Improve Your Multichannel Marketing Attribution With Machine-Learning

Global spending on media is expected to reach $ 2.1 trillion by 2019, according to McKinsey. But is all that money effective in producing improved ROI? Without knowing which channels are driving sales—or, more specifically, which individual efforts are working—marketing spending is like a black box.

In today’s world of digital commerce, it’s not uncommon for a transaction to involve as many as 30 marketing activities, or “touchpoints.” Yet many marketers take a convenient shortcut and credit each purchase to the last touchpoint before the sale. To properly attribute the influence of all marketing activities, businesses need an enterprise-grade methodology capable of quantifying each touchpoint’s impact on the sale.

Why you should worry about marketing attribution

You may know that in aggregate your marketing “is working,” but without true attribution, you don’t know which marketing activities have the greatest impact and which are simply eating into limited marketing budgets. Attribution allows you to…

  • Justify marketing budgets and optimize marketing activities
  • Make smarter bids for digital campaigns
  • Effectively measure key performance indicators

Proper attribution solves common marketing inefficiencies


As the saying goes, “You can’t improve what you don’t measure.” Marketing with little or no information and analysis leads to spending marketing dollars inefficiently and ineffectively. Proper marketing attribution solves at least four major problems:

1. Cluster analysis can result in flawed targeting

Until recently, marketing best-practices dictated the use of cluster analysis, a statistical technique that assigns prospects and customers to groups, or clusters, based on characteristics that are common to individuals within that group. Although the goal of cluster analysis is to group consumers with similar traits into a target segment, in reality each individual within a cluster is just that—an individual. Even though individuals may be grouped in the same cluster, their purchasing behaviors may be completely different.

Despite such obvious limitations, cluster analysis became popular because it was simple and practical to calculate clusters using commonly available software running on the computing power at the time. Basically, something was better than nothing. But as business has become more competitive and customers have become used to customization, cluster analysis is no longer adequate. Modern customers expect to be treated as individuals.

2. Attributing sales to the last touchpoint is flawed

Standard practice in the industry has been to attribute a sale to the most recent marketing touchpoint. But by creating brand awareness, earlier touchpoints may have had a significant impact on the consumer’s decision to purchase. In fact, a consumer may have been planning to purchase regardless of whether he or she saw the latest online ad.

3. Separating the effects of multiple touchpoints is difficult

In today’s digital world, the average sale results from more than 30 touchpoints. Without better predictive models, marketers simply don’t know which activities were most effective in driving the purchase decision.

4. Knowing which ads reached a specific consumer

In a perfect world, we would know exactly which customers saw a particular online or TV ad. But, to date, this information has been difficult or impossible to come by. The nature of digital marketing and media placements can often be at odds with data-minded marketing professionals. Views and clicks in the digital world are aggregated and anonymized, and television and print media show only broad demographic information about potential influence.

Marketers are forced to create landing pages with information-collection forms, but that practice offers concrete information only if the visitors are motivated to submit the form. If the visitor doesn’t complete the form, all you’re left with are website analytics—raw numbers that don’t provide a picture of the full impact of a particular ad in an integrated campaign with multiple touchpoints.

Use machine-learning to map the path to purchase

For effective marketing attribution, marketers need to develop highly accurate predictive models. Statistical methods, which add a score for each separate personal characteristic, are not up to the task of modeling complex human behavior. But machine-learning models capture the complexity of human behavior, analyze the impact of many touchpoints, and identify which marketing activities most influence a sale.

Traditionally, data scientists built machine-learning algorithms manually. That process can be frustratingly time-consuming, with some projects taking months to deliver. By the time the algorithm is ready, it may already be obsolete.

Marketing needs a faster way to build the algorithms—a process that isn’t so manual. The answer is automated machine-learning (AML), a technology that automatically constructs algorithms from historical data, sometimes in as little as a few hours instead of days or months.

Attribute accurately with AML

AML empowers users—including marketers—of all skill levels to make better predictions faster. By automating many of the skills traditionally applied only by data scientists, AML provides the fastest path to data science success for users who understand the business and the data.

AML allows you to create sophisticated marketing attribution models to perform complex “what-if” analyses that quantify the effectiveness of different kinds of marketing activities and different combinations of marketing touchpoints. It can help you…

  • Start with a baseline. To begin, you need to determine a baseline—the sales that would naturally occur without any marketing activity. You can use AML to analyze the impact on sales if you remove all the marketing touchpoints.
  • Determine the marketing contribution to sales. This is the difference between actual sales and the calculated baseline sales. The more effective your marketing activities, the more sales are boosted above this baseline.
  • Assign a contribution for each touchpoint. AML performs a variety of what-if calculations to determine the impact on sales if you remove one, or multiple, touchpoints.

By using historical touchpoints and outcomes, AML automatically finds patterns, creating a model that predicts sales depending on the touchpoints that apply to each lead. Using the model, you will run a number of “what if” scenarios using different touchpoints to predict how different combinations of touchpoints impact sales.

Attribution performed in this way lights up your sales funnel, giving you insights not possible only a few short years ago. Attribution creates a clear guide to show you which marketing programs are worth spending money on and which are not. With this information, you do more of what works and reduce or eliminate what doesn’t.

To learn more, download the new report or watch the on-demand webinar, Multichannel Marketing Attribution, to find out how automated machine learning easily finds the most accurate models to achieve precise, meaningful attribution.

Let’s block ads! (Why?)

MarketingProfs Daily: Marketing Strategy

Add Comment