Part 2: Why Analytical Modeling Will Never Be The Same

“Why Go To A Machine When You Could Go To A Human?”

Ray Bradbury poses the exact question I want to discuss – and answer – in this post.

As I discussed in Part 1 of this series, in the world of consumer lending, machine learning has the potential to be a significant disrupter to the traditional methodology and infrastructure of analytical modeling. As I asked then: is there a need for a development sample in the world of machine learning and adaptive predictive models?

The business value extracted from creating development samples in traditional model building methodology is still relevant today. The difference is that with today’s capabilities, the development sample is merely the starting point of a continuous learning process fueled by sophisticated technologies and human expertise.

The power of predictive analytics is the ability to accurately answer questions by processing copious amounts of complex data. The power of the human mind is in our ability to formulate relevant and meaningful questions…

• What specific business problem are we addressing?
• What specific business goals are we setting?

Let’s look at why humans still play critical roles in creating development samples for predictive models.

To Get Analytical Modeling Results, You Need A Strong Cast

To achieve financial impact, three “human” roles are necessary for effectively using predictive analytics and machine learning. In some cases, roles may overlap, or a single individual or team may assume more than one role – but each is critical to results. These roles are:

The Business Expert: Prepared with specific subject matter knowledge about processes, opportunities and challenges of her industry.

The Data Specialist: Familiar with what data is available and the quality and breadth of the data, based on both the industry and the specific organization.

The Computer Scientist: Equipped with statistical and mathematical background for building models and structuring continuous improvement through machine learning.

To better understand these human roles and how the people that fill them perform within the process of analytical modeling, we reached out to a colleague at Starpoint Solutions. A company rooted in technology staffing, business applications solutions and career development, Starpoint is now breaking new ground in offering predictive data analytics as a service.

Certainly, here was the mix of human expertise and new technology that we wanted to see in action. We sat down with Henry Zelikovsky, Starpoint CTO, to listen and understand exactly how they manage “human capital” and expertise with their new technology in the predictive analytics process.

Starpoint’s cloud-based service model reinforces the complementary nature of technology and people. People must have the expert knowledge and understanding of how prediction technologies work to improve the accuracy of business outcomes and identify opportunities for better business performance. As Zelikovsky explains:

“We use technology and machine learning in order to process an enormous amount of data in a predictable manner to give to a person better qualified information within our decision support tool…but the decision is from the person.”

When we asked Zelikovsky to provide an example for our readers, he provided a case study in consumer lending.

Humans And Machines Team Up: A Case Study

A financial institution supplied a data set of people who applied for a loan and asked the following specific question:

Who are the people that are most likely to fail to repay their loan?

If the business experts can predict with a high degree of certainty which people will not be able to repay their loan, the financial institution would save significant expenses by not offering the loan to that segment — or by offering a smaller loan.

What is our capability to address our question from a data perspective? What is the meaning of the data?

First, the business experts need an understanding of what data is available and what data is relevant within the organization. This requires both industry-wide and organizational knowledge.

How then can we use this data to answer the specific business question: “Who will fail to repay their loan?” Humans team up. The data specialists and business experts create a data structure compiled of sample data sets. Constructed with “human” experience, specific knowledge and expert judgment, this sampling technique – similar to the process of creating traditional development samples – is a critical step in translating a business challenge to a mathematical model.

Henry reinforced the importance of sampling the data sets on a small scale to determine the possibilities in the data. From this small sample of data, a knowledgeable person can calculate and interpret results, which becomes the basis for how the machine will “learn” as it processes larger amounts of data.

The next step is putting the computer scientists to work to identify existing algorithms relevant to solving this specific business problem – the risk of offering a loan – or to create new algorithms for the predictive model.

Once the specialists determine the appropriate methodology for the algorithmic process, the machine is used to automate the processing of the full data set. As these samples are evaluated, sophisticated technologies build and continuously improve predictive models through a technology-driven learning capability.

What did Starpoint’s solution deliver?

Starpoint created a five-tier ranking system that determined the likelihood of the financial institution’s customers to repay their loan. The system identifies with more than 95% accuracy those customers who will not repay their loan.

Analytical Alchemy Drives The Process

The business experts from the financial institution must fully understand and actively challenge the algorithms’ “reasons” for the ranking. In my opinion, this process is the most critical (and most fun!) – as it will ultimately determine future decisioning about loan offerings. The level of human expertise and sophistication within the business to judgmentally augment the algorithm’s prediction is the intersection of “art and science.”

The future value of this machine learning solution is that as new prospects approach the financial institution for a loan, they will be “ranked” based on the continuous learning, experience and existing rankings of like customers’ ability to repay. As more customers are included in the model, the learning process continues and as a result the accuracy of the rankings improve.

The financial institution now has the predictive power to:

• Decline the loan application
• Give a different amount for the loan
• Give a different rate for the loan

“Cognitive computing systems learn and interact naturally with people to extend what either humans or machine could do on their own. They help human experts make better decisions by penetrating the complexity of big data.” —IBM

Of course, machine learning for predictive analytics is a systematic approach that enables one to utilize large and unruly data sets, but the real value is deeper. In every business case, the first round of results leads you to another question – and, as Henry Zelikovsky reminds us, “there is always a question behind a question.”

Machine learning is a game changer, but not because it will replace people, or drastically change the way in which we approach analytics, but because it will enhance our ability to do those things we already do…

• Asking more questions
• Learning through processing of large, unruly data set
• Interpreting the results
• Determining actions to take

And all this is still done by experts with the subject matter knowledge qualified to judge the end result. People are the key to making machine learning meaningful. The relevancy of findings is in the interpretation and subsequent action.

This process – articulating, challenging and seeking to answer critical business questions – is something we are passionate about. So, stay tuned, and we’ll bring you the latest information and insight.

Part 3 of this series will look at the processes, management discipline, and strategy for creating value through analytical models in our real-time, machine-learning age filled with new and complex data sources.

*************************************************

Widely recognized for creating Citigroup's Global Decision Management, Marcia Tal is founder of her data analytics consultancy, Tal Solutions. Marcia specializes in helping organizations create revenue and profit from analytical insights. For more on her analytics solutions, click here.