Back to Gartner for Marketers Blog

Pick A Personalization Engine Your Data Scientists Will Love

By Jason McNellis | July 28, 2020 | 0 Comments

MarketingData and Analytics

Predictive analytics and data driven optimization are getting easier. At least if you have invested in the right personalization engine (for recently completed research on personalization engines see Critical Capabilities for Personalization Engines – subscription required). These platforms give marketers control over three powerful types of analytics:

  1. Advanced product recommendations that leverage customer profiles, contextual data and even unstructured data streams to improve what content or products are shown to a given customer.
  2. Diverse predictions including discount sensitivity, customer lifetime value (LTV) and product affinities for marketers to better define audiences, improve targeting or segment results.
  3. Strong support for optimization through advanced testing. This includes multivariate testing and multi-armed bandits that can help marketers achieve positive results quicker.

These features are powered by a slew of machine learning techniques, including collaborative filtering, gradient boosted machines, deep learning, contextual bandit algorithms and natural language processing (NLP) (for an explanation of a few of these and why they are important to marketer’s see Turbocharge Your Marketing and Personalization With 3 AI Algorithms – subscription required).

 

Machine learning is becoming simpler and more user-friendly as these advanced techniques are built into personalization workflows or given dedicated UI’s to facilitate no-code customization. Will approachable machine learning diminish the value of data scientists in your organization?  I don’t think so.

 

You can select a personalization engine that will help both your data scientists and marketing technology stack perform at a higher level. Look for these four characteristics:

  1. Support for bring-your-own-algorithm. Most data scientists get much more satisfaction from building machine learning as opposed to deploying machine learning. Deployment can be particularly tough for models that use in-session contextual or behavioral data. Several personalization engines simplify this challenge by importing and using external models built in python, Jupyter notebook or R; the same languages likely used by your data science team.
  2. Simple algorithm testing. Which is better: the propensity model built into your personalization engine or the one from your data science team? Test them head-to-head and find out! Challenge your data science team to identify audiences and use cases where they can beat your platform. Every time your data science team wins so do your customers.
  3. Built-in performance reporting. Automated dashboards that demonstrate the impact of machine learning is another widespread feature of personalization engines. Reporting on algorithm testing and holdout testing (personalized by machine learning versus a control) are common, including tracking performance over time. Some personalization engines also offer built-in segmentation reports so you can easily view model performance by segment, a useful resource to generate hypotheses around potential model improvements.
  4. Access to consumer data. Personalization engines often contain valuable customer data that can be used for machine learning and generating insights. This includes unified customer profiles which likely contain linkages to data sources not available in enterprise data warehouses, such as session-level response data. Some personalization engines provide API access to this type of data and several have invested in organizing this data into data table structures that are easier to understand and access.

 

A symbiotic relationship is one that benefits both parties. With a little planning, you can enhance your POC process to ensure a symbiosis between your personalization engine and data science resources.

Leave a Comment