Unilever is expanding digital twin technology to 40 manufacturing sites worldwide, and the company is explicit that the programme is judged on measurable business results, repeatable processes and workforce training rather than on technology deployment. The framing matters because industrial digital twin programmes have a long history of delivering impressive demonstrations without durable financial returns. For retail and consumer goods supply chain teams, the lesson is that the value of a digital twin depends less on the model than on the metrics attached to it.
Key Conclusions
A digital twin is a computational representation of a physical asset that updates as the asset changes. In manufacturing, the most common applications are throughput optimisation, energy management, predictive maintenance and changeover scheduling. Each of these produces a different return profile, and the return depends heavily on whether the site has the instrumentation and data quality needed to support the model.
The gap between demonstration and return usually appears at the data layer. Models require consistent sensor coverage, reliable timestamps and stable naming conventions across equipment. Sites that have grown through acquisition often carry a mixture of legacy systems, which means the first phase of a digital twin programme is data standardisation rather than modelling. Programmes that skip this phase produce models that work in a pilot and fail at scale.
The decision to prioritise repeatable processes is therefore a maturity signal. Rather than building a bespoke model for each site, the programme defines a standard set of use cases, a standard data schema and a standard deployment sequence, so that each new site can be commissioned in less time than the previous one. Repeatability, not sophistication, is what turns a pilot into a portfolio.
Workforce training is the third pillar and the one most often underfunded. A model that recommends a changeover sequence is only useful if shift supervisors trust and act on it. Building that trust requires training, visible success on low-risk decisions, and a feedback channel that lets operators correct the model. Without these, recommendations accumulate in a dashboard and the measured benefit stays near zero.
Choosing Use Cases That Produce Measurable Returns
The use cases with the clearest financial return in consumer goods manufacturing are throughput bottleneck removal, energy consumption per unit produced, and unplanned downtime reduction. Each has a directly attributable cost line, which makes the business case easy to construct and easy to verify after deployment.
Changeover optimisation and quality prediction have higher ceilings but weaker attribution. Their returns depend on downstream effects such as inventory reduction or scrap avoidance, which are harder to isolate from other initiatives. Programmes that lead with these use cases often struggle to demonstrate value even when the models perform well.
The practical sequencing rule is to start with use cases whose savings appear on an existing cost line within one reporting period. That produces a verified benefit early, which funds and legitimises the harder use cases. Leading with the most sophisticated application, however attractive technically, tends to stall the programme.
Building Repeatability Across Sites
Repeatability requires three standardised assets: a defined data schema that every site maps to, a defined set of use cases with fixed success metrics, and a defined commissioning sequence with target durations. Together these let a central team deploy each new site with progressively less bespoke work.
The data schema is the foundation and the most expensive to get right. Equipment from different vendors reports different variables at different intervals with different units. A central mapping layer that normalises these into a common model is what allows models built at one site to be reused at another, and reuse is where the cost curve bends.
Commissioning duration is the metric that shows whether repeatability is real. If the tenth site takes as long to commission as the first, the programme has produced bespoke projects rather than a repeatable capability. Tracking commissioning time as a first-class metric keeps the programme honest about its own efficiency.
Workforce adoption should be measured with the same rigour. Track the share of recommendations accepted by operators, the time from recommendation to action, and the number of operator-initiated corrections to model outputs. A rising correction count is not a failure signal; it usually indicates that operators are engaged enough to improve the model.
The pattern generalises beyond manufacturing. Retailers deploying forecasting or replenishment models face the same sequence: data standardisation first, use cases with attributable savings second, repeatable commissioning third, and operator adoption throughout. Programmes that invert this order tend to produce dashboards rather than results.
Best Practices
Lead with use cases that have an existing cost line
Throughput bottleneck removal, energy per unit produced and unplanned downtime reduction each map to a cost line already tracked by finance. Starting there produces a verified benefit within one reporting period, which funds the harder applications later. The sequencing also builds credibility with site leadership, which matters more than model sophistication when the next investment decision is made.
Standardise the data schema before modelling
Build a central mapping layer that normalises equipment variables, intervals and units into one common model. Reuse of models across sites is where the cost curve bends, and reuse is impossible without a common schema. The mapping layer should be owned centrally and version-controlled, because uncoordinated local changes are the most common reason a reusable model becomes site-specific again.
Track commissioning duration as a core metric
If the tenth site takes as long to commission as the first, the programme has produced bespoke projects rather than a repeatable capability. Treating commissioning time as a first-class metric keeps the claim of repeatability honest. Commissioning time should be reported alongside the financial benefit, so that improvements in deployment efficiency are visible rather than assumed.
Measure operator adoption, not just model accuracy
Track the share of recommendations accepted, the time from recommendation to action, and the number of operator corrections. Rising corrections usually indicate engagement, not failure, and adoption is what converts model output into financial return. Adoption metrics should be reviewed with the same rigour as model accuracy, because an accurate model nobody uses produces no measurable business result.
Common Mistakes
Treating a digital twin as a technology project rather than an operating change is the most common failure mode. The model is the easy part; consistent instrumentation, standardised naming and operator trust are what determine whether the model produces a return. Programmes that measure deployment rather than results tend to produce demonstrations rather than savings.
A second mistake is building a bespoke model for each site. Bespoke work produces impressive pilots and a cost base that scales linearly with the number of sites, which means the programme never reaches portfolio economics. Standardising the data schema, the use case set and the commissioning sequence is what converts a pilot into a capability.
Summary
The decision to judge a digital twin programme on measurable results, repeatable processes and workforce training rather than on deployment count is the most consequential design choice in the initiative. It sets the programme's incentive structure, and it determines whether the tenth site costs less than the first.
For retail and consumer goods supply chain teams, the transferable sequence is clear: standardise data first, choose use cases with attributable savings, build repeatable commissioning, and invest in operator adoption throughout. The same sequence applies to forecasting, replenishment and store operations models, which makes it a general framework for scaling analytics rather than a manufacturing-specific recipe.
Data Sources
- Unilever puts business results first in factory digital twin rollout
- Tractor Supply opens Idaho distribution center with automation, dedicated AI team
- GlobalFoundries building $21B New York fabs under US-Japan trade deal
FAQ
What is a digital twin in manufacturing?
A: It is a computational representation of a physical asset that updates as the asset changes, used for throughput optimisation, energy management, predictive maintenance and changeover scheduling.
Why do digital twin pilots fail to scale?
A: Usually because of the data layer. Models require consistent sensor coverage, reliable timestamps and stable naming conventions. Sites with legacy systems from acquisitions need data standardisation before modelling.
Which use cases deliver the clearest returns?
A: Throughput bottleneck removal, energy consumption per unit produced and unplanned downtime reduction. Each maps to an existing cost line, making the business case easy to construct and verify.
What makes a programme repeatable?
A: A standard data schema, a fixed set of use cases with defined success metrics, and a defined commissioning sequence with target durations. Together they reduce bespoke work for each new site.
Why does workforce training matter?
A: A model recommending a changeover sequence only creates value if supervisors trust and act on it. Training, visible early wins and a correction channel convert model output into operational decisions.
How should adoption be measured?
A: Track the share of recommendations accepted, the time from recommendation to action, and the number of operator-initiated corrections. Rising corrections usually indicate engagement rather than failure.
