Redpoint Logo
Redpoint Logo

Jun 26, 2020

What is Automated Machine Learning?

Definition: Automated machine learning (AML) automates the process of applying machine learning to real-world problems. AML covers the complete pipeline from the raw dataset to the deployable machine learning model. The high degree of automation in AML allows non-experts to make use of machine learning models and techniques. (Wikipedia)

This two-part blog series will examine the Redpoint approach to machine learning with a solution that eliminates the need for data scientists to build complex models that need to be re-built as dynamic customer journeys evolve. By stripping away complexity, the Redpoint solution puts the power of analytics into the hands of operational marketers – and thus closer to the customer.

The jury on artificial intelligence (AI) is in. Data-driven organizations using new technology to drive new revenue streams are fully on board with AI as a competitive differentiator. According to a recent McKinsey & Co. survey, 80 percent of companies that have adopted an AI marketing and sales use case report a revenue increase – with half reporting an increase of 5 percent or more. MIT Technology Review examined AI use cases in a “Global AI Agenda” survey of more than 1,000 AI leaders and found that personalization of products and services – arguably the heart and soul of marketing – was a leading contender. In financial services, for instance, 58 percent of respondents cited personalization as the leading AI use case.

Despite AI having proven its efficacy across the board and with widespread confidence in its ability to deliver business results, organizations struggle with scaling an enterprise use case. One reason for the struggle is the general misconception that AI still must depend on data scientists to build, code and program highly complex models. Companies that espouse this route often discover a self-fulfilling prophecy; the complexity they invite at the outset becomes an insurmountable obstacle to enterprise wide scaling.

Evolutionary Approach to Machine Learning

At Redpoint, we believe there is a better way. Automated machine learning (AML) in the Redpoint rg1 solution scales beyond proofs of concepts and special projects with code-free models and automated algorithmic optimization that democratizes the utilization of analytics for the everyday operational marketer.

This blog series will break down the important characteristics that define automated machine learning, examine the process itself, explore the types of models favored by marketing and sales, look at the opportunities, and how algorithmic optimization allows for automated re-training without human intervention. A full understanding of how AML delivers enterprise value will allay the misconception that AI must entail a repeat cycle of manually built models that go stale over time, requiring a continual infusion of resources that deliver minimal value.

Three Tenets of a Machine Learning Model

There are three basic characteristics important for any machine learning model, and a closer look reveals where the problem of scale first asserts itself. The first characteristic for success is that a model be predictive. This may seem obvious, but only insofar as it’s true as the core of what analytics is all about; if you can predict something even a little bit better than random, you can better align resources to take advantage of opportunities. This is true for any line of business and any AI use case.

Second is optimization and adaptation, which is where the struggle to scale an AI use case begins to emerge. Companies tend to invest heavily in the initial building of models, but because the world is dynamic they are then faced with having to repeat the process again and again as time goes on and new data brings new patterns to be detected – and they balk at throwing good money after bad. Particularly in marketing and the creation of a personalized customer experience, customer journeys are becoming more dynamic by the day. A static model constructed for today’s journey loses predictive value – goes “stale” – quickly, which will entail an expensive rebuild/retrain of the models in a few short months. Which takes us to automation as the third characteristic for a successful approach, and an imperative for delivering on optimization and adaptation. Automation scales, people do not. Later in this blog series, we will explore in more depth how Redpoint differentiates with automation, in particular automated re-training where models are tuned to perfectly optimize for a chosen metric.

An Automated Process, Reduced Complexity

A closer examination of the standard machine learning (predictive modeling) process itself helps explain why some companies turn to armies of data scientists, and why Redpoint eliminates this necessity by putting its AML solution in the hands of the citizen marketer.

First, of course, is the acquisition and curation of data in preparation for activation. Traditionally, this is where companies lean on data scientists to correctly prepare the data. It is where the Redpoint solution sets itself apart with powerful analytics that determine what data is valuable and what isn’t, without human interaction. Next, the preparation phase – transforming, joining, filtering – similarly requires intensive resources to create the features in the data necessary for comparisons, aggregations and other metrics that need to be composed into the data set.

Typically, the next step is selecting code – the modeling approach, accounting for considerations such as the quality or richness of the data, the sparsity of the data, and the type of questions you’re trying to answer. Here again, most companies rely on very skilled resources to make these judgment calls, which the Redpoint solution does as a matter of course and eliminates the mistakes made by relying on human judgment and experience.

Next is configuring parameters and inputs for the models, which is another step traditionally put into the hands of experienced data scientists vs. having an algorithm that automates the process. After training and review, you select code again and repeat the final steps as necessary until you’re comfortable that the model predicts what you want it to predict. Only then do you deploy the model for activation.

Again, for a code-based algorithm (Scala, Python, etc.) the entire process requires an enormous amount of resources, financial, temporal and human capital. Automating the entire process with code-free algorithms is the key to the Redpoint AML solution democratizing analytics for the everyday marketer. It leverages computing speed to analyze thousands of variations in the models and resolve the most powerful model in short order. This removes the built-in complexity of building models that has historically posed the biggest challenge for organizations to scale an enterprise AI use case.

Hopefully, this short outline has set a context that reveals the current state of the market when it comes to developing predictive assets to improve business performance. In the next blog in this series, we will explore the various types of models in the Redpoint AML solution. We will also look at how the entire machine learning process is fused with an algorithmic optimization that allows for automatic re-training without human intervention and a simulation process that guarantees outcomes are measured against a perfect solution for whatever business outcome you’re trying to achieve.

For more on the topic, watch Make Personalization Easy with Automated Machine Learning, a webinar hosted by Steve Zisk and Bill Porto of Redpoint Global. Steve is Senior Product Marketing Manager, and Bill is Senior Engineering Analyst.


Humanizing a Customer Experience without Humans: Let Machine Learning Take Control

Evolutionary Programming: The Survival of the “Fittest” Data Models

Beyond the Hype: Put the Power of AI into the Hands of Marketers

Be in-the-know with all the latest customer engagement, data management, and Redpoint Global news by following us on LinkedInTwitter, and Facebook.