Digital twins are making companies more efficient

They will also help them reap the benefits of advances in AI

Aug 28th 2024|San Francisco and Seattle

WHEN A PASSENGER in search of a taxi orders an Uber, all it takes is a few taps on a smartphone to make a car appear, as if by magic. Traffic permitting, they are soon whisked to their final destination. But the magic tricks do not end there. As soon as that screen is pressed, the passenger—along with all of Uber’s other riders, drivers and the systems that connect them—becomes part of a comprehensive digital replica of the firm’s inner workings.

This digital twin, one of the most sophisticated of its kind, allows Uber to adjust its operations in real time. Annoyed passengers may think that this enables the firm’s “surge pricing”, when fares suddenly spike to balance ride demand and driver supply. This is partly true. But the more immediate and more positive effect is that the digital twin allows for up-to-the-minute route optimisations through ever-changing city traffic.

If current technology trends hold, such end-to-end digital representations of a company’s inner workings—and, increasingly, its ecosystem of customers and suppliers—will no longer be the speciality of tech firms such as Uber. Artificial intelligence (AI), in particular, will make it much easier for all sorts of businesses to build virtual replicas and oversee them on a scale managers alone never could.

As a result, digital twins will redefine what it means to run a company. Instead of co-ordinating disparate islands of automation, as is the case today, bosses will manage a constantly churning “flywheel” fuelled by data. With access to information from all over the company’s operations, as well as from its customers and suppliers, a corporate twin will not just help managers make better plans. It will also implement them, learn from the outcomes and optimise itself to achieve certain corporate objectives—over and over again.

Companies have long tried to model and automate key parts of their business. Even before the global financial crisis hit in 2007, Goldman Sachs, a bank, built a system called SecDB which, among other things, regularly calculated the different types of risk facing its different financial assets. When Lehman Brothers, another bank, went bankrupt in 2008, this system allowed Goldman quickly to understand its exposure to the failing firm.

As such systems multiply and interconnect, companies are in effect building digital twins of themselves—equivalent to recreating a human one organ at a time. What distinguishes these models from their predecessors is their ability to continuously monitor (and influence) their real-world equivalents. Amazon, a big online retailer, is considered to have pushed this process the furthest. After dominating e-commerce for nearly 20 years, the company has amassed vast amounts of sales data, enabling it to build a single model that can forecast demand for 400m items two years into the future. It can even anticipate how a new book by a famous author such as Michelle Obama will fare and how a Taylor Swift concert will impact local demand.

But this model is only part of Amazon’s supply-chain optimisation. “Once we have a reliably accurate sales forecast, we can use this as the basis for all planning,” explains Ping Xu, who leads forecasting. She oversees a “gym” in which models that optimise different parts of Amazon’s supply chain—from how many of a certain item to keep in stock to where to build new warehouses—train together to learn how to act as one coherent model.

What took Amazon years to put together is now becoming much easier to build. Cloud-based databases have helped, allowing companies to store their data in one place for large-scale analysis. So have data-harmonisation techniques, designed to ensure different bits of information are mutually compatible. Molham Aref, the founder of RelationalAI, a startup, aims to turn business processes into what he calls “Lego blocks of digital twins” that can together produce a replica of any company.

The greatest impact on the development of corporate digital twins, however, will come from AI and machine learning. For one, these tools make it easier to grasp the internal processes in need of modelling. Celonis, another startup, currently designs software that trawls a company’s internal data for useful insights. In due course AI will be able to perform this discovery process more flexibly and with minimal prior instruction. Just as large language models (LLMs), which power services like ChatGPT, can extract patterns from vast amounts of text, corporate models fed on business data could discover what makes a firm tick, predicts Dario Gil, head of research at IBM.

LLMs will also allow digital twins to adapt. Enterprise software used to involve rigid rules, which made finding workarounds tricky. If a customer wanted to return a product, for example, that would have to be handled in the software-approved manner. Large Action Models, as some call such LLMs, could change that. Trained on complaint messages and other unstructured data, they may be able to offer customer-support workers flexibility, or even perform tasks themselves. “Enterprise software will become more generated-on-demand and self-assembled,” says Charles Lamanna, who leads the development of such software at Microsoft.

On the double

Most important, however, AI and digital twins will each enable the other to flourish. Just as fragmented computer systems hamper data analysis, they also constrain what task-performing algorithms known as agents can do. Digital twins offer, in effect, a level playing-field for agents to move on. Such tools will only become more important as agents become easier to build.

What’s more, as AI becomes better at capturing what happens inside companies, an ever-bigger part of their internal processes could be turned into software. This could launch a virtuous virtual cycle, in which new enterprise software generates more data, enabling yet deeper AI insights and creating an ever-more detailed digital twin. Firms that jump on such a bandwagon early may well have a lasting advantage.

Such companies are also likely to shift shape. The past 25 years saw the rise of huge tech platforms, including Uber, Google and Meta, most of which are marketplaces that match consumers with goods, services and content. As non-tech businesses, from carmakers to insurers, become more and more embodied in software, they will turn into large platforms. By embracing their digital twins, companies will be able to do more than just match buyers and sellers, orchestrating complex relationships between them too.

If businesses can increasingly be digitally replicated, why stop there? Some firms have started to build digital twins of entire sectors of the economy. J.D. Power, a data-analytics firm, is gathering reams of data on the American automobile industry—including information about individual cars, which dealers stock them, how they are configured, and so on—and how such factors influence sales. With the help of Palantir, a software-maker, J.D. Power is now developing a system that can indicate the current state of the market, as well as show carmakers what is likely to happen if they adjust certain variables, such as increasing incentives in a specific market or supplying more vehicles with luxury packages or in particular colours.

Such opportunities also come with risks. As businesses become ever more reliant on digital twins fed on their most sensitive information, they also leave themselves more vulnerable to being hacked. A well-targeted attack could, in theory, not only grant rogue actors access to a company’s deepest secrets, but also allow such data to be secretly manipulated—with real-world consequences. This is magic to be handled carefully. ■

Curious about the world? To enjoy our mind-expanding science coverage, sign up to Simply Science, our weekly subscriber-only newsletter.