Harnessing AI as a Paternalistically Dictatorial Force for Good

Last week, I delivered a presentation at Summer Conference that focused on what banking might look like in 2025. Realizing that I had limited time, I had to pick a particular area to focus on, I chose to focus on how AI could positively affect banking services.  I hope you had a chance to attend my session in person, but if you missed it (and even if you were there to hear it, read on …) here are some additional thoughts for your consideration.

Firstly, I acknowledged that AI services such as the much-heralded ChatGPT have some limitations.  The current version commercially available, GPT-4 can still get things wrong. Since it is incredibly good at writing in human readable language, it can make wrong things “sound” right.  Potentially dangerous, right?  Further, the level of the dialogue generated by GPT-4 is not completely focused on the dynamic range of how we as humans communicate. There is a lot of nuances in how we phrase things, even our voice inflections and tone can impart critical information to the listener. AI is not fully capable of interpreting that nuance.  And of course, the possibility of fake pictures, audio and video files could wreak havoc on FIs ability to perform adequate KYC and related risk management.

But remember, my talk was about where we likely will be in 2025, not today.  Assuming that AI has not taken over everything (ala The Terminator), it is a good bet that the technology will have advanced to where we can properly harness the good elements while eliminating (or greatly minimizing) the bad. If OpenAI has version 4 of ChatGPT available today, what version will be available in June of 2025?  GPT version 21?  How good will the 2025 version be compared to today.  Consider what OpenAI has done between ChatGPT version 3 and GPT-4:

  • GPT-4 is 82% less likely to respond to a request for disallowed content
  • GPT-4 is 40% more likely to produce factual responses

But here is the more impressive statistic: it took the Internet seven years to acquire 100 million users.  ChatGPT achieved that level of users in two months.  Meaning that this technology is not going to be some fad, it is an important tool that needs to be taken seriously.  With full recognition of the potential risks and with the assumption that the future will have appropriate guardrails to mitigate those risks, here are the key areas where I think FIs should be focused on now to harness AI technology to provide a superior experience to banking customers.  They are: Enhanced Online Chat, Fraud / Anomaly Detection, Personalized Next Best Product Recommendation and Enhanced Risk Management for Lending and Investments.  Let’s examine these 4 uses of AI in more detail:

Enhanced Online Chat – I will admit that I am a BIG fan of chat. In most cases where I need support, doing so via chat is the more convenient for me. Because while chatting, I can multi-task, catching up on other activities while waiting for a response. The problem with the current options of chat is two-fold. The first fail is most chat options are really just bots and not really intelligent ones; they can handle standardized inquiries but pose a question not in the list of bot responses and you are out of luck.  Which leads to the second chat fail which is actually getting to a human via chat who does not have the experience or expertise to answer your question. It happens more than you would think.  In 2025, the power of AI could bring millions of data points to understand the specific question that is being asked and in seconds can provide a contextual answer appropriate for that individual inquiry. It would be written in a way that read like a person wrote it, not a robot.  Getting chat right might be the single greatest achievement for a financial institution that wants to have the highest quality customer service available.

Fraud / Anomaly Detection – Bankers are nothing if not concerned about the security of customer data. To that end, we have deployed many systems that portend to assist in the ability to mitigate fraud.  Sadly, most of these are not in line in real-time with how customers are transacting.  In utilizing the advanced AI tools available in 2025, billions of data points would not only would detect potential anomalies in how an individual customer transacts, but more globally detect the anomalies that identify potential bad actors on the other end of transactions.  This will enable bankers to provide customers the one thing that they most desire – financial wisdom.  The ability for the bank to provide timely information that predicts potential bad outcomes and alerts the user in real-time.  After all, what is better service: helping a customer resolve a fraudulent transaction or providing the wisdom to enable them to prevent it from occurring?  I don’t need to access ChatGPT to answer that question.

Personalized Next Best Product Recommendation – It’s no secret that I believe that we could be doing much more as financial institutions to market our services. One of the best ways to get more wallet share is to determine what product or service would best enhance an account holders experience and make that recommendation at a time and place that would capture the user’s attention.  This involves two elements: the nature of the offer and the placement and timing or delivering the offer to the customer.  Today, some online banking systems allow for a pop-up ad to appear while a user is logged in.  But often, the ad is not contextual. It is quite common for a user to see, for example, an ad for a Gold MasterCard, yet that customer already has the institution’s Gold MC.  Further exacerbating the situation, while a user is actively using OLB is not the time to ask them to break away from time critical tasks to click on an ad.  In 2025, AI will greatly enhance both of these elements, and while working together, they will provide a much higher level of clicks and conversions.  Using data points including all transaction activity of the user, the full knowledge of thousands of other users that have a similar account profile and even online search, surf and social information can lead future AI solutions to make highly accurate contextual recommendations that are perfectly suited for an individual customer. Instead of interrupting online banking activity, the offer will be presented at the request of the user or offered on an interstitial page that appears when the user has logged off, when they are more open to clicking on an offer.

Risk management for Lending – One of the toughest jobs we have as bankers is predicting the future, but we do it more than you might think.  Each time we book a loan, we are predicting a positive future outcome, which is that the borrower will pay us back.  While the absolute future outcome cannot be accurately predicted, we use tools to assist us in predicting the future. And yet, the tools we have at our disposal today seem absolutely prehistoric compared to what the future holds. In 2025, advanced AI will compile vast amounts of historical trend analysis that can alter our thinking about what the future might hold for a particular borrower. Moreover, the ability to track thousands of individual data points can alert us to a potential problem before it occurs.  Tools already exist to score overall business health, but adding the power of advanced AI will take these tools to new heights of predictive capabilities.

Are the four areas I have summarized above the only positive future uses of AI?  Hardly.  But it’s a pretty good place to start as each of these has real-world impacts that would assist the bank immediately, not some long off, down the road payback.  The real question is how an FI embarks on the path to harness AI to enable the elements I am describing. For most community financial institutions, this mean talking to your vendors, particularly those that create your customer-facing tools and finding out what they are doing to harness AI to bring out better customer outcomes in service, risk management and product / service advocacy.  Those companies that are busy right now working with the likes of OpenAI will likely have services ready for you by Summer Conference of 2025.  Speaking of the 2025 Summer Conference, let’s make a pact to meet up there and compare notes to see how many of the elements I have outlined are available and/or in use to community banks.

BTW, unlike the previous week’s article, none of this article was written by ChatGPT.  Perhaps if you compare the two articles, it will highlight the difference between how a subject matter expert and ChatGPT tackle the same topic…today.  However, it’s a good bet that by June 2025, ChatGPT may write this article even better than I could.  We’ll see if that comes to pass and if so, lament over that fact at the 2025 Summer Conference…