Why machine learning is critical to multi-touch attribution

Columnist Alison Lohse notes that in today's complex marketing atmosphere, marketers need tools that can quickly and accurately make sense of myriad and disparate data -- and machine learning does just that.

Chat with MarTechBot


Until six or seven years ago, econometric models offered the best way to measure multi-touch attribution. These methodologies, like MMM (marketing mix modeling), turned statistical analyses into predictions and answers to high-level questions: How much revenue is generated from each channel? How much do I need to spend in each channel to optimize my mix? Econometric models rely on complex information and assumptions by human experts, and these models did (and still do) provide valuable insight into big-picture forecasts.

Two recent shifts, however, have necessitated a new way to address multi-touch attribution: big data and user-level analysis. Both require processing power far beyond traditional modeling — beyond, in fact, what humans are capable of on our own. This is where machine learning comes in.

Beyond traditional modeling

Attribution powered by machine learning delivers the flexibility, speed, automation and scalability that big data demands. Traditional models could only operate as quickly as the human brain. An expert chose a methodology that presumably was best suited to his team’s data inputs, put the data through the model, and analyzed the outputs. This process was limited by the expert’s assumptions (which model is best), his or her ability to construe the data, and the time required to adjust the model if and when required. Unsurprisingly, results came in slowly. It was not unusual to run a marketing campaign and only understand its performance a full month later.

Machine learning models, on the other hand, don’t rely on human assumptions. The most precise methodologies, such as the ensemble method, are “smart.” They use different and/or multiple algorithms, depending on the circumstances, to return the most accurate results.

This flips the paradigm. Instead of saying, “We think that logistic regression will give us the best answers,” and then feeding the data into the model, it says, “Here is what we spent, here is what our users did, here are all the conversions,” and decides which model or models can figure out the interactions most successfully. Machine learning models automatically go through hundreds of combinations to find the best fit for the data set, so marketers always receive the best possible analysis.

Faster insights for changing times

Not only does a machine learning approach to attribution serve up better insights, it does it dramatically faster. All the computations, across those hundreds of models, occur in real time or near-real time. Time to results drops from months to hours. This is especially critical when your business and/or the environment are changing quickly.

During an election cycle, for instance, a lot can change in a month. Similarly, if you’re launching a new product, you can’t wait a month to learn that your messaging is tanking with your target market. Or say your campaign is converting beautifully — until your biggest competitor launches a BOGO promotion. Machine learning models can account for, adjust and measure all of those variables on the fly, before you even know that they’re impacting your campaign.

Personalized models

Finally, machine learning makes it possible to analyze conversions at the individual user level. Unlike econometric models, which operate primarily at 30,000 feet, models such as the ensemble method can process the required volume of data at the necessary speed and level of granularity to tell us what is happening on the ground.

Machine learning allows marketers to track and build personalized models for each user or user segment, for example, then integrate tightly with real-time bidding platforms to serve the best ad for that person, get instant feedback and update the algorithmic model based on what happens. No more A/B testing. With machine learning, you automatically and almost instantly know the most efficient vendors, channels, placements and creative for your target audience. The insights and economies of scale are tremendous.

In a big data, user-centric world, marketers need tools that can quickly and accurately make sense of myriad and disparate data. Machine learning does exactly that. When used for multi-touch attribution, it provides the breadth and depth that marketers need to make the most informed decisions at the most critical moments.

Contributing authors are invited to create content for MarTech and are chosen for their expertise and contribution to the martech community. Our contributors work under the oversight of the editorial staff and contributions are checked for quality and relevance to our readers. The opinions they express are their own.

About the author

Alison Lohse
Alison Lohse is COO and Co-founder of Conversion Logic. Alison spent the last 18 years focused on digital strategy for a number of Fortune 100 companies across many industries including telecom, retail, travel, B2B, CPG and tech. Her expertise and focus on client service, advanced analytics, media planning and optimization lends Alison a unique ability to drive digital strategies that scale brands helping them reach a wider audience. Cutting her teeth on digital starting in 2000, she worked across the interactive media practices at Starcom IP, then Avenue A, Razorfish and SMG with a focus on sophisticated media buying through analytics and optimization. Most recently, Alison was the Regional VP of Visual IQ, Chicago where she worked with Conversion Logic’s CEO, Trevor Testwuide. Alison earned an MA from the University of Manchester (UK) and holds a degree in art history from Lawrence University.

Fuel for your marketing strategy.