Embed energy data in production planning to cut emissions

Submitted through the COP31 Sustainable Transformation Awards · SKD Türkiye (WBCSD Global Network Partner)

The company manufactures technical textiles and reinforcement materials in Türkiye and employs more than 1,000 people.

Production in this sector is energy intensive, and electricity is the largest single component of that consumption. Rising energy costs and low-carbon production requirements made energy performance a commercial issue rather than only an environmental one.

The specific gap was in decision making rather than in measurement. Energy data was already collected and reported, but production planning decisions were taken on capacity, delivery date and operational suitability alone. Energy efficiency and emissions performance were not active inputs, so the plan that looked best on capacity could be the most energy intensive option available and nobody would know.

Energy management therefore functioned as a passive reporting activity: figures were produced after the event, at a level of aggregation that could not be traced back to the scheduling decisions that caused them.

The objective set was to integrate production, machine and energy data so that the conditions under which a given item can be produced with the lowest energy consumption can be identified, and so that sustainable production becomes part of daily operational decisions rather than a separate reporting cycle. The project started in 2025, using 2021 as the baseline year for the data analysis.

Location of the initiative: Türkiye

The platform is a data-driven decision support system that sits inside production planning rather than alongside it.

Its foundation is an integrated dataset. Records from the plant's manufacturing execution system and from its energy metering infrastructure were brought together, producing 4.09 billion rows of production and energy data covering the period from 2021 to 2025. That volume is what makes it possible to model consumption at the level of an individual machine running an individual product style rather than at plant level.

From that dataset, validated models establish which machine and which operating parameters produce a given product style with the lowest energy consumption. The models reach 93% accuracy on 10 minute interval data and 96% on order-based analysis, which is what allows their output to be used in scheduling rather than treated as indicative.

The models feed a recommendation mechanism. When a schedule is built, capacity and operational suitability remain constraints, but energy efficiency and emissions performance are now active inputs to the same decision, so the planner sees the energy consequence of a routing choice at the point of making it.

An anomaly detection layer runs on the same data. Deviations in energy consumption are identified as they occur and the relevant maintenance and operations teams are informed proactively, so that losses and inefficiencies are addressed early rather than appearing in a monthly total.

The platform was deliberately built on infrastructure the company already owned — the existing manufacturing execution system, energy management systems and corporate data infrastructure — and the people, data and processes needed to run it were established in house. It is designed to extend beyond electricity to natural gas, compressed air, water and other utilities, and the natural gas optimisation phase is under development.

Sustainability impact

Climate

The initiative targets Scope 2 emissions, because it works by reducing the electricity consumed in production and therefore the emissions associated with purchased electricity. The baseline year is 2021, and the analysis covers the period from 2021 to 2025.

The results reported so far are identified potential rather than realised abatement, and are stated as such. Analysis of the recommendation mechanism identified an energy saving potential of approximately 7%, which corresponds to approximately 8.5 million kWh of electricity a year and to more than 3,500 tonnes of CO2 emission reduction, on the basis that the recommendations are followed in scheduling.

Emission effects were assessed using internationally accepted methodologies and reported as a carbon reduction effect. The electricity saving is converted into tonnes of CO2 in accordance with the GHG Protocol accounting basis, using Türkiye's 2025 emission factor of 0.0004340 tonnes per kWh published by the Energy Efficiency and Environment Department of the Ministry of Energy and Natural Resources.

Realised savings measured against that identified potential are not yet reported.

The indicator set is total energy consumption in kWh, specific energy consumption in kWh per tonne, CO2 emissions in tonnes, data accuracy as a percentage and financial saving in USD. Reporting specific energy consumption alongside the total is what allows the result to be read independently of production volume.

Business impact

Benefits

The identified saving carries a financial value of approximately USD 777,000 a year, arising from the same 7% reduction in electricity consumption, so the environmental and commercial cases rest on one set of operating changes rather than two.

The anomaly detection layer produces benefit independently of the scheduling recommendations. Deviations in consumption are flagged as they occur and maintenance and operations teams are informed proactively, which prevents energy losses and equipment inefficiencies from persisting between reporting cycles.

Planning quality improves in a more general sense. Adding energy performance to capacity and delivery date as a decision criterion gives planners a further basis on which to choose between otherwise equivalent options, and the models make explicit a relationship between product style, machine and consumption that was previously held informally.

Because the platform runs on the existing manufacturing execution system, energy management systems and corporate data infrastructure, no parallel technology stack has to be funded or maintained, and the capability transfers to the natural gas, compressed air and water optimisation phases without new foundations.

Senior management receives the energy optimisation results and tracks the rate at which the recommendation mechanism is actually used in planning, which keeps the benefit tied to adoption rather than to the existence of the tool.

Costs

The cost base is analytical and organisational rather than physical. It consists of the data engineering required to integrate production and energy records into a single dataset, the development and validation of the consumption models, the effort of maintaining data quality across 4.09 billion rows, and the internal team that operates the platform.

The platform was built entirely with the company's own internal resources over approximately six months, by a five-member internal team that also carries its operational maintenance. Because it runs on existing corporate infrastructure, no external development or licensing cost was incurred.

There is an organisational cost in changing how schedules are built. Planners have to weigh an additional criterion, and the recommendation is advisory: an energy-optimal routing can conflict with a delivery commitment or a capacity constraint, in which case the saving is not taken. The realised benefit therefore depends on adoption rates rather than on model quality alone, which is why usage in planning is tracked as an indicator.

Model maintenance is a continuing obligation. Accuracy of 93% on 10 minute data and 96% on order-based analysis has to be re-validated as machines, product mix and operating conditions change, or the recommendations will drift away from the plant's actual behaviour.

Costs are contained by building on systems the company already operates rather than procuring a separate platform, by keeping the people, data and processes in house so the capability is retained, and by extending the same infrastructure to further utilities instead of repeating the integration work for each one.

Impact beyond sustainability and business

Co-benefits

The same dataset supports work beyond energy. Once production and consumption records are integrated at machine level, the infrastructure serves scheduling, maintenance planning and process analysis, so the investment is not consumed by a single application.

The approach extends to other resources by design. The platform is built to cover natural gas, compressed air, water and other utilities, and the natural gas phase is already in development, which means the method rather than the model is the transferable asset.

Anomaly alerts change the relationship between energy management and maintenance. Consumption deviation is frequently an early symptom of a mechanical problem, so the same signal that protects energy performance also supports equipment reliability.

Because the platform is designed to be applied at other plants and in other geographies, an energy optimisation model proven at one site becomes a template for a group rather than a local tool.

Potential side-effects

The reported figures are potential, not delivered. An identified 7% saving, approximately 8.5 million kWh and more than 3,500 tonnes of CO2 assume that the recommendation is followed; a company reading these as achieved results would overstate its position, and the distinction has to be maintained in every disclosure until realised savings are measured.

Optimising a schedule for energy can work against other objectives. Consolidating production onto the most efficient machines affects delivery dates, changeover frequency and equipment utilisation, and those trade-offs sit with the planner rather than with the model.

A model that is 93% to 96% accurate is still wrong some of the time, and accuracy measured on historical data does not guarantee performance on a product mix that has changed. Continuous re-validation is a requirement rather than a refinement.

Dependence on a single integrated dataset creates a dependency on data quality and on the systems that feed it. If metering or production reporting degrades, the recommendations degrade with it, and the effect is not immediately visible to the planners relying on them.

Typical business profile

The model suits energy-intensive manufacturers running many product variants across machines with different efficiency characteristics, where the same item can be produced on more than one route and where the choice between routes is currently made without reference to energy.

It is most relevant where a manufacturing execution system and machine-level energy metering already exist, and where several years of history can be assembled, because the models depend on a dataset large enough to separate the effect of the machine from the effect of the product and the operating conditions.

Delivery engages a digitalisation and production systems team as the owner, working with production, planning, maintenance, energy management and sustainability functions, with senior management sponsorship to ensure the recommendations are actually used.

Approach

1. Connect production and energy records at machine level: Integrate the manufacturing execution system with the energy metering infrastructure so that consumption can be attributed to a specific machine running a specific product style, which is the resolution the modelling requires.

2. Assemble a history long enough to model: Build the dataset over several years — here 4.09 billion rows covering 2021 to 2025 — so that seasonal variation, product mix changes and machine ageing are represented rather than treated as noise.

3. Model specific consumption per product style and machine: Establish which machine and which operating parameters produce each style with the lowest energy consumption, and express the result as specific energy consumption per tonne so it is comparable across volumes.

4. Validate accuracy before letting the model influence decisions: Test the models against measured consumption at more than one time resolution — here 93% on 10 minute data and 96% on order-based analysis — and publish the accuracy alongside the recommendation.

5. Turn the model into a recommendation inside the planning tool: Deliver the output where scheduling decisions are made rather than in a separate report, so that energy efficiency and emissions performance become active inputs alongside capacity and delivery date.

6. Add anomaly detection and route the alerts: Monitor consumption for deviation and inform maintenance and operations teams as it happens, so that energy losses are corrected at the point of occurrence.

7. Fix a five-part indicator set: Track total energy consumption, specific energy consumption per tonne, CO2 emissions, data accuracy and financial saving, and keep data accuracy in the set because the credibility of the other four depends on it.

8. Report adoption to senior management, not only savings: Track how often the recommendation is used in planning as well as the energy performance itself, because an unused recommendation delivers nothing regardless of its accuracy.

9. Extend the platform to other utilities: Apply the same integration and modelling method to natural gas, compressed air and water, reusing the data infrastructure rather than rebuilding it for each resource.

Stakeholders involved

• Project leads: The project is sponsored by senior management as a strategic digital transformation initiative aligned with the company's sustainability and operational excellence objectives. Energy optimisation results are shared with senior management on a regular cycle, and the usage rates of the recommendation mechanism within production planning are tracked alongside energy performance. That combination is what keeps energy efficiency positioned as an element of corporate performance rather than as a technical matter delegated to a specialist function. The digitalisation and production systems team is responsible for delivery.

• Company functions: Production, planning, maintenance, energy management and sustainability teams take an active role alongside the digitalisation and production systems team. Planning applies the recommendations and reports where they cannot be followed; maintenance and operations act on the anomaly alerts; energy management and sustainability own the indicator set and the reporting; production provides the operating context that the models have to reflect. Working this way has established shared responsibility for energy efficiency across functions that previously held separate objectives. Continuity is secured by building on the existing manufacturing execution system, energy management systems and corporate data infrastructure, and by making the necessary people, data and operational processes permanent within the organisation. The solution is in active use in production planning and is being integrated into the next generation of scheduling applications.

• Main providers: No external solution provider led the development. The platform was built in house on the plant's existing manufacturing execution system and energy metering infrastructure, both supplied by established industrial software and equipment vendors, and the modelling, integration and operation were carried out with internal capability. That choice is significant for replication: the barrier for another manufacturer is analytical skill and data quality rather than procurement of a proprietary platform.

• Other: The project was advanced with internal stakeholders. External parties were not engaged in its development, which reflects a deliberate decision to retain the data, the models and the resulting capability inside the company. The roadmap covers natural gas optimisation, energy management for utility services and the extension of the application to other plants, so the intended reach grows beyond the site where it was built.

Key parameters to consider

The baseline year is 2021 and the analysis covers 2021 to 2025, drawing on 4.09 billion rows of production and energy data. The project itself started in 2025.

Model accuracy is 93% on 10 minute interval data and 96% on order-based analysis. Data accuracy is carried in the indicator set rather than assumed, because the value of the recommendations depends on it.

The indicator set is total energy consumption in kWh, specific energy consumption in kWh per tonne, CO2 emissions in tonnes, data accuracy as a percentage and financial saving in USD.

Prerequisites are a manufacturing execution system, machine-level energy metering, a corporate data infrastructure capable of holding several years of high-frequency records, and analytical capability retained in house.

The results stated — approximately 7%, approximately 8.5 million kWh and more than 3,500 tonnes of CO2 a year, with approximately USD 777,000 of financial benefit — are identified potential conditional on the recommendations being applied in scheduling, and should be treated as targets until realised savings are reported.

Implementation and operations tips

Deliver the output where the decision is taken. Energy dashboards have existed in manufacturing for years without changing schedules; what changed here was placing the recommendation inside the planning process so that it competes with capacity and delivery date at the moment of choice.

Publish model accuracy with every recommendation. Planners will not override an established scheduling practice on the basis of a model whose reliability they cannot see, and the 93% and 96% figures are what earned the models their place in the decision.

Measure adoption separately from savings. The gap between identified potential and realised benefit is usually a usage problem rather than a modelling problem, and it is invisible unless usage is tracked as an indicator in its own right.

Keep specific consumption per tonne in the indicator set. Total consumption moves with production volume and will hide an efficiency improvement in a growing year, whereas the specific figure isolates the effect of the change.

Build on the systems already in place. Using the existing manufacturing execution system and energy metering meant the effort went into modelling rather than into infrastructure, and it is why the same foundation can now be extended to natural gas, compressed air and water.