Unilever, working with Accenture, is scaling AI-driven digital twins across its global factory network. It plans to add more than 40 new twins over the next 18 months, on top of at least five already in operation. For plant operations and S&OP leaders, the more useful takeaway is the sequence Unilever followed to get there.
How the twin delivers value: A digital twin is a live virtual model of a production line that runs on shop-floor sensor data. The value comes in three ways:
First, it predicts bottlenecks or quality drift before they happen, allowing operators to intervene before scrap is produced. That is where waste and defect reductions come from.
Second, it lets engineers test changes in the model before touching the real line. That is how capacity increases without trial runs that burn product.
Third, it continuously adjusts setpoints such as fan speed, temperature, and ingredient dosing. That is where energy and material savings appear.
Unilever’s new “agentic” layer goes further. Accenture says that as the system learns and staff confidence grows, the twin can make some of those adjustments automatically, with human oversight.
The value per site: Unilever shared results from specific plants:
Raeford, North Carolina (deodorant): predicts 95% of process-flow restrictions, cut waste 20%, lifted capacity 10%
Poznan, Poland (mayonnaise): up to 20% fewer minor stoppages, and nearly 30% less waste
Gandhidham, India (Dove soap): 30% drop in quality defects over four years.
Cu Chi, Vietnam (OMO): 1-2% savings on premium ingredients.
Haldia, India (detergent powder): an "energy twin" cut thermal energy use over two years, though no percentage was disclosed.
The prerequisite: Unilever could only add the simulation layer because it had already standardized its data foundation.
The company says its Unilever Manufacturing System already runs across 124 factories, 2,100 lines, and more than 75% of production capacity with reported gains of 3% OEE, 5% labor productivity, and 8% lower cost. The digital twins sit on top of that standardized data layer.
The pattern: The CPG smart-factory race is on, but each company is using a different technology stack.
PepsiCo is building physics-based plant simulations with Siemens and NVIDIA and reports gains of up to 20%. P&G is focusing on platform integration and warehouse robotics.
The playbook is similar: build a template, then scale.
What to watch: Unilever frames the 40-plus new twins as "a scalable blueprint for global rollout," so the test is whether the per-site results hold when the same model is copied across different products and regions.
For anyone weighing the same move, two things come first:
Get your plants onto one clean, consistent data feed before building a twin. The twin learns from that data, and Unilever needed its 124-factory system in place first.
Prove the savings beat the cost on a single line before scaling it across the network.
