What Comes Before AI? PI.
April 3, 2024
|
6 minute read – When it comes to newer technologies in the financial services industry, the usage of artificial intelligence (AI) is front and center, but it comes with its own set of challenges including not always achieving promised performance levels and unanticipated pitfalls. But the challenges don’t mean that the opportunities aren’t worth pursuing. Institutions are seeking to learn faster as they take advantage of new AI capabilities and they’re finding that applying process improvement (PI) principles before implementation is key to gaining the full advantages of AI without creating unexpected issues.
AI is already commonly used in lending decisions, fraud prevention, and customer service, and as the field advances, new AI applications seem to appear daily. While the following questions apply to AI implementations, many of them are also relevant for less-complex solutions such as robotic process automation (RPA) that simply automates manual processes.
6 PI questions to ask before implementing AI:
1. What is the clear objective of this AI implementation?
This question isn’t limited to PI, of course. It should be answered for almost any strategic initiative, project, or endeavor the institution undertakes. It is especially important that the objective is crystal clear with AI implementations since there is still so much to be learned. As you prepare to articulate the objective, it helps to ask, what problem are we trying to solve or opportunity to take advantage of? How will this support our strategy? What will be the tangible outcomes? What benefits will be realized? How will we measure success? Starting with a clear objective and measurable success outcomes that align with the strategy is the first and most important step and is critical to hitting the mark.
2. Are we paving the cow path?
A common pitfall is what is referred to as “paving the cow path” or automating what we’re currently doing without considering whether that’s still the best path. When pieces of existing processes are automated or enhanced with AI, a question that sometimes remains unasked is, do we really need this step at all? PI is used to reveal steps that are no longer necessary and highlights business decisions for reconsideration. For example, when automating report generation, find out if anyone is still using the reports. Or when automating the payoff of multiple customer credit accounts for a debt consolidation loan, a business decision might be made to instruct the customer to pay them off, eliminating the need for automation.
It’s tempting to think that there’s really not much harm in automating an unnecessary step since it will function on its own, but the cost of automation or AI enhancement is not limited to its initial cost. Just like other technologies, it must be maintained with updates and any changes made to the data stream coming in or out may necessitate changes in the interface. This creates inefficiency if the automation or AI wasn’t strictly necessary in the first place.
3. How would we approach this enhancement or problem if AI didn’t exist?
In PI, it’s important to consider various ways to achieve the objective, so this question is extremely helpful with AI because so many new solutions are being created. The newness and novelty make it easy for the thought process to be dominated by the capabilities of the solution rather than the problem to be solved or enhancement to be offered. Take some time to think through other ways to achieve the benefits. Some have identified less complicated, less expensive solutions by asking this question.
4. What impacts might this implementation have that we aren’t thinking about?
One of the principles of PI is to seek out unintended negative effects on the customer or employee experience such as, system efficiency, time spent, or costs. For example, an AI implementation that provides personalized customer experiences could create friction in the account opening or lending processes as they are modified to gather additional data. Or perhaps the system slows down due to the additional demands. PI is extremely helpful in uncovering this type of conflict, which can then be weighed carefully against the benefits or adjusted to minimize added friction.
5. Are our sources of data and data governance policies ready for AI?
The institution’s processes for gathering, protecting, and maintaining data need to be stronger than ever. Sophisticated AI solutions are built to learn from data without being explicitly programmed and often require large volumes of complete and accurate data to learn from. This could be internal or external data and the potential exposures to privacy, transparency, and bias issues are magnified with AI. In addition, regulations are still evolving for how to protect against these and other risks.
Just scratching the surface on data governance, assuming you are using a third party for the AI solution, get clear on how the third party uses data (internal and external) and whether they have adequate guardrails in place. For usage of internal data, consider whether there is enough for the solution to learn from without creating biases or incorrect answers. Similar to business intelligence initiatives, there is often work to be done on capturing the right data accurately and consistently as well as clean-up of existing data.
6. Is our vendor management process up to the AI challenge?
The complexities of working with AI vendors may make this the right time to do PI on the vendor management process. Strong vendor management can not only track how the institution is using AI, but can also help ensure that your governance policies are applied and tracked consistently.
AI is an exciting, rapidly-developing, area that holds much promise for the financial services industry. Since it’s relatively new, there is much to learn. Using these process improvement questions and examining the processes that AI is a part of, as well as other processes it affects, could save some headaches and put you on the road to better results for the whole organization.