Uplift Modeling Can Reveal Alternative Paths To New Customers

"Data-Driven Thinking" is written by members of the media community and contains fresh ideas on the digital revolution in media.

Today’s column is written by Ellen Houston, director of applied data science at Civis Analytics.

Marketers are increasingly shifting to performance and optimizing their investments with the right audience to drive business growth.

This is no small task. With the proliferation of data sources and marketing technologies, decisions are increasingly nuanced when deciding who to target and what to message.

Today, improving a campaign's effectiveness and efficiency means marketers need to better understand who would have just showed up anyway and whose behavior was actually changed to drive incremental results. This means relying less on traditional “all or nothing” methods of credit and toward concepts such as persuadability and uplift.

Persuadability matters most

Persuadability and uplift modeling have long been a staple in politics. The idea is that there are people who already support a candidate, people who will never support that candidate and the persuadables: those who might support the candidate if they are reached in the right way. In politics, these are the people who matter and where nearly 100% of efforts are focused.

Brands are increasingly seeing the value of thinking like political advertisers. There are people who will buy their products, people who will never buy their products and people who might buy their products if they are reached in the right way.

Their likelihood of being swayed to buy a product is their persuadability. Uplift modeling is a predictive technique that forecasts the likelihood that a person’s behavior will be swayed by a given communication. It helps to think in terms of four categories:

The sure things: people who will buy a product whether you market to them or not.

The never-gonna-happens: people who will never buy a product, whether you market to them or not.

The do-not-contacts: Counterintuitively, these people will buy a product only if you don’t reach out to them. These consumers represent not only a waste of ad spend but also potentially lost sales if you’re not respectful of their space.

The persuadables: customers who matter most in your marketing efforts. These people will buy a product if they receive the right types of outreach.

Uplift modeling is about investing to drive the most impact, focusing on the persuadables and reducing ad spend wasted on customers who would have purchased anyway. That’s not to say that existing customers shouldn’t receive messaging, but brands should be thoughtful about the types and frequency of messaging these customers receive.

Marriott should send platinum members a different message than those who spend just a few nights a year. Communicating with sure-thing customers is about fostering loyalty, creating relevance and driving deeper lifetime value. Communication with persuadables should focus on sharing the most meaningful message to drive new behavior.

The mechanics

To measure uplift, marketers must compare the actions of similar-looking consumers who have been exposed to messaging vs. those who have not. An easy way to start is to test messages before they are sent – if one message is highly persuasive only with a specific persona, spend can be directed toward a digital campaign highly targeted to that persona. If a different message persuades a broad audience, it could be a good fit for a mass channel like TV.

It also helps to think of persuadability on a spectrum. There aren’t just two buckets – those that can be persuaded and those that cannot. Rather, persuadable consumers vary according to how much and what type of effort might be required to sway them to purchase. Refining marketing efforts for persuadability and uplift takes time and testing.

When done at scale, this comparison helps to make more informed campaign decisions. For example, marketers can increase the rotation of their most broadly appealing messages in mass channels such as TV. Or, they can identify smaller and more targeted audiences for digital activations and optimize frequency and budgets to eliminate wasted spend on tactics that don’t move the needle.

Let’s say a beverage company is launching a new sparkling tea, and its market research says the ideal target skews younger, female and more urban. Historically, the company may have built a campaign focused on young, urban females using heuristics and demographics. However, this isn't the full audience. Uplift modeling identifies the most persuadable audiences across both men and women in their early 20s, driving the company to spend on more diverse platforms with more specific audiences, which may have previously been missed.

When wading into uplift modeling, marketers shouldn’t be afraid to take a crawl, walk, run approach. The first time they apply this technique will be the hardest. They should start with a single campaign, glean learnings, test and then scale across campaigns.

Ultimately, success with uplift modeling requires marketers to have hard conversations with their partners. Shifting from gathering a high volume of cheap (and potentially unqualified) conversions to driving real, incremental gains represents a shift in incentives and a new definition of success. No longer is a campaign all about driving the most actions and outcomes, but about acquiring new customers who wouldn’t have taken action otherwise.

Ultimately, this will cost more on a per-conversion basis, but in a performance-driven marketing landscape, where growth is the name of the game, uplift modeling is the technique that delivers.

Follow Civis Analytics (@CivisAnalytics) and AdExchanger (@adexchanger) on Twitter.

 

Add a comment

XHTML: You can use these tags: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <s> <strike> <strong>