How to Help Clients Embrace the Use of AI in Financial Decisions

Algorithms, particularly in the form of artificial intelligence and machine learning, are proliferating in finance and changing how we make decisions.

Algorithms have been around for a while in the form of risk models, forecasting, and robo-advisors, but now generative AI is making waves in how people make choices about money. Even the Morningstar Medalist Rating received a machine learning makeover to help extend coverage and build on human expertise.

Although the power of algorithms is hard to deny, many people may not be completely comfortable being dependent on an algorithm’s output when making an important choice.

This is not a new phenomenon. In fact, for the last decade, behavioral science researchers have been studying algorithm aversion—a phenomenon when people prefer advice from humans over advice generated from algorithms, even though they are aware that algorithmic-driven recommendations often outperform humans on average.

However, given the power of algorithms in financial decision-making, it’s important to bring more people onboard.

A few key tips on how to help people come around to the use of algorithms in decision-making:

1. Try using a simple model first.
2. Start by explaining how the user benefits.
3. Incorporate human interaction.
4. Use algorithms only on things they are demonstrably good at.

Simple May Be Just as Good (or Even Better)

At a high level, AI allows professionals to throw data into a system, let a model find patterns in the data, and wait for the system to spit out an answer.

The problem is, given how complicated these models are, it can be hard to understand them and even harder to explain how they operate. Naturally, many people aren’t comfortable with algorithms that even experts don’t understand, especially when the algorithms in question are used in high-stakes contexts.

Luckily, research finds that more-complex algorithms aren’t always better. In fact, in some situations, simple algorithms based on a few straightforward rules may outperform complex models and also be more robust.

When looking to incorporate algorithms in business processes, don’t assume more complicated means better. Instead of opting for a complex black box model right away, try testing that model against a simple rules-of-thumb approach. A simpler model may perform similarly in terms of accuracy—and have the added benefit of being better understood by humans.

Start With the (User’s) Benefits

Humans, even experts, aren’t perfect and face a slew of biases when making predictions and giving advice. For example, research shows that many financial professionals are not well-calibrated, and they hold many of their predictions with unjustifiable confidence. The financial industry is not alone in this: Decades of research have found similar results with experts from other fields, like psychologists, medical doctors, lawyers, geopolitical experts, and even professional golfers.

Fortunately, many professionals have found a way to improve accuracy by incorporating algorithms into their decision-making. By helping experts make better decisions when advising individuals, algorithms then improve outcomes for end consumers. Although the improved accuracy means the end consumers are better off when algorithms are used, this benefit is often not well-communicated to individuals.

When presenting algorithms to users, focus on how the user is benefiting from the algorithm. For example, clearly state the performance quality of the algorithm and how it has helped other individuals make better decisions. Another option is to compare the algorithm’s accuracy to that of a human, which gives the user a frame of reference to compare it with. In our own research, we found that people are receptive to their financial advisor using generative AI if the advisor explains how the use of AI makes it possible for them to spend time on other things that benefit their clients.

Where’s the Human?

At the end of the day, no matter how accurate an algorithm is, it is still an abstract procedure that many people have trouble wrapping their brains around. When algorithms are used in decision-making, some people may feel like they are giving up control to a complex entity they don’t understand. When phrased like this, it is no wonder people can be algorithm-averse.

One key step that professionals can take to assuage this concern is to put a human back into the process.

For example, our research finds that investors may be more comfortable with their financial advisor using generative AI if their advisor checks the AI’s output before incorporating it into their decision—in other words, using a “human accuracy check” layer before accepting an algorithm’s output.

Other research finds that users are more comfortable if they can exert some control over the algorithm—for example, if the user can help a recommendation system learn what they like and dislike, or if the user is able to use an algorithm’s output as a suggestion that they can then choose to act on, versus the algorithm determining the outcome.

Task Selection Is Key

The power of algorithms, especially in regard to generative AI, is seemingly endless. But before implementing an algorithm, we should all ask: Is this an activity we should be outsourcing to an algorithm?

Research suggests that people believe some tasks simply shouldn’t be handled by an algorithm. Generally, people are receptive to an algorithm handling objective tasks but believe subjective domains should be reserved to humans. In finance, this means that investors are onboard with outsourcing functional tasks, like portfolio construction and managing taxes, to algorithms. However, they would rather reserve activities that require a personal connection and human interaction to actual humans.

In other words, let machines handle the numbers, but don’t outsource human connections, relationship building, and empathetic communication.

Wrapping Up

Given the efficiency and effectiveness of algorithms, their prevalence in the finance industry is inevitable. However, especially as algorithm-fueled solutions are introduced to individuals, we must be cognizant of their impact on trust and receptiveness.

Although algorithm aversion is a well-documented phenomenon, that doesn’t mean these solutions are doomed to fail. Research has already pointed to some interventions that can help people warm up to algorithms: opting for simpler algorithms when possible, being explicit on how the algorithm benefits the user, incorporating some form of human oversight, and carefully selecting the tasks that are outsourced to algorithms.