Unilever's deodorant line in Raeford, North Carolina, now produces about 10% more from the same equipment, and its waste is 20% lower. The tool behind it is a digital twin, a live computer copy of the line that uses AI to predict where product will start to back up.
A line is only as fast as its slowest step
A deodorant stick is filled, capped, labelled and packed. Whichever of those steps is slowest decides how many sticks leave the building each hour, and every step before it has to wait. The trouble is that the slowest step moves. A different product, a worn part or a small jam in one place turns into a queue somewhere else ten minutes later, and the person standing at the line sees only the queue.
A digital twin works like a traffic app that shows the jam ahead on your route. It takes live readings from the machines, runs the line forward in the copy, and flags where product will pile up before it does. At Raeford, the copy predicts 95% of these slowdowns, so the fix happens while the line is still running smoothly.
What 10% is worth
Take a line that runs 16 hours a day, five days a week. That is 80 hours of production, and 10% more is eight extra hours of output every week. It is one full shift, with no new machine, no new hire and no Saturday overtime. Waste that is 20% lower is a second saving, because every discarded stick has already been paid for in materials and labour.
The gain is not a one-product accident. On a mayonnaise line in Poland, a twin cut small stoppages by up to 20% and waste by nearly 30%.
Why more factories will get this
The 2019 twins were custom-built by Unilever's own engineers with a Microsoft partner. Now Unilever is working with Accenture to build more than 40 further twins over 18 months, and the system is meant to take on some adjustments itself, with people overseeing it. When a tool moves from custom work to a service that a consultancy repeats, it reaches smaller companies a few years later. A mid-sized factory with one bad bottleneck will be able to buy this, and a 10% gain will stop being something only a company of Unilever's size can reach.
The same question works in any process with a queue
You do not need a twin to start. A warehouse shift manager could take a month of dock appointments, truck arrival times and unloading times and ask an AI assistant where trucks waited longest, and on which days. The answer might be that the waiting happens at the two doors that handle mixed pallets, whenever two such trucks arrive within the same hour. Moving one appointment costs nothing, and it frees up the same capacity that a new dock door would have bought.
That is the principle behind Unilever's approach: it starts from the problem it needs to solve and then looks for the AI that fits. A twin applies that principle with sensors on every machine. Before you ask which AI tool to buy, find the place in your process where work waits. That is where the capacity is hiding.