Financial Institutions generally have two camps related to the deployment of AI, more specifically, Generative AI (“GenAI”). In one camp are those ignoring any real benefit of GenAI and have tried their best to exclude GenAI from any daily business activities. This camp views GenAI as inherently risky, untrustworthy and sees only negative outcomes associated with its use. In the other camp are FIs who are legitimately looking at GenAI to determine how it can best be used throughout the enterprise. While I don’t see FIs in this second camp throwing care to the wind and using GenAI haphazardly, there does seem to be some disconnect between the prospect of what GenAI can reasonably do and the use cases in which it is being deployed. For those who are not regularly following my writing on AI, let me remind you of a couple of quick facts. GenAI is based on probability, not on concrete absolute facts. Thus, GenAI cannot accurately calculate a simple math problem that any $5 calculator can. The fact that GenAI is based on probability and not absolutes means that depending on the desired outcome, the output of GenAI may be non-repeatable or just flat out wrong. Thus, it was with interest that I read an article in The Financial Brand called 8 Mistakes That Will Guarantee AI Fails at Your Bank. A link for that article can be found at the end of this article, but I believe that one or more of these 8 potential mistakes could be affecting any FI’s ability to successfully deploy GenAI solutions that can be positive for the organization. Let’s examine the 8 “mistakes:” Technology Overkill — Using AI to Solve Simple Challenges – The article points out that in some cases, more emphasis is placed on the technology chosen to be used versus the problem itself. While it is possible to use GenAI in an “overkill” mode to test the viability of it to solve a problem, in most cases, there is a business need to solve a unique or specific problem and thus that is the focus of the problem-solving quest, not researching capabilities of a GenAI tool. Unless you are a university researcher, it is not likely smart to use a bazooka to hunt mosquitos. Focus GenAI tools on problems appropriate for their use. Putting Too Much Faith in AI Techniques – The hype surrounding GenAI has made it such that some business leaders figure it is the answer to virtually any business problem. While demonstrably untrue, it will be extremely frustrating for problem-solving teams to have unrealistic expectations about the ability of GenAI to solve problems. Further, failure of GenAI to produce on a problem it was unsuited for may cause the organization to forgo using it on a different problem where it could be very effective. Trying to Solve the Wrong Business Problem Using AI – This is a big issue I have highlighted in the past; the failure of clearly documenting what the desired outcome is for any GenAI project. Poorly articulated business requirements allow data scientists to make their own interpretation of the problem. While what they create with GenAI may solve “that” problem, the actual problem the organization has is left unsolved. Expecting AI Models to Learn Your Business Overnight – The amount of training and tuning of a GenAI model is greatly underestimated. Especially in the instance of a small language model AI, just presenting related documents to a GenAI tool does not equate to it giving the results to align with expected outcomes. In general, there is an expectation of instant success, like a magic drug, when the better analogy might be a workout routine that is strictly adhered to and is altered as needed to achieve a desired outcome. Treating All the Data in Your Bank as If It Is Gold – The quality of the data inputted into the GenAI model is critical for its success. Poorly organized and/or incomplete data will leave the GenAI tool with missing pieces as it attempts to determine probable outcomes to queries. There is also the issue of whether those that have access to data stores are fully cooperative with a GenAI project if they feel that the purpose of the GenAI tool is to eliminate their job. Not Employing Enough ‘Plumbers’ to Get the AI Job Done Right – There is a disconnect with those identified as “data scientists” who are directing GenAI projects and “data engineers”, those that work with datasets and who can perform needed data cleanup and data manipulation needed to make a GenAI project work. Many GenAI projects fail to have the requisite data engineers working alongside data scientists to achieve overall success. Investing Insufficiently in Tools to Use What AI Builders Develop – GenAI projects cannot be viewed as a “once and done” model. Data will need to be constantly updated and new data fed into the GenAI model. More importantly, there is a big difference between the laboratory development of a GenAI tool and the production version rolled out to multiple end users. In many cases the production version is not setup to enable the end users to access or use the tool in the manner in which it could actually bring value. This goes beyond just the typical issues that FIs face in getting tools to be supported by IT, but even that is still an issue at most FIs. Letting Data Scientists Learn about New AI on the Company Dime – There is a strong urge to find potential problems to which a GenAI tool can be applied. I am certainly one who is always looking for how GenAI could be a game changer across numerous FI departments. But again, in most cases, there are simpler options available for problem solving. It is important to resist the urge to use GenAI as some type of universal problem solver and try to shoehorn it in every which way. The fact is that GenAI can and is being used to solve problems that exist in financial services. It is not appropriate to stick our heads in the sand and ignore the obvious benefits that GenAI can bring. But it is also inappropriate to use GenAI as a panacea that somehow can solve virtually any problem. Start with a clear definition of the outcome that you need to achieve. Do you regularly have loan operations staff that is confused about how to properly document and process loans across many different loan and collateral types? This represents a great option for a small language model GenAI project. Conversely, using GenAI to make next best product recommendations where the institution will not allow offers to be presented to customers while they are active in online or mobile banking may not make any sense. Align your use of GenAI with reasonable outcomes that align with your overall strategic goals. Deploy the right resources and provide technical staff with the business requirements that clearly document the desired outcomes. Then hold the groups working on GenAI projects with documentation of their success (or not) according to the desired outcome. Don’t get mad if early attempts are unsuccessful, figure out what went wrong, make adjustments and try again. GenAI is a dynamic tool that can bring tremendous benefits, but you can’t fully trust it nor give up on it too early in the process. Resources https://thefinancialbrand.com/news/data-analytics-banking/artificial-intelligence-banking/8-ways-to-make-artificial-intelligence-fail-in-your-bank-181159/ The views expressed in this blog are for informational purposes. All information shared should be independently evaluated as to its applicability or efficacy. FNBB does not endorse, recommend or promote any specific service or company that may be named or implied in any blog post.