specific custom audiences you’ve built in one media cannot be specifically targeted in another at the same time. A holistic approach in digital marketing faces a new brickwall of challenges.
Moreover, third-party cookies are already things in the past since Apple, Google, and Mozilla had announced that they would remove support for third-party cookies in their browsers since 2020. The very thing that allows for reaching a highly-defined audience is now no more. Marketers are left only with their own CRM data (first-party data); for some, they have some sort of second-party data to leverage. A holistic approach can’t seem to go easier in this post-cookies era.
One of Oppna’s partner, a rising Indonesian digital bank, had a similar problem — they are trying to build an omnichannel experience. For a festive campaign, the company wants to acquire their offline target market of brick & mortar customers into online digital bank users as they have better behavior, such as higher usage of debit card. However, there are 2 problems: their users’ data of those offline market are very limited and they want a scaled audience that could be reached through different platforms.
The question is now how to build an effective omnichannel marketing despite higher restrictions on audience data connection and third-party cookies downfall. One of the answers is by collaborating with data providers, be it with another company you’ve partnered with or third-party data providers like DMP (Data Management Platform) or DSP (Demand-Side Platform). Data collaboration means there are lots of opportunities that you can leverage to build an omnichannel marketing — new segmented audience and user enrichment to name a few.
The possibility doesn’t stop there. You can create a second-party or third-party audience segment that is tailor made for your company based on customer profiles, and you can use it across platforms! What’s a better tool than a look-alike audience that can be used to replicate your data to millions of users while still being able to use it in various media channels, offline and online, for you to scale your marketing efforts. You can leverage a look-alike audience in multiple ways, from doing it on your own to collaborating with data providers for the data enrichment and model solution. Let’s understand what and how a look-alike model works so you can leverage it optimally.
Lookalike model is a popular method to expand the size of an audience that can be targeted by companies from a much smaller segment. The scaled audience mimics the profiles and behaviors of the original smaller audience. With the way the model works, there are several usecases where look alike audience could thrive:
- Reach new prospects that look like the company’s best customers (gaining new audience)
- Find hidden gems from existing leads that mimics the behavior of converted leads (prioritization of sales funnel)
- Decide which existing users or customers who has similar profiles with high-value users (user retention or increasing user’s value)
Before getting to know how lookalike model works, you need to understand a few concepts that are critical to lookalike model:
- Seed audience is data filled with existing customers based on whom we are interest to find the look-alike audience (e.g. converted leads and high-value customers)
- Reference set is a user database in which we want to find customers who are look-alikes to the seed data previously mentioned. Reference set can be accessed from different sources, namely data providers (DMP and DSP), social platforms (Facebook), or your company’s population userbase
The lookalike modeling process starts by combining a relatively small seed audience into a much larger pool of data, which is reference set that you’ve learned, and by creating different labels between seed and reference set. The seed audience will then be enriched with attributes or features similar to that of the reference set. The rich attributes of the reference set are then used to train a machine learning model so that it can identify the attributes most predictive in terms of similarity to the seed. The reference set is then scored individually on their similarity to the seed using these predictive features. With this process, you can unearth valuable attributes that might be hidden in the first place and unavailable in a generic pre-built segment. Sometimes, the predictive attributes might not be the one you think!
Going back to the digital bank case, the company, then, needs to leverage their small amount of offline user data to be brought online. The company reached out to Oppna for an objective of scaling their limited offline users’ data for acquisition. The company chose to use Oppna’s Look Alike Audience as it has what it takes to build a great look alike audience: data sources from multiple industries/segments, 100+ unique data attributes, and robust look alike model.
The most fundamental point that is needed to be taken care of is the seed audience that you put into the look alike model. It is still necessary whether you build your own look alike or use the solution from data providers. So, what’s necessary for you to build the optimal seed audience. Here is a checklist for you to consider:
- Multiple pseudonymized identifiers: Without identifiers, no connection will be made so make sure that you have a complete identifier for each of your users. Do it better, have multiple identifiers, such as phone number, email, and advertising ID, to allow a high march rate. You need to also make sure you have it hashed for internal and external privacy purposes.
- Be specific with your segment: Define what you precisely want to replicate. Don’t be too general with it. For example: create a seed segment of converted leads in the last 30 days compared to seed segment of any converted leads (as long as you have enough data)
- Complete user information: Information that can be inputted by each provider is different. If you want to know more how you can set multiple identifiers or different information for a single person/entity, check this out: Approach to Understanding User Better (An Introduction)
If you want to build your own look-alike audience, there are two additional things you need to prepare, namely:
- Rich attributes for a better predictive power: Set up numerous attributes for your user data that comprises demographic, interest, behavior. The combination would result in better model
- Optimized ML classification model: Build and test several models for classification, such as logistic regression or random forest, and see which works for you!
This kind of preparation was also done in the digital bank case in Oppna. Their seed segments have multiple pseudonymized identifiers and a pinpoint-level segment. What did it result into, you may ask? During the festive campaign, the digital bank managed to gain 30 times higher average balance, additional 2.5% in debit card owner, and 2.5 times higher in install to register ratio. That’s the power of great seed segment combined with robust look alike model and rich attributes!
Oppna provides third-party look alike modeling solution and other solutions related to lead generation and customer profiling which focus on Indonesian market:
- Oppna Look Alike Audience: Submit your seed segment and the rest is ours to do. Using our 100+ unique data attributes across industries, we are able to build a robust look alike model. Customizing your audience based on similarity and reach is not a problem anymore.
- Bring your own data: Oppna also allows you to replicate seed segment to your own population data for retention or cross-selling purposes.
- User attributes enrichment: If you can build the model on your own, Oppna is here to enrich the attributes for your seed and reference set, resulting in a better-optimized model.