A digital twin is a live virtual copy of a company's factory, supply chain or computer systems, fed by the same data as the real thing, so people can break it on screen without breaking anything real. Many large companies already have one for planning and monitoring, and very few use it to rehearse a crisis. That is a bigger loss than it sounds.
A rehearsal finds the crisis nobody imagined
Most crisis exercises are meetings. A team reads a made-up scenario and talks through what it would do, so the exercise covers the crisis that someone could imagine. In July 2024 a faulty security update crashed 8.5 million Windows devices, and airlines, hospitals and banks lost their systems. Six months earlier, Change Healthcare, which handles insurance claims for much of the US health system, was shut down after criminals got in through a login with no second step of verification.
In both cases the cause was small and came from outside: a supplier's update, one unprotected login. The damage came from everything that depended on it. A rehearsal on a copy of the real systems asks what stops working next when one thing goes dark, and the answer comes from the connections the systems really have, not from what the people in the room remember.
Netflix's tool shows what this does over time. It runs every working day, so the practice never fades between yearly exercises. The trouble is that most organizations cannot switch off the real thing at noon. A hospital group cannot turn off patient records to see what happens. A digital twin lets it do that on the copy, which brings the Netflix habit to work that cannot be switched off.
You do not need a digital twin to start
A twelve-shop bakery chain has no digital twin. It has a card payment provider, a delivery app, one shared email login and a cold room. A manager can write those four on a page and pick one: the card provider goes down at ten on a Saturday morning and stays down until four.
She then gives an AI a plain description of how each shop takes orders, handles cash and schedules staff. She asks it for three different ways the day could go, what breaks second in each, and which assumptions in her plan are weakest. It might point out that two shops keep no cash float at all, or that the delivery drivers cannot be paid until the card system returns. She decides what to change, and writes it down within a few days while the problem is fresh.
That is about an afternoon, roughly what one meeting costs, and it can be repeated every quarter with a different thing switched off.
AI helps only when it is used with structure
In 24 simulated crises, AI working alone spotted 41% of the critical developments planted in the scenarios. Human teams alone found 48%, and human teams with AI found 49%. Teams that could search for outside information found 81%.
So the large gain came from outside information, and a casual chatbot question added almost none of it. A crisis team's weakness is that everyone in the room imagines the same crisis. A search engine and an AI are cheap ways to bring in someone else's imagination, but only if the team asks properly. The best teams that used AI wrote many messages with full context and asked it what a team shaken by the event would overlook. The weakest team that used AI sent one short question with no background.
The practical step is small. Pick the one thing your work cannot run without for a day, write down how the work really depends on it, and ask an AI to argue with your plan for losing it. The first time, you will find a dependency you did not know you had.