Will Your Most-Trusted Financial Advisor Be a Robot?

Artificial intelligence (AI) is literally everywhere. Causing seismic changes to industries, the technology is displacing jobs in some sectors while creating them in others. Yet overall, AI actually stands to make our lives much easier.

It will do this in part by making machines understand—and imitate—us better. For example, AI technology can help us handle mundane tasks such as setting up appointments. Currently, 60 percent of small businesses don’t have online booking systems, which means appointments and reservations must be made over the phone. Rather than place these calls ourselves, Google has created an AI technology called Duplex that can hold voice conversations with an employee at a hair salon or restaurant. Able to understand the complicated nuances of natural language, Duplex not only handles the calls, but it sends a confirmation once a reservation has been made. As Duplex is refined, it’s going to be harder and harder for the person on the other end of the line to discern whether they’re speaking with a human or a machine.

As we use AI for more complicated and creative processes, we need to make sure the enormous amounts of data it leverages can offer personalization at scale, engendering trust between consumers and companies. The best, and perhaps only way to do this is by teaching machines to anticipate future human behavior across every aspect of transaction.

Studying our own behavior to build more thoughtful machines

AI is designed to copy intelligent human behavior. Therefore, it makes sense that we first analyze the actions of people before building machines that can handle a wide range of unexpected human outcomes. Behavioral analytics is an area of data analytics that allows machines to collect demographic, geographical and other important information from users. Profiles are not only created for every single user account, but they are tracked and updated every time a new transaction occurs.

Already used in gaming, e-commerce and social media, the data compiled using behavioral analytics allows companies to examine more than single events—instead, they can look at seemingly unrelated past activity to develop reasonable predictions about what might happen in the future.

In an effort to prevent fraudulent sales, Apple released Touch ID, a payment security system that validates users’ identities through their fingerprints for its Apple Pay application. Yet fingerprints can be stolen, making TouchID an imperfect system. In fact, it was so imperfect, that it was killed after a five year run. A profile of a person’s behavior, however, is more difficult to duplicate, as it’s been compiled over a period of time. In other words, while a machine could never know if someone were using a stolen fingerprint, it could potentially detect if a user were acting outside of the normal behavior of a profile.  

Over time, profiles become more robust, offering granular insights about the actions of either an individual or even an entire demographic. Such profiles can mitigate costly mistakes and predict future behavior that can serve both consumers and companies.

Decreasing errors in data reporting by using hybrids

Of course, it’s impossible for all AI and machine learning predictions to come true, as they can only be “approximations of [what will happen in] the real world.” Yet companies are finding that developing AI hybrids, which combine machine learning-driven platforms with human analysis, yield better results than traditional AI systems.

Humans don’t have the capacity to find all errors in our automated world, but machine learning-driven solutions have their weak points too. Since they are trained on specific data sets, they may trigger false positives and other errors when unleashed on the much messier data in the real world - errors that may require human investigation. In 2016, however, researchers from MIT revealed that a machine learning anomaly detection program that incorporated input from humans was far better at predicting cyber attacks than when left to its own devices. In fact, this hybrid system, called AI2, markedly reduced false positives and predicted 85 percent of cyber attacks, which was about three times better than previous benchmarks.

Real world applications for such hybrid solutions are still being developed, yet innovative companies have already been trying them on for size. In 2017, for example, Amazon created Amazon Macie, a machine learning-powered visibility service that uses predictive behavioral analytics to understand and access data to automate security. The system, which was inspired by observing real human behavior, monitors user log data to detect anomalies, and is proven to yield very few false positives.

Hybridization not only helps mitigate risk, it improves trust

Amazon is by no means the only company using such models. In fact, wealth management services now routinely partner human financial advisors with robo-advisors to improve trust while keeping clients secure.

The AI that drives robo-advisors uses algorithms to create profiles for their investors while allocating assets based on modern portfolio theory. Unlike humans, they work 24 hours a day to handle tedious tasks such as rebalancing investment portfolios, which can help improve returns and reduce volatility. Many of these automated systems are equipped with sensitive fraud-alert features—some companies even reimburse clients if there are unauthorized transactions.

Although no humans are necessary for automated, AI-driven portfolio management, employees haven’t been made redundant. In fact, they’re more necessary than ever, as they use the data compiled from robo-advisors to provide better, more efficient customer service. Allowing AI to partner with traditional family advisors is big business, as evidenced by robo-advisor behemoths like Betterment LLC (with over $10 billion in assets under management), Wealthfront and Personal Capital, with $10 billion, $7.5 billion and $5 billion in assets under management, respectively.

It’s not a stretch to imagine that hybridization may soon extend beyond payments, investing and financial services. Ayasdi, for example, a company that uses AI to analyze complex data sets to discover seemingly disparate relationships, not only creates interesting risk models in the financial services sector, but also in healthcare and the public sector.

It’s all about relationships

Currently, AI offers tremendous personalization at scale, allowing us to handle tasks such as placing reservations to managing assets. Yet when it’s not built to predict future inconsistencies, this technology can deliver costly false positives. Using behavioral analytics to foster relationships between machines and humans can help further personalize services, mitigate risk and decrease false positives.