Article adapted by John Lindsay. Original co-authored by Joshua Koran and Mike Ross. Photo credit to Nolan Williamsons www.Flikr.com.
When it comes to intent-based audiences, optimization, and customer acquisition, it’s important to get one thing straight from the start:
There are two separate cycles to the process.
1. Generating an intent-based audience.
2. Engaging that intent-based audience to generate new customer acquisition.
How do most marketers build audiences?
Whether they’re focused on interest, intent, behavior, or something else, an audience is on as valuable as it is useful for a marketer, and most audiences are built by looking in the rear view mirror. I other words, when a marketer wants to build an audience they usefully look backwards in time and say:
· What are these people doing a lot of
· What are these people doing more than others?
So, if someone visits a sports-related website (e.g., ESPN.com) more than the general population, a marketer might build a “sports intender” audience around that individual and that sel that audience to a brand like Jeep.
Why is “looking back” the wrong audience-building tactic?
Brands don’t want to reach audiences built on “what people have done”, but “what people are likely to do.” Moreover, they want to reach individuals who are going to take some kind of action. Put another way, a brand like Jeep Doesn’t want to talk to people who are “interested in sports”. What Jeep wants is to talk to people who are going to take a specific action that delivers a desired result, such as a website visit, test drive or request for quote.
Yes, someone’s interest in sports could be correlated to their likelihood of taking an action, but it’s no guarantee. Success for a marketer working at Jeep is measured by how many consumers take a desired action, not how many people with an affinity with sports see an ad.
Enter predictive audiences
A predictive audience is an audience composed of individuals who are likely to take a desired action. It uses the same inputs as other forms of audiences, but it follows different modeling. Instead of just saying a consumer does something “more” than others in the general population, and then rank the ordering of that consumer, predictive modeling tries to figure out what activities correlate to a marketer’s given success rate.
So, using Jeep as an example, there could be three separate audiences to achieve three separate outcomes. There could be a:
· Brand-awareness audience
· And a SUV-buying audience
· And a website-visiting audience
Making the most of audiences to drive acquisition
Once a marketer builds their predictive audiences, the question is how can those audiences be used to drive acquisition. To be clear, this isn’t a duplicative process. Marketers do not do the same thing as what a demand side platform (DSP) already does. The difference is a generic background default propensity – using the historical default propensity versus what the DSP does, which is to decide what the propensity for an action is right now, in the moment.
If the ad server can be supplemented with predictive audience behavior and background default propensity, then it can do its job even better than before. The nuance is that one form of modeling decides if you’re the right person in general, and the other decides if right now is the ideal moment for you to be engaged. It’s the predictive model’s job to determine if you’re the optimal individual and it’s the ad server’s job to determine if now is the right time to connect with you.
This is the distinction between building the audience and using the audience.
Lower Funnel Audiences for re-targeting: Media and CRM
A similar approach could be applied to the lower in the funnel with regard to returning organic visitors to Jeep’s website. With machine learning profiling applied to the visit behaviors, genuine purchasing interest of a specific Jeep model can be detected determining the right audience. The same machine learning can build a visitor profile across multiple sessions using a dynamic predictive score to detected an increasing interest to buy as a hand-raiser around the right timing. This builds an intent audience for the lower funnel. A data platform that has AI build at the core, has the potential to aggregate all these signals, apply the intent logic to generate the audiences and from there, to activated retargeting campaigns towards both the unknowns and the knowns: The anonymous unknows via Facebook, Search, and the DSP; and the knowns via the relevant CRM channels including email, regular post and by phone.
An additional benefit of website behavioral based intent profiling is that acquisition campaigns can build Look-a-Like (LAL) predictive intent audiences. These tend to be much more effective than conventional sources of LAL audiences, and can enlarge the LAL pool by a factor of 10 or more since it is based on a the predictive intent attributes and not based on website conversions.
When the brand’s first party data are stored in a CPD, profiles are unified around a single UID, deduped, and can be enriched from 1st, 2nd, 3rd party sources. The predictive intent logic applied to these data sets and signals has the capacity to generate audiences and syndicate them across channels for activation within fractions of a second. This approach becomes increasingly valuable in a Post-3rd Party cookie world
The extra firepower of a closed-loop system
The best marketers and most successful brands always invest in a closed loop systems of customer acquisition versus a background-based approach. What this boils down to is a difference in philosophies. With background modeling what you’re effectively saying “If I talk to these people at some time, somewhere, its going to be better than random”. This is a true statement, but it’s not as effective as a closed-loop system which is what giants like Amazon, Facebook and Google use
The Jeep example continued
Let’s hypothetically assume a marketer for Jeep identifies an audience and runs a campaign targeting that audience with a given ad (or behaviorally triggered email). It’s the right audience in general for the needed output (for the example, presume the needed output is “clicking”). But unfortunately, now isn’t the ideal time (or it’s the wrong context) to engage with the audience to get them to take the desired action. So the marketer decides to wait until its the right time to not only achieve the desired output, but to do so at the optimal price as well.
But there’s another step to this process.
Once an action is won, and the consumer executes the desired output (clicking) the consumer will also interact with the brand’s property (in this case, Jeep’s website or a Jeep owned landing page). Once that happens, it behooves the marketer to continue leading the consumer to Jeep’s ultimate goal – buying a product and becoming a customer.
It is in this additional phase that the value of a closed-loop system really comes into play. It allows marketers to not only win an auction, but also incorporate additional tools (e.g. website personalization) into their marketing strategy.
Looking at the closed loop and predictive audiences
Purveyors of predictive audiences and closed-loop solutions with AI embedded at the core, such as Zeta, that expedite and improve the customer acquisition process are extremely valuable to enterprise brands. Owning an integrated DSP/DMP/CDP coupled with insight optimization, built with the same profile, and having a consistent understanding of what’s going on maximizes the likelihood of the consumer continuing their journey towards the penultimate outcome.
Why predictive + intent targeted audiences for digital ads & e-mails – work better than conventional email campaigns?
The Context: Is this the right time to talk to them?
The Interest Level: What is the interest in a given area?
The Message: What are we showing them to achieve the desired outcome?
Assessing the consumer’s state of mind
Being able to understand whether the consumer is in the right frame of mind for interaction or not is key to any marketer’s success. A closed-loop system can evaluate consumers who look similar to a given prospect. It can identify when they clicked on a given ad and in what context to predict their state of mind.
Then, it can look at all the similarities across the different dimensions, time, geography, context, device, bandwidth speed, etc. – to figure out the best time to strike. For example, people using a high-speed connection in the evening on a desktop with a propensity to be interested in sports, are more likely to take the marketers desired action than a person who’s on a slow connection, on a mobile device, during the morning rush hour even though they have the same propensity to be interested in the sports.
Messaging matters, not because it’s specific to a prospect, but because other consumers who share similar attributes to said prospect behaved similarly when exposed to the same messages. Scoring is how we try to enable a brand to talk to the right audience at the right time. Its not just what they were doing yesterday or last week, it’s what they’re doing right now – whether offsite and/or on Jeep’s website.
In addition to what they’re doing right now and what we think they’re going to be interested in the current moment of matching a message to the audience is something Facebook, Amazon, Google and ZetaGlobal do exceptionally well.
To learn more about how intent signals and modeling drive customer acquisition, talk to John Lindsay at WON about intent-based marketing, CDPs, and/or Zeta’s Data Platform and AI predictive intent audience capabilities.