Despite all of the recent buzz, big data isn't new. As far back as the 1880s, the U.S. Census Bureau was collecting and analyzing data on citizens. Even then, the volume of data was overwhelming. In fact, the Bureau estimated that it would take at least ten years to process all the information. That was until Herman Hollerith created the Hollerith Tabulating Machine.
From that moment on, data and technology would intertwine forever. Now the world is saturated with technology, and each piece of technology is collecting and sending data.
Insurance companies collect data on individual drivers to calculate their risk and insurance rates. We do the same with credit scores. Facebook looks at who we are friends with and suggests who else we might like to connect with. Google and Amazon analyze past searches to determine what products we might want in the future.
Many people use data at their jobs to make financial projections, develop consumer segmentation, or perform some other ad hoc analysis. They look through past data to find past patterns and ask themselves "What will the future look like if this pattern continues?" As technology infiltrates more of the average consumers’ activities, more personal data will be collected and used by businesses. This will change the game.
What does tomorrow hold?
The best predictor of the future is the past. Knowing why a segment of consumers took a particular action can be valuable for planning future campaigns or product developments, but the act of applying past data to future scenarios has largely been manual and involved some guessing and assumptions. Today, data sources are everywhere, and they are a part of a connected world: the internet of things.
Predictive analytics will take information from the multitude of data sources, analyze it and apply insights to anticipate how consumers will act next. These insights help answer questions such as "What is going to happen next?" or "What is the best that could happen?" For example, if a consumer recently changed their job title, got married and they have been making steady contributions to their savings account, this person might soon be in the market for a home loan. The results of a similar analysis can be seen on YouTube when the site recommends the next video to a user based on their viewing history.
For many years, businesses have focused on historical reporting and ad-hoc analysis. With the proliferation of data generating devices and connected consumers, knowing what happened in the past or why it happened are not enough anymore. Predictive analytics is powering data-driven decisions and giving many companies a competitive edge.
Performing predictive analytics on thousands of customers involving tens-of-thousands of data points is only feasible with machine learning and artificial intelligence (AI).
How to develop predictive analytics
As beautiful as this scenario sounds, achieving it requires a herculean effort. First, companies need a mountain of data, because the more they have, the more likely it is that they'll be able to find true causality.
Once companies can collect data, they need:
- a place to keep it,
- a way to secure it,
- a way to organize it,
- an easy way to access it, and
- a method to analyze it.
Data at this scale and the demands of real-time analysis require always-on AI and machine learning.
Machine learning is similar to the scientific method.
- Data comes in.
- A model is made (hypothesis or prediction).
- The model is tested.
- Results from the test are gathered (learning).
- The model is refined based on those results (parameters are adjusted).
- The process is repeated.
Through this process, the models close in on the truth of what is happening which allows for more accurate predictions.
How the megabanks use machine learning
This scenario isn't exactly the future of data. It is in action today. In fact, community financial institutions’ biggest competitors, megabanks, are already taking advantage of their mountains of data to eliminate fraud, provide better service, develop products and sell more effectively. For example, HSBC improved their fraud detection and false-positive rates by using AI to monitor trends of their cardholders, and J.P. Morgan has designed an AI called COiN (Contract Intelligence Platform) to review annual commercial credit agreements. While it is still new, early research has shown that what once took 360,000 hours can be processed in seconds.
How can smaller institutions keep up?
Machine learning is only as good as the data that powers it. This requires a vast amount of data, sophisticated management systems and the development of models. Implementation of machine learning at a community financial institution is costly and laborious. So, how can community banks and credit unions incorporate machine learning?
With the right strategic direction and resources, community financial institutions can develop capabilities in-house for historical reporting and ad-hoc analysis. However, when it comes to predictive analytics, limited data volumes create a glaring gap. Thus, partnerships are the way to go.
Partnering with a company that has data from hundreds of community banks and credit unions from across the country is crucial for machine learning to be effective. While it is possible for community financial institutions to purchase data, this solution lacks tailored learning to suit your institution. Purchased data lists are mostly generic, and the data models are one-size-fits-all.
Finding a fintech partner for machine learning gives community financial institutions the best of both worlds. A partnership allows them to access data models developed on a large data set that would otherwise only be available to the largest financial institutions, enabling community banks and credit unions to utilize predictive analysis to reach consumers and compete with megabanks.