Beyond a DMP: Audience Discovery and Persona Indexing
Abstract
If we were to single out the one reason why many clients fail to get value out of their DMP, it would be the inability to create audience segments that make a difference. Audience segmentation is a crucial step of any marketing and advertising strategy and unfortunately, it is where we fail the most. It is 2020 and we are still building segments based on someone visiting a page, making a purchase or clicking through a display ad. I hope you don’t misunderstand us, this is not wrong but not exactly meaningful. These are signals but not audience segments.
Over the past 2 years, we have been working with different clients (from Travel and Hospitality, CPG, Technology and Healthcare) who are trying to get more out of their DMP and mainly around creating actionable segments. They are shoving so much data into their platforms from behavioral data (site/app) to customer onboarded data to 3rd party prospect data. The one question all of them kept posting was:
How can I turn this data into meaningful audience segments that I can trust and use?
Many DMPs struggle to come up with segment recommendations or build intelligent segments based on patterns and trends. We say “many” and not all because Salesforce and Adobe have made progress to solve this problem. Salesforce with Einstein Segments and Adobe with Predictive Audiences. Both offerings are promising but have their own share of limitations.
Overview
Meet our imaginary client, Ooda, who is a global hotel chain with multiple sub-brands that target different budgets, amenities, audiences and locations. Ooda knows who books their hotels but they didn’t know why someone would book one brand versus another beyond the common intuition (leisure vs traveler, family vs single, city vs suburb…). In efforts to narrow down the requirements, we concentrated on an upcoming campaign that they wanted to launch for two of their brands: Hamra & Bayda. The campaign’s KPI is to drive the highest click-through-rate (CTR) and bookings, offsite and onsite. To determine the best possible creative to serve each site visitor, they had to analyze consumer behavior and preferences to answer the following questions:
They wanted to segment their consumers to create distinct personas/audiences, which would allow them to simplify marketing efforts while simultaneously increase effectiveness.
Ooda is fully invested into the Adobe Stack (Analytics, AAM and Target) as well as Google Advertising (Display and Video 360). Given the limitations of what we can do in the DMP, we built a data pipeline to export the data from the DMP into a data science environment to analyze, model and train. Once the characteristics and audience personas were identified, we funneled them back into the DMP to activate through offsite targeting (Ad Network) and onsite personalization (Adobe Target or any site optimization tool).
Data Analysis and Modelling
Overview
Advertising strategies should consider not only the advertisements being served and the channels and tactics used to serve them, but the audiences being targeted. CRM, DMP and CDP systems all provide audience data that can be used to identify groups that are considered to be high-value for a specific marketing initiative. Traditionally, a small number of variables from these systems would be selected by a marketing team and high-indexing individuals would be targeted for some action. A more modern approach leverages statistics and machine learning (ML) to define audiences. After assessing the available data, appropriate methods can be chosen to produce an algorithmic approach to modeling audiences. The goal of algorithmic audience modeling is to identify behaviorally similar groups of consumers (or potential consumers) in a way that a human cannot define. For example, traditional audiences may be defined as LTV segments, loyalty membership tiers, individuals who have purchased certain products, etc. Well-designed audience definition algorithms will identify commonalities between individual characteristics and combine them in ways that produce audiences with distinct characteristics. These characteristic audience differences can be directly leveraged to tailor marketing strategies to unique groups of customers. In our pursue to find these audiences, we worked with two very distinct types of data (3rd party and 1st party). Let’s discuss the hurdles and findings we faced with each.
3rd Party Data
Our client Ooda might see 3rd party data as a way to enrich their underlying segments to produce more meaningful segments and learnings about their customers. The exportable 3rd party data from providers like Bombora or Eyeota, is an aggregated index of your selected segments unique customers/member population overlapped with the providers’ 3rd party traits unique population.
1st Party Data
Due to the restrictions cited above, pivoting from 3rd party data to 1st party data seemed to be the most logical next step. Leveraging the data Ooda had collected on consumer behavior and characteristics made some of the more advanced statistical methods feasible, including the construction of an algorithmic audience model. The methodology agreed upon prior to receiving data from Ooda was to provide them with distinct personas for each of the brands – Hamra and Bayda – to be actioned against in the upcoming campaign.
Activation
Now we get to the fun part, turning the insights above into actions. When it comes to activation, Ooda initially wanted to use this data to enhance their creative assets and media buying but with the shift from 3rd party data to 1st party, we also shifted to activating these insights only on owned property (website and app) rather than the open exchange. In order to do so, we leveraged the integration between Adobe Audience Manager and Adobe Target, Ooda’s testing and optimization product of record.
Personalizing and testing on website
Now that we have new audiences at our disposal, we can test tailored messaging on the site toward these new audiences. The persona audiences should be available in the Audiences section of Target and display with the same Segment name as shown in AAM. Here are a few examples to consider for our hotel client Ooda:
Personalizing and testing within a native mobile app
These new audiences are not just limited to personalizing the web experience. Adobe Target can also be used for testing and personalization within a native mobile app as long as the Adobe AEP SDK is used, and Target and Audience Manager are properly implemented within the app. Since AAM and Target communicate server-side, the workflow for using these new persona audiences is identical between web and app activities. The difference comes with the experience setup in Target using the Form Composer. A native mobile app should receive JSON from Target instead of HTML. The mobile app must be setup to know how to interpret the JSON sent from Target.
To learn more about how overcame the DMP limitations to create meaningful audience segments from 1st and 3rd party data, download our case study. We will walk you through this entire case study from beginning to end including architecture, analysis, modeling and all the way to activation and execution.
