Managing energy use across 85 factories with minute-by-minute data.
Meter data, a 48-hour forecast and waste alerts in one platform; the energy team sees waste before the bill arrives.
Problem and approach
Problem
In a company with multiple facilities, energy consumption often becomes visible only when the bill arrives.
Meters from different manufacturers use their own protocols, leaving facility comparisons to manually assembled spreadsheets. A compressor idling overnight or a load running during peak hours disappears into the month-end total.
Approach
We connect the 85 facilities in stages, starting with pilot sites rather than all at once.
In a two-week discovery sprint, we map pilot-site meters to machines and test a prototype with actual users in week three. The jointly scoped first phase, such as monitoring and waste alerts at pilot facilities, targets a launch in 6–12 weeks; forecasting and other facilities are added in later phases.
Manage energy every minute, not just at month-end.
Minute-by-minute consumption monitoring
Each meter sends a reading every minute; consumption is traced from facility to line to individual machine.
Anomaly and waste alerts
The platform learns each machine's expected consumption by shift. Deviations and their costs are routed to the relevant team.
48-hour consumption and cost forecast
Consumption and cost are forecast two days ahead with a confidence interval. Tomorrow's peak load is visible today.
Peak-period load-shifting recommendations
Identify loads that can be deferred from peak hours: which machine, at what time and with what monthly saving.
Scenario and scope
Example users
Energy manager, facility manager and maintenance team.
Example workflow
Meter data is collected, missing readings are flagged and measurements are compared with shift information. Deviations are assigned to the relevant team; the facility manager decides whether to apply load-shifting recommendations.
Decisions for a real project
Choose meter protocols, measurement frequency, historical data requirements, facility and machine mappings, alert thresholds and human approvals. Forecasting or recommendation models are enabled only if pilot evaluation is sufficient; the first phase may cover monitoring only.
Related service: Custom software development; for forecasting and anomaly scope, see AI integration. You can list your own users and systems in the project brief tool.
Timelines, teams and technologies in this scenario are illustrative planning choices; actual scope, schedule and team are defined during discovery. Product names do not imply partnerships.
What we measure
These targets are example measures defined for this scenario; actual results vary by project.
Technology and team
Technology
- Python
- Kafka
- MQTT
- TimescaleDB
- PyTorch
- React
- Kubernetes
- AWS
We explain data preparation, human approval and pilot planning for forecasts and anomaly alerts on the AI integration page, and other scenarios in our AI in business processes guide.
Suggested team 7 people
- 1 data scientist
- 1 product designer
- 4 software engineer (web, mobile, backend)
- 1 data engineer
Your process
Do you manage a similar operation? Let's schedule a 30-minute conversation about turning your factories' energy data into decisions.
Let's discuss this conceptThis is an example, not a limit; we apply the same approach to similar processes in other industries.