Home / Banking Strategies / The lean, mean learning machine: The rise and resonance of machine learning

The lean, mean learning machine: The rise and resonance of machine learning

Jul 6, 2017 / Technology

Lately there’s been quite a lot of talk about artificial intelligence (AI). Publications are quickly assembling perspectives, ideas, suggestions and conclusions on how AI can and will be used over the next few years. Current AI research has a theoretical arm that tackles broad questions about modeling basic human thought patterns.

But a vibrant academic and business community is making more mature advances and applying them to very real, tangible problems to achieve quantifiable successes. And machine learning, an AI subset, represents the most breathtaking of those focus areas. But what is machine learning exactly, and how does it work?

Welcome to the machine: Learning the basics

Broadly defined, machine learning teaches computers to recognize patterns in a way similar to how humans do. A computer program, focused on a specific task, performs an operation and calculates its performance of that function—while making adjustments along the way to increase performance and thus minimize the program’s “error” in subsequent steps. This is characterized as “learning” since the program modifies its behavior based on input received and output produced.

Numerous examples already exist where we use machine learning technologies, possibly without even knowing it. Here’s how my typical day might look:

  • In the morning, I use Alexa’s natural language processing to tell me the day’s weather.
  • With morning coffee and my iPad, I’m presented top news items to read, selected specifically for my tastes and interests.
  • Driving to work, Waze updates me on the road conditions ahead.
  • In the office, I unlock my MacBook with a fingerprint scan, as it has learned my various finger positions from active use.
  • I check my email, disregarding anything sent to me but archived in my spam folder automatically by our system’s filtering function.
  • LinkedIn provides me with a list of “people I may know.”
  • Checking an Amazon order, I’m provided with a list of recommended items.
  • My Facebook activity shows that I’ve been recognized and “tagged” in my friend’s photos from a baseball game I attended.
  • Heading home, I ask Siri to tell my wife that I’m on my way. Siri recognizes my voice and texts my wife.

The FinTech space: Where AI is A-1

Financial technology (FinTech) already uses machine learning in services such as check image capture and fraud detection. Via traditional analysis, a consumer’s common usage can be compiled and any behavior that falls outside the norm gets flagged for further review. Machine learning comes in as it draws broader conclusions over usage. For example, if the common pattern is disrupted by an occasional vacation, a machine learning algorithm might initially see this as an anomaly but would adjust over time, and learn to see it as less unique in the future.

Image analysis presents another area where machine learning thrives. Google Photos uses models it derives from learning algorithms, applies them to our photographs to detect faces, and tags those faces so we can do image searches for specific people. Called Image Tagging, this technique is also being applied to financially-oriented image capture such as with a check, credit card and I.D. Training the models to better identify and categorize these images makes them smarter and better able to detect slight modifications or irregularities. 

More broadly speaking, banks and credit unions are finding opportunities to help inform business and product decisions via machine learning. Since those institutions sit on years of user financial data, the detailed analysis of it can yield some very interesting, targeted opportunities.

McKinsey and Company’s “Executive Guide to Machine Learning” highlights such usage by European Banks:

“In Europe, more than a dozen banks have replaced older statistical-modeling approaches with machine-learning techniques and, in some cases, experienced 10 percent increases in sales of new products, 20 percent savings in capital expenditures, 20 percent increases in cash collections, and 20 percent declines in churn. The banks have achieved these gains by devising new recommendation engines for clients in retailing and in small and medium-sized companies. They have also built micro-targeted models that more accurately forecast who will cancel service or default on their loans, and how best to intervene.”              

Since the models are trained on usage data, and the algorithms are tuned to analyze enormous data sets, the programs can look for and predict patterns all the way down to an individual customer or member. Compare that to a data analyst, who would spend weeks or months trying to determine that through traditional statistical methods.              

You’ll also see progress as the financial world moves towards a more image-rich environment, where remote deposit capture of checks, and the digital collection of receipts and other traditional paper-based documents for later categorization. Here, machine learning techniques are increasingly used to bolster not just efficiency but also the accuracy of processing that data. Think of optical character recognition software (OCR) that learns from mistakes and improves itself with every read.

Putting it all together: Subtle differences, profound outcomes

               

So what’s the difference between machine learning and a traditional statistical software-based approach? It’s subtle—but the outcome profound. 

With traditional software, recognition of any phenomena requires the developer to essentially think through all possible scenarios. If a developer hasn’t written the software to account for that unusual scenario, then unintended consequences present themselves. 

But with machine learning, the software is taught what the ideal outcome is and then figures out all the possible paths to get there. It makes mistakes, learns from them and adjusts itself to minimize those mistakes in the future. This results in software that developers can build without thinking through every possible scenario or permutation. And it adds up to far-reaching, exciting implications.

As the software gets smarter, our offerings to financial institutions will benefit; less brittle, more hardened applications will also be more future proof because of their in-built ability to recognize and adapt to subtle changes. Machine learning in the financial services space will usher us into an AI age in a tangible, real-world way that produces real results.

And so if the attitudes of the industry were once skeptical, or its leaders puzzled, there is good news: We too can learn, adjust and minimize any mistakes in the future as we bring AI into the service of a greater financial good.

Want more Banking Strategies? Sign up for our free newsletter!

Danny Piangerelli is the co-founder and chief technology officer for Austin-based Malauzai Software. He leads the technology team and is responsible for all aspects of Malauzai’s technology infrastructure and has more than 15 years of engineering and enterprise-class technology experience.