What Credit Unions Must Do Today to Pursue AI

It is not hard to find credit union leaders who are enthusiastic about artificial intelligence. Go to any conference or seminar and you'll find vendors already promising the future of AI. The enthusiasm is well-placed, but depending on which AI thought leader one cites, the percentage of jobs that will be replaced by AI varies. Serial AI Venture Capitalist Kai-Fu Lee said 40% of jobs will be automated by AI. Oxford University's researchers found the figure to be even higher, suggesting 47% of today's jobs are candidates for AI obsolescence.

Jobs prime for automation are those heavily reliant on data, many of which are readily found in the financial industry. Among them are underwriters, loan officers, accountants and auditors, respectively ranked in likelihood of automation at 99%, 98%, 94% and 94% by the same Oxford study. Time to panic? Not exactly – the Oxford study suggested the invasion, or revolution, will take about 10 years, starting as early as 2025.

Recommended For You

Understanding the state of AI is important in order to contextualize this opportunity. Despite what marketing professionals at the last conference you attended said, AI has still not passed a basic test. This test is called the Turing Test, where humans are fooled into thinking they are speaking with another human, when in fact it is a computer. While this difficult feat has not been achieved at any satisfactory level, there are exciting developments in the AI field, namely in machine learning.

Machine learning adopted a new approach using a technique called artificial neural networks. This new approach uses a programming model to mimic organic brains. Training data is then used to teach the model. Models are tools already applied in credit unions. Things like financial projections and what-if scenarios are typical models, just to name a couple. The industry is also already using training models to recognize pictures at surprisingly accurate levels. How does this translate to finance and banking? Simple. Pictures are abstract representations of data, and data is something credit unions are rich in.

So what is the most important thing a credit union can do today to be ready for what is being called the fourth industrial revolution? Start working on data quality and data hygiene. There are several ways to do this, but ultimately it all comes down to data governance. Whether a credit union achieves that through a single person, a committee or another way does not matter. What matters is that it starts to clean the data and puts controls in place to keep it clean.

The importance of clean data is paramount because it will eventually be used for training and inputs into models. The poorer the data quality, the poorer the insights produced by the models – just like the old adage of garbage in, garbage out. For most credit unions this problem is already present in simple deliverables like reports. Imagine what damage dirty data would do to a model that does not know the difference between good and bad data – especially if it's being used for making decisions or recommendations.

Another step a credit union can take today is to document the models currently in use. Simply put, this means documenting what the credit union does now to make decisions for things like who is credit-worthy and who is not. Consider what data is important and what data is noise. These models are actually already in use, but most organizations do not realize it. What is understood as human experience or written procedures are actually applied models, just not in a format that is ready for machine learning.

How does a credit union document this knowledge for conversion into a future model? It needs to approach the endeavor with humility and patience. Nate Silver's book, "The Signal and the Noise," an important primer on the subject, reminds the traveler on this journey that the road is long and comes with many frustrations, as human experience and written procedures are reluctant to give up their secrets. Converting these processes to models is a must for any credit union serious about pursuing AI. These models, once extracted into their pure mathematical form, will grant any credit union many options. Some options may be exercised today, like automated decisions and prescriptive analytics, without the benefit of machine learning or AI.

The final thing credit unions must do to prepare for AI is to simply manage expectations. Ask any AI enthusiast about the future of applied AI and the answer will nearly always be two to five years. However, the promise of tangible proximity is a well-worn promise heard during each generation of AI, which only leads to the delivery of AI moving further away like a mirage in the desert. While this ever-retreating delivery date serves as a prudent warning, the promise of the AI-enabled credit union is too great a promise to ignore.

An AI-enabled credit union will be more agile and aggressive than any organization previously known. The cold, objective analysis of the machine will drive business decisions and find value that is currently hidden in data sets too large for any human to wield. The machine never sleeps. It never eats. It is relentless. So for the time being, creative and curious people will be necessary for the foreseeable AI future.

While the state of true AI maybe substantially overpromised, there are steps a credit union can take today to be ready for when the technology catches up. The first is cleaning up data – an effort that is rewarding regardless of the future of AI. The second is to document the models already in use by the credit union; even without AI, these models have value. A credit union also needs to manage its expectations – as the joke goes in AI, machine learning is written in programming language, while AI is written in PowerPoint. All this along with a creative and curious workforce will position any credit union to reap the benefits of true AI when it finally arrives.