By: Namratha Rajagopalan, Sasha Amini and Pete McNamara
Published: April 24, 2024 | Updated: May 17, 2024
Read time: 4 minutes
The racing line: the fastest way around a racetrack. Brake just right, turn in just right, hit the apex just right, and line up for the next corner just right.
One misjudgment, and your track time is suboptimal.
So too with interest rates offered by banks on deposits. Profit is optimal and deposits are maximized by offering just the right interest rate for just the right product to just the right customer all of the time.
Slightly misjudge the line, and customers will either stray or money will be wasted.
Find the line, and you are winner. But not for long. Changes to market conditions and customer behavior conspire like variable track conditions and other cars to knock you off line.
The situation becomes even more challenging when today’s interest rate dynamics have little by way of precedent.
Experimentation instead of prediction
Changes in interest rates are by no means unprecedented. They go up or down as central banks adjust rates for inflation, and banks pass the adjustments onto customers when setting interest rates on deposits.
The degree of rate adjustment by banks is based on expected customer tolerance. Predictive AI, which can help banks predict a customer’s likelihood to default on a loan, might seem like a useful tool for establishing those degrees.
Drastic rate changes make forecasting largely impossible
But predictive AI relies on historical precedents and thrives on continuations of the status quo. Drastic rate changes make forecasting largely impossible. For example, the average US federal funds rate was 0.58% between 2011 and 2021; its average for 2023 was 5.03%.
Moments such as the dot-com bubble at the turn of the millennium or the Great Recession of 2008–2009 show similar dynamic rate shifts. But even just ignoring the different market conditions, they are isolated snapshots that provide little of worth to predictive AI.
Predictions are focused on the future, but their inputs come from the past. When those inputs are not available, it falls on business experimentation to create new contemporary data instead of relying on past data.
Test & Learn — more than just business experimentation via A/B testing
It sounds easy when posed as simple questions: What is the lowest interest rate I can offer to a customer without losing them? Will deposits increase if I increase the rates on offer?
The answers are complex. Deciding how much to adjust rates is not as simple as say decoding to tweak a website visual based on whether test group A respond favorably to a change versus control group B.
First, there are many variables, which mean constant rate tweaks up and down to find the racing line. Second, there are many customers, which mean constant accommodation of changing conditions and behavior to stay on the racing line.
Test & Learn (T&L) addresses the complexity around interest rates through small-scale experimentation on test customer groups for matching with control groups. The approach enables banks to easily learn which customers respond to which interest rates and interest-bearing products.
The feat relies on customer segmentation based on general attributes, such as customer behavior, and specific bank-related attributes, such as depth of relationship. By carefully separating signal from noise, a bank can identify the exact racing line for each rate adjustment and roll them out case by case to each and every customer.
T&L can also conduct “natural experiments” if there have been different rates or offers provided to some customers. These natural experiments can offer initial insights into how far a bank has already exhausted particular approaches. The learnings can then guide decision making for subsequent test design.
The perfect rate
Account switching is quicker and easier today than ever. Open banking allows for automated application form filling, and neobanks offer alternatives to traditional banks.
And so, finding the racing line is also more important than ever. Yesterday’s tolerance levels are too lax for today’s fickle customers.
The perfect rate is as mythical as the perfect lap. Just when you think you have nailed it, the racing line has moved again. T&L allows for continuous improvement so that the learnings from one experiment feed into the tests that follow.
In the end, all that matters is that the pursuit is optimal every time. That way, the rates you offer to each customer will be too.
Contact us for more on how Test & Learn® helps drive smarter decisions across all banking initiatives, ranging from interest rate adjustments and operational efficiency to loyalty programs and cross selling.
Related resources
