Case Study: Advertising. Supplementing Human Judgement with Machine Learning

There’s a great deal of interest in the business world in adding machine learning (ML) capabilities to data analysis applications so computers can leverage the terabytes of data companies generate on a daily basis to improve their products, services and/or internal processes. Some futurists have speculated that the combination of ML and big data has the potential to create autonomous systems capable of extracting knowledge from data and applying that knowledge to improve the system’s performance, all without the need for any human intervention.

There has been a lot of FUD and hype surrounding the topic of autonomous systems, and the purpose of this blog post is not to rehash those debates. Rather, let’s examine exactly what ML is capable of today and how it can be used to supplement, but not replace, human judgement when it comes to improving business results. For while computers are indeed much faster than humans when it comes to processing data and making complex calculations, they’re not better than humans at applying knowledge and experience to improve an outcome. That said, combining machine learning with human domain expertise can deliver remarkable results to a variety of industries. In this blog, I’ll explain how machine learning can improve results for advertisers, and in future posts I’ll touch on how it can benefit other industries.

Improving campaign effectiveness with Machine Learning

Thanks to Ad serving applications like Google’s DoubleClick, companies now have access to all kinds of data that can be leveraged to deliver extremely targeted advertisements and determine how effective those ads were in driving consumer engagement. Ad server logs can quickly tell businesses not only how many consumers are clicking on a banner ad, but also whether or not those clicked ads were converted into actual purchases.

But analyzing ad server data alone is not enough to improve a campaign’s results. Other factors to be considered include a customer’s purchase history and their demographics, but that data is typically captured by other systems (say a CRM platform like Salesforce), which a human would then analyze along with the ad server data. By leveraging data from other sources, a more detailed analysis is possible, like identifying a previously untapped customer base or determining how to divide campaign budget between traditional, retail and digital advertising for best results. For while machine learning algorithms are quite adept at sourcing and curating data from different databases, they currently can’t analyze that data and use the findings to adjust a campaign. That kind of domain expertise only comes from the experience and applied learning that a human trained in advertising and marketing gains over time.

spml_big_query_info_graphics_v3

Using this model, Advertisers can help answer complex questions such as what is my white space? and quickly adjust the campaign to improve the conversion rate.

Where ML can serve to augment human judgement in this use case is in reducing the amount of time needed for a human to conduct the analysis. As mentioned, much of the advertising data that can be used to improve the results of future campaigns is typically stored across different databases, often in different formats and in staggering amounts. Gathering the required data, formatting it for consistency and developing dashboards to present that data for easy analysis is a time-consuming task that requires an IT department to write scripts to source and present the data. The time and effort involved in this process requires ad campaign data to be processed in batches that can take weeks or even months to complete, so the lessons learned from a campaign are only realized after it’s over. What’s worse, by the time that data is used to guide a future campaign, it may be out of date and inapplicable.

But what if ML were brought in to help an analytics platform learn how to identify, classify and present data without the need for human intervention? What if the data from those different systems were pulled and curated in real-time? A campaign manager could see results as they occur, analyze the data to see if the campaign is effective and tweak the campaign to improve its performance while still in the field. The potential cost savings are significant, as faster analysis through ML could eliminate the need to invest in a future ad spend to address the missed opportunities from an earlier, less-effective campaign.

Stay tuned for future examples of how machine learning can help businesses speed up their data analysis to more quickly improve results and improve profitability.