[twitter-follow screen_name=’@dinojain’ show_screen_name=’no’]
If you don’t manufacture aircraft engines, wind turbines, or any other industrial product, the concept of digital twins—creating a virtual model of a physical asset to track and optimize its performance—may not be on your radar. Maybe it should be.
General Electric (GE) pioneered digital twinning by building complex and remarkably accurate models of its aircraft engines. During the manufacturing process, GE installed hundreds of sensors in each of its engines to capture data on metrics such as temperature, pressure, flow rate, and many other variables that affect an engine’s performance.
Data from all aircraft engines was linked and the timing and chances of an engine malfunction—down to the individual part level—could then be predicted. Digital twins also helped develop more accurate and efficient maintenance schedules to improve performance of future versions of the engines.
Now, many other manufacturers of everything from rail cars to refrigerators hooked into the Internet of Things (IoT) also use digital twin technology to improve performance and build better products.
What’s more, over the past several years, digital twinning has moved out of manufacturing into other sectors. A growing number of non-industrial companies are harnessing the power of digital twin technology to better model their customers and improve performance and revenue growth.
For years, companies have used analytical models to perform predictive and prescriptive analytics across customer segments. However, digital twin technology takes modeling to a new level. It combines analytics software with machine learning algorithms to create individual customer models. For example, major insurance companies are using digital twin technology to create discrete models of their policyholders. Yes, it’s a one-off model of each policyholder. That’s a huge difference.
By using their own customer data, acquiring public data, and purchasing data from brokers, companies can now aggregate millions of data points about their customers and use that data to construct a digital twin of each customer. Those models enable them to predict individual customer behavior, as well as future performance and revenues, with uncanny accuracy.
Even better, the models get more intelligent over time. Via machine learning algorithms, as new pieces of data come in, each digital twin synthesizes the data contextually and “learns” more about its needs, spending patterns, life events, preferences, and other habits that further enhance the capabilities of the predictive model.
Again, this is not at the segment level; this is at the individual level. It’s almost as intimate as human knowledge. How well do you know your wife or husband or kids? Probably well enough to use past behavior to predict future behavior with near certainty.
Digital twins deliver a close doppelganger. The models you develop will have a tremendous payoff in terms of efficiencies gained in crunching those millions of pieces of data, making accurate predictions, and testing products and services against customer desire and behavior before you bring them to market.
And, it’s not just insurance companies that can build digital twins of their customers. Any company—from financial services, to retailers, to airlines, to healthcare systems, etc.—can use digital twinning to model customer behavior. It’s a matter of collecting enough information and investing in the requisite analytics and machine learning capabilities.
However, as with any machine-enabled venture, there are some caveats:
- Humans are not digital models. There will be times when the models’ predictions are wrong and the humans they model won’t make rational decisions. Learn from mistakes, and remember that good old-fashioned customer knowledge and business experience still needs to be heeded.
- Garbage in equals garbage out. The models are only as good as the data they crunch. The more data you have, the less a few pieces of inaccurate data will affect your models’ analyses and decisions, but—as with any IT system—it’s good to have the best data you can get.
- Don’t go cheap. Digital twins require lots of power and storage. This technology is large scale, writ large. The more (and better) processing power you have, and the more cloud storage you have, the more efficiently models will run and the more data they can crunch, which will lead to better decisions.
Digital twin technology isn’t easy to implement, nor is it cheap. But if it’s done well, it’s worth every ounce of brain (and machine) power, and every penny you put into it. The technology affords you the capability to accurately predict individual customer behavior and learn based on experience. You can build sandbox where you can solve problems, test theories, and simulate the results of decisions BEFORE you make mistakes at the customer level. That’s a win-win for everyone.
I’d love to hear what you think. Please comment or DM me on Twitter, and please follow me! [twitter-follow screen_name=’@dinojain’ show_screen_name=’no’]