Contextual Integration Is the Secret Weapon of Predictive Analytics

by   |   January 8, 2016 5:30 am   |   1 Comments

Dominik Dahlem, Senior Data Scientist, Boxever

Dominik Dahlem, Senior Data Scientist, Boxever

The past decade has brought services such as Amazon and Netflix to the fingertips of nearly every consumer, inviting impulse purchases and marathons of TV binge watching. What made these services so successful was not necessarily product availability or quality of content. The real differentiator was context and the use of predictive analytics. Amazon has long been praised for its recommendation engine, which has prompted many users to fill their shopping carts with additional items that Amazon suggests based on their purchasing profile. Netflix introduced a similar concept by providing contextual suggestions to home entertainment. The technology ensures that subscribers stay firmly planted on their couch when one show ends by offering up a recommendation developed by a continuously learning, preference-based predictive algorithm.

Context can be nearly as important as price when consumers are making purchasing decisions. A recent Boxever study found that, outside of price, offers have the most impact when they address what the customer is already doing. Data scientists are now mastering ways to create machine-learning algorithms to provide real-time, highly personalized offers for different customers and, in many cases, customers who may fall into different profiles for different purchases. For example, a person probably has vastly different preferences when it comes to price and convenience when traveling for business than when booking a family holiday. Machine-learning algorithms can identify patterns that know which profile that particular person fits at any given time and provide an accordingly enhanced and relevant experience.

At the core of many predictive analytics technologies are approaches that have been around for decades or even centuries, such as Bayes’ theorem, the method of least square, artificial neural networks, and backpropagation. Predictive analytics has been around for years, but only now have data teams begun to refine the process to develop more accurate predictions and actionable business insights. The availability of tremendous amounts of data, cheap computation, and advancements in artificial intelligence has presented a massive opportunity for businesses to go beyond their legacy methodologies when it comes to customer data.

Moving From Traditional Aggregation to Contextual Transformation

Related Stories

How to Harness Predictive Analytics and Become a Big Data Dynamo.
Read the story »

Predictive Analytics Captures Value in Every Business Sector.
Read the story »

With Predictive Analytics in the Cloud, the Sky’s the Limit.
Read the story »

Predictive Analytics Takes B2B Marketing from Good to Great.
Read the story »

In most organizations, customer intelligence data typically is collected and analyzed in silos. Web teams look at different statistics than the loyalty team, which looks at different data than the sales, marketing, and customer service teams. This fragmented approach to customer intelligence not only lengthens the time it takes to act on data, but also makes the actual intelligence less accurate and actionable, and strips the data of context.

Today, that’s all changing. As predictive analytics continues to rapidly evolve, the organizations that cling to yesterday’s data aggregation practices will put themselves at a significant disadvantage.

A major part of this transformation is the realization that data needs to be looked at from as many angles as possible in an effort to create a multi-dimensional profile of the customer. As a consequence, we view recommendations through the lens of ensembles in which each modeled dimension may be weighted differently based on real-time contextual information. This means that, rather than looking at just transactional information, layering in other types of information, such as behavioral data, gives context and allows organizations to make more accurate predictions.

Machine learning also is contributing to the improvement of accuracy in predictive analytics. Algorithms have the ability to learn in real time and find solutions and connections among the incoming data sets. Accuracy is not only improving, it’s bringing a new element to the table – contextual immediacy.

For example, using traditional, data-based marketing approaches, a retailer would blast mass promotions based on high-level demographics, such as age, location, and sex. This “DEAR NAME”, because you are living in “CITY”, you should buy tickets to “EVENT” approach is hardly personable and, often times, makes people tune out your communications.

Today, retailers have the ability to do so much more. Leveraging customer intelligence and predictive analytics, a retailer could send notifications to users’ phones whenever they are within a mile of the store, offering them a discount on a new product line that they know the consumer will be into based on past shopping experiences. Another example is the airline that knows you fly back home every June for your parents’ anniversary, and sends timely booking reminders when fairs are low, and offers recommendations and deals on things to do with your parents after you arrive.

With trails of data being dropped by consumers every second across social media, retail websites, mobile apps, and more, it is imperative that data teams enable their organizations to capture layers of both traditional and behavioral data to provide the business with a complete and highly intelligent view of the customer. The challenge lies in analyzing the wealth of unstructured data – which typically is text heavy and/or event-based and much more difficult to organize than transactional data – that’s available due to the boom in consumer technology over the past decade.

Like the long history of predictive analytics, the core process of collecting data and developing predictive insights is not new either. A sound approach follows the scientific method, starting with understanding the business domain and the underlying data that is available. Then data scientists can prepare to test a particular hypothesis, build a model, evaluate results, and refine the model to draw general conclusions.

Likewise, the desire to predict behavior is nothing new for scientists, analysts, or even marketers. The key difference in predictive analytics today is the need for accuracy, relevance, and real-time learning. Using demographic and transactional data no longer will be enough for businesses to compete in the age of the hyper-digital consumer. Analyzing that added layer of behavioral data will provide the context needed to offer accurate, efficient, and actionable insights.

Dominik Dahlem is a Senior Data Scientist at Boxever, a customer intelligence and predictive marketing company for travel companies and retailers. Dominik specializes in network analysis, machine learning, high-performance computing, data mining, and travel analytics. Prior to joining Boxever, Dominik served as a Research Scientist at IBM and as a postdoctoral fellow at both MIT and Trinity College Dublin.

Subscribe to Data Informed
for the latest information and news on big data and analytics for the enterprise.

Big Data predictions for 2016

Tags: , , , , , , , ,

One Comment

  1. MM
    Posted January 9, 2016 at 11:58 am | Permalink

    Great article, thank you!

Post a Comment

Your email is never published nor shared. Required fields are marked *

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