In the last two years, almost every technology company has started using the terms artificial intelligence and, related to that, machine learning. In fact, there has been a 1,400% increase in the mention of these two terms during earnings calls of public companies during the same period. Gonzo readers who follow fintech companies know that they rely heavily on these capabilities.
First, some definitions:
An algorithm is simply a defined series of steps or tasks to be performed. For those of us who don’t hang out with the IT wonks, a simple non-technical example would be a cooking recipe.
Artificial intelligence is essentially the process of taking pre-defined algorithms, applying them to your data, and using the results to make decisions or recommendations.
Machine learning is the use of algorithms and analytical models to perform specific analysis without any explicit instructions, relying instead on previous results and patterns. In this discussion, we are going refer to this machine learning powered artificial intelligence as simply intelligent algorithms. If you just thought of HAL in 2001: Space Odyssey telling Dave he was getting pissed, you’re there!
The speed at which capabilities are growing in this area make it something that bank managers need to understand and manage, which leads to our first “Need to Knows”:
Banks and credit unions can leverage the power of these intelligent algorithms today! There is no need to wait for some magical and massive scientific breakthrough. In fact, large banks and fintech firms are using these capabilities heavily for marketing, delivery, pricing and other things. Your turn.
This is not something that will magically appear one day from the bowels of IT. It will be used well because management teams get involved and guide it.
Avoid the “Road to Nowhere.” Don’t do any of this just to see what you can find. All algorithms need to be purpose-built. If a financial institution needs a portfolio loss prediction model for commercial loans, the predictive model has a clear purpose, and we know what data would be needed to make such predictions, what outcomes would be considered significant, and what modeling techniques would be relevant and meaningful for this algorithm. This is often referred to as “hypothesis-driven AI,” where you define what you need/might want to know, and AI aims at that result. By just fishing around without such clear methodology, it is easy to end up on the Road to Nowhere.
Beware the Black Box of Magic. Several vendors are touting the prowess of their tools where they can somehow magically garner insights from any data source. While this approach may be successful to mine standard information such as social media feeds, this type of black box approach, which is not custom built for industry specific and business model specific needs, generally fails at banks.
You will need internal data scientists. If a financial institution wants to pursue intelligent algorithms but has no one on the team who understands other concepts such as linear regression, correlation or statistical significance, then the odds of success are quite low. The notion that somehow machine learning techniques can magically usher in artificial intelligence at the bank or credit union without the need for human data scientist resources is folly. Oh, and you can’t just outsource this.
There are lots of use cases that you can leverage today! Some examples:
Predictive Models: Firms such as Capital One have been successfully leveraging predictive models since the 1990s (the stone age of artificial intelligence) to sell very specific credit cards and rewards programs to new customers based on spend, usage and other data. A financial institution can create and implement its own predictive models by utilizing off-the-shelf tools available in the market today, especially on its card portfolio, where the profitability and transaction volumes are rich.
Prescriptive Models: In just about all banking related conferences, it is close to impossible to escape the buzz phrase next best action, which at its core is a prescriptive model driven outcome. Next best action is simply an integrative thought process that analyzes recent events/encounters with customers and produces a recommendation for a next action step/offer, like a savings product, car loan or wealth management service, at the right time for the customer. You can start providing next best action prompts to sales employees today, or, if you already do this, you can analyze results and tweak algorithms to make the prompts better.
Voice Mining: An often-overlooked area of analysis that matured in the mid-2000s is voice mining, also called speech mining. The English language only has 44 unique sounds, called Intelligent algorithms can review voice recordings and mine these phonemes in real time to provide the financial institution with in-depth information on why a customer called, identify patterns, and help resolve root causes of issues to reduce operating costs. The days of walking around the institution’s call center asking staff, “What is going on today?” can become history by implementing these highly intelligent and highly mature algorithms.
Text Analysis: The same phoneme-based algorithms can intelligently query a vast amount of freeform text information and provide in-depth insights amongst copious amounts of such data that financial institutions possess. This same technology is what is behind several chat-bots that automatically respond to email and text-based customer inquiries and subsequently reduce operating costs for many organizations.
Bottom line: all financial institutions can get out there today, follow a disciplined hypothesis driven methodology and leverage existing technology to create and deploy purpose-built intelligent algorithms through clearly defined use cases. There is no need to wait for some amazing artificial intelligence technology to arrive, because it is already here and is eagerly waiting for us to deploy it in a disciplined manner.