AI-Driven Energy Automation Cuts Multi-Site Manufacturing Costs

Executive Summary 

A global manufacturing enterprise faced rising energy costs, fragmented visibility across plants, and limited intelligence in its legacy BMS. Akira deployed an AI-driven Energy Optimization Assistant using AWS IoT, Greengrass, and SageMaker to forecast demand, detect anomalies, and autonomously optimize HVAC and equipment loads.

The hybrid edge-cloud solution delivered a 28% reduction in energy consumption, 40% lower peak-demand charges, and significant cuts in manual interventions, achieving ROI within six months while enabling unified, real-time visibility and ESG-ready reporting. 

Customer Information 

• Customer: Confidential Global Manufacturing Client

• Industry: Industrial Manufacturing

• Location: India, Middle East

• Company Size: 4,500+ employees across 7 plants

Business Challenges 

The customer operated multiple manufacturing plants with energy-intensive equipment, HVAC systems, and complex load patterns. Due to rising tariffs, operational expansion, and dynamic production schedules, energy costs were increasing 12–17% annually. Their BMS offered limited intelligence, relying on manually configured rules that could not adapt to real-time variability such as weather conditions, occupancy, or machinery load. 

Key challenges included: 

• No unified visibility across sites → fragmented BMS dashboards 

• High peak demand resulting in additional penalties 

• Overcooling/overheating in non-critical zones 

• No predictive insights on equipment inefficiencies 

• Manual control interventions leading to human errors 

• Difficulty meeting internal sustainability KPIs and ESG reporting 

• Lack of integration between energy data, production cycles, and shift patterns 

• Inefficient load distribution causing unnecessary operating costs 

The customer needed an automated, scalable, AI-driven optimization solution that could intelligently balance cost, comfort, safety, and operational continuity. 

Technical Challenges 

The customer’s infrastructure had several limitations: 

• Legacy BMS systems without API-level access 

• No standardized data model across sensors, meters, and controllers 

• High data latency from edge devices 

• Inadequate storage for historical consumption and event logs 

• Non-uniform integration with PLCs and SCADA systems 

• Lack of forecasting capabilities for load and environmental dynamics 

• No automated anomaly detection for consumption spikes or equipment leakage 

Security and compliance requirements added further constraints: encrypted data flows, role-based access, network segregation, and integration with the customer’s internal IAM. 

Partner Solution 

Solution Overview 

Akira implemented its Energy Optimization Assistant—an autonomous AI agent that continuously monitors, predicts, and optimizes energy usage across sites. The agent ingests data from IoT sensors (temperature, humidity, occupancy), smart meters, BMS/PLC systems, and AWS IoT Core. It uses AWS SageMaker models to forecast demand, detect anomalies, and recommend (or autonomously trigger) optimization actions such as HVAC setpoint adjustments, load shifting, and equipment duty cycling. 

The solution runs in a hybrid architecture: AWS cloud for analytics and model training, and AWS Greengrass-enabled edge nodes for low-latency control. Dashboards built using QuickSight unify visibility for energy managers, sustainability teams, and plant operators. 

AWS Services Used 

• AWS IoT Core: Secure ingestion of sensor and meter data 

• AWS Greengrass: Edge-based execution of control logic and optimization actions 

• Amazon S3: Historical data storage + audit trails 

• Amazon Timestream: Real-time time-series queries for dashboards and insights 

• Amazon SageMaker: Model training for forecasting and anomaly detection 

• AWS Lambda: Event-driven triggers for agent decisions 

• Amazon CloudWatch: Monitoring and logging 

• Amazon QuickSight: Central dashboard for energy KPIs and ESG reporting 

Architecture Diagram 

1. Real-time Processing: Sub-50ms edge inference with AWS Greengrass for autonomous control decisions 

2. Multi-Agent AI: Coordinated forecasting, optimization, and anomaly detection agents 

3. Scalable Architecture: Serverless auto-scaling handles millions of IoT events per second 

4. Hybrid Edge-Cloud: Local processing for latency-critical actions, cloud for advanced analytics 

Implementation Details 

The implementation followed a 10-week Agile delivery model. 

Week 1–2: Discovery & Data Mapping 

• Audit of BMS, IoT devices, PLCs, and SCADA 

• Data schema standardization 

• Baseline energy performance modeling 

Week 3–5: AWS IoT + Data Lake Integration 

• Setup of IoT Core rules and device onboarding 

• Pipeline creation using Kinesis → S3 → Timestream 

• Edge Greengrass deployment at plants 

Week 6–8: ML Model Development 

• Load forecasting model built in SageMaker 

• Thermal comfort estimation model 

• Anomaly detection for energy leakage 

Week 9–10: Automation + Dashboard 

• Integration with HVAC/BMS control loops 

• Unified energy optimization agent built using Lambda + Greengrass 

• QuickSight dashboards for ESG and operational KPIs 

• Testing, validation, and controlled rollout across zones 

Security & compliance integration included IAM policies, encryption (KMS), network isolation (VPC endpoints), and audit logging. 

Innovation and Best Practices 

The solution leveraged AWS best practices across the Well-Architected Framework: 

• Edge computing deployed for sub-second control responses 

• Continuous model retraining using SageMaker Pipelines 

• Secure IoT onboarding using X.509 certificates 

• Real-time optimization using event-driven Lambda agents 

• Scalable time-series analytics using Timestream 

• Automated deployment using CI/CD + Infrastructure as Code 

• Modular multi-agent architecture enabling future extensions 

The project introduced unique innovations including hybrid ML inference, dynamic HVAC balancing based on occupancy prediction, and peak-demand prediction 45 minutes before occurrence. 

Results and Benefits 

Business Outcomes and Success Metrics 

The Energy Optimization Assistant delivered significant measurable impacts: 

• 28% reduction in total energy consumption across 7 plants 

• 40% reduction in peak-demand charges 

• 34% improvement in equipment efficiency 

• 55% reduction in manual interventions by facility operators 

• 20% improvement in ESG reporting accuracy 

• Immediate ROI with payback in <6 months 

The customer now has unified, real-time visibility into energy usage, equipment behavior, and zone-level performance across all facilities. 

Technical Benefits 

• 87% faster detection of consumption anomalies 

• Scalable data ingestion pipeline supporting 500k+ daily data points 

• Lower latency through edge-deployed inference 

• High availability architecture across multi-region setup 

• Improved data quality via Timestream-based time-series modeling 

• Strengthened security posture with fully encrypted data pathways 

Customer Testimonial 

"Akira’s Energy Optimization Assistant has transformed how we manage energy across our plants. The AI-driven automation delivered immediate savings and deep visibility we never had before." 
— Head of Operations, Leading Manufacturing Group 

Lessons Learned 

Challenges Overcome 

• Integration with legacy BMS systems required custom OPC-UA bridges 

• Normalizing inconsistent sensor data formats 

• Managing hybrid cloud-edge deployment due to low-latency requirements 

• Fine-tuning predictive models for varying plant conditions 

Best Practices Identified 

• Always start with baseline energy profiling 

• Deploy edge inference early for smooth automation 

• Use time-series storage (Timestream) instead of RDS for performance 

• Build modular agent components for future scalability 

• Align dashboards with both Ops KPIs and ESG metrics 

Future Plans 

The customer plans to: 

• Extend the Energy Optimization Agent to two additional plants 

• Integrate solar forecasting for renewable optimization 

• Use Reinforcement Learning for autonomous multi-objective optimization 

• Expand the ESG module to include Scope 3 reporting 

• Implement predictive maintenance using vibration and thermal imaging data