Agentic Workflow for Automated Calculation of Market-Based vs. Location-Based Emissions

Dual Scope 2 reporting—calculating both location-based and market-based emissions—is a complex, data-intensive operational bottleneck. It requires ingesting Energy Attribute Certificates (EACs), contractual instruments, and grid emission factors from fragmented sources like utility portals, procurement systems, and third-party databases. Manual consolidation is error-prone and consumes hundreds of analyst hours annually, delaying disclosures and increasing assurance costs. A custom automated workflow directly addresses this by orchestrating specialized agents to perform these calculations continuously, ensuring methodological rigor and audit readiness.

Implementation centers on a LangGraph or similar orchestration layer that sequences agents for data retrieval, applies calculation logic, and routes exceptions. Key architectural decisions involve integrating with SAP or Oracle ERP for spend data, browser agents for portal scraping where APIs are absent, and a centralized emission factor database. Controls are critical: each data point requires lineage tracking, calculations must be versioned, and outputs undergo automated validation against historical ranges before approval gates. This creates a production-grade system that turns a quarterly manual crunch into a continuous, controlled process, reducing reporting effort by over 80% and materially lowering compliance risk.

This orchestration layer is triggered by a data-ready event, such as a monthly utility bill ingestion or a new Energy Attribute Certificate (EAC) transaction. A central orchestrator agent activates specialized sub-agents, each responsible for a discrete task: fetching grid emission factors from a database like eGRID, parsing contractual instrument data from procurement systems, and validating EAC serial numbers against registry APIs. This decomposition ensures each agent operates within a bounded, auditable context, isolating failures and simplifying updates to methodology or data sources.

Post-calculation, a validation agent reconciles the dual figures, flags anomalies against historical trends, and routes results through a mandatory human-in-the-loop approval gate before finalization. This gate, integrated into platforms like ServiceNow or Jira, provides operational control. The final, approved data is written to an immutable audit database with full lineage, ready for framework-specific reporting packages or direct API publication to CDP and CSRD platforms, ensuring defensible disclosure.

Phase 1 establishes the core calculation engine and data ingestion agents. We deploy a containerized orchestrator, such as LangGraph, to manage specialized agents for pulling Energy Attribute Certificate (EAC) data from registries, grid emission factors from authoritative databases, and consumption data from utility APIs or ERP systems like SAP. This pilot validates the dual-calculation logic (market-based vs. location-based) against a controlled set of facilities, creating a trusted baseline and defining exception-handling patterns for missing or anomalous data before scaling.

Phase 2 scales the workflow across the enterprise portfolio, integrating with core financial and operational systems. Agents are configured to pull from hundreds of data sources, with robust retry logic and data quality scoring. A human-in-the-loop layer is introduced for outlier review and sign-off before figures are committed to the reporting database. The final phase hardens the system with immutable audit logging, automated reconciliation checks, and observability dashboards, ensuring the workflow delivers audit-ready, framework-mapped emissions data for both internal management and external disclosure.

The initial rollout phase focuses on a single facility or business unit, using a parallel-run strategy where the automated workflow's outputs are compared against manual calculations for a full reporting period. This sandbox environment validates the agentic logic for ingesting EACs, grid factors, and procurement contracts, while establishing baseline performance for data completeness and calculation accuracy. Key controls here include detailed calculation logs and a manual override interface in the orchestration layer (e.g., LangGraph) to pause or correct agent decisions.

Post-validation, the workflow scales under a phased geographic or divisional expansion, governed by automated approval gates. Each calculation run is preceded by a data-quality check, with exceptions routed to a designated sustainability analyst via ServiceNow or a custom console. Final outputs are written to an immutable ledger and the central ESG platform (e.g., Salesforce Net Zero Cloud), triggering dashboard refreshes. This controlled approach mitigates disclosure risk, embeds human oversight, and creates a repeatable operating model for global deployment.