February 13, 2017
by Anand Rao
How AI-based virtual replicas can help financial institutions better predict customer behavior.
Ever since wind farms became part of the US energy market in the 1980s, operators have worked hard to compete with more established energy providers. And they’ve found that when you’re trying to improve operational efficiency, manually inspecting broken parts on a wind turbine just won’t cut it. Instead, wind farm owners use artificial intelligence (AI) to find ways to produce more energy with less wear and tear on their equipment. Meet digital twins: models that are virtual replicas, or “twins,” of each physical asset (the wind turbine), the underlying parts (the rotors), and the entire system (the fleet of wind farms). The models simulate real-world conditions, respond to changes, predict problems before they occur, and help managers make better decisions.
Generating customer insights
Financial services firms can use the digital twin concept too. Most financial institutions create models based on customer segmentation—groups of people who share a set of common traits and behaviors. By applying the digital twins concept, a firm can actually create individual profiles for each customer. In essence, it’s a model of one. The twin can simulate the decisions that a real-life person might make, achieving a level of predictive analytics that is unique and more accurate than previous models.
How digital twins work
At a recent Mind + Machines event, Colin J. Parris, vice president of software research at the GE Global Research Center, explained that digital twin models operate in three stages:
- See. The digital twin gathers data. The model constantly updates itself with new information about individual assets, the system, and external data sources. It analyzes data using machine learning and other AI techniques.
- Think. Users can ask the digital twin questions. Based on this input, the digital twin runs simulations looking at historical data, current data, and forecasts. The model then considers what might happen, as well as the relative risk and confidence level associated with each option.
- Do. The digital twin proposes a course of action for a person to review. The person, or even the twin, can then take action.
Creating a model of one
This goal of modeling individual behavior, of course, isn’t new in financial services. For example, to predict a person’s spending needs after retirement, a financial advisor has to consider each client’s specific financial, social, and health concerns, and then project them into the future. What financial service providers and advisors need most is “a model of one.”
We’ve used digital twin technology to model individual policyholders and simulate their future balance sheet and cash-flow statements. (See our paper on behavioral simulations for the Society of Actuaries.) In this case, we’ve combined data from multiple public and proprietary sources to create a synthetic data set at the individual consumer level. This digital twin synthetic data has more than 330 million rows (one for each US resident) and more than 4,000 sociodemographic, behavioral, financial, and health factors spread over time. Wealth managers, asset managers, and insurance companies can use this data to construct a more complete view of their own customers’ financial profiles and needs. The customer insight platform allows our clients to:
- See. The insurer or wealth manager creates a twin using the synthetic data set as well as the specific transactional data that the company has. This digital twin is constantly learning about its real-world counterpart, and it’s updated as any new information comes in with respect to the consumer’s life events, spending and savings decisions, personal preferences, and changing financials needs.
- Think. The digital twin considers all different spending and saving decisions at each critical life event. It projects the impact of its decisions into the future and then makes a choice based on its preferences and needs.
- Do. The digital twin proposes an optimal savings strategy that the consumer can accept or reject. Once the consumer has determined a plan and made a decision, the twin can carry it out.
Proceeding with caution
If you’re a financial institution that wants to use digital twins to help you think through decisions, here are some things to consider:
- Models are only as good as the data added, and this is especially true with digital twins. You’ll also need to follow proper model validation and model management processes to make sure that your data stays current and relevant.
- Once you’ve implemented digital twins, you need to be able to trust the output. Apply a strong AI governance program to make sure the people, process, structure, and technology meets the standards of regulators, stakeholders, and customers. In particular, you want to make sure that the AI simulations aren’t biased, their decisions can be explained by the financial advisor, and the results are verifiable.
- Obviously, sensitive client data used for digital twins should be kept secure, and you’ll need to follow relevant privacy rules. Make sure you understand all the restrictions that might apply.
- It takes a lot of processing power to run the behavioral digital twin simulations. You’ll need distributed and parallel processing units that can handle the computational load.
- You can model individual consumer preferences and behaviors, but do not underestimate the difficulty of fully capturing the human factor. Don’t be caught off guard by the occasional surprise.
Applying the digital twin model
Digital twin technology can help firms get to know their customers’ wants, needs, and expectations on a more personal level. Yes, humans are infinitely more complicated than a wind turbine. And much of what drives human decision-making is unseen and difficult to model. But the digital twins concept gives financial institutions a real opportunity to gain deeper insights when they model customer behavior. Digital twins might just be what your firm needs to power growth for years to come.