— May 18, 2026 · 9 min read
CSRD/ESRS data pipeline: 2026 reporting architecture
The CSRD directive (Corporate Sustainability Reporting Directive, 2022/2464, adopted on 14 December 2022) requires large European companies to publish detailed reports in line with the ESRS (European Sustainability Reporting Standards) issued by EFRAG. First reports due in 2025 for fiscal year 2024, progressive widening until 2028.
ESRS data scope
ESRS published by EFRAG (delegated act of 31 July 2023, applicable from 2024) cover twelve standards: two cross-cutting (ESRS 1 General Requirements, ESRS 2 General Disclosures) and ten topical standards in E (environment: E1-E5), S (social: S1-S4) and G (governance: G1). The most voluminous is ESRS E1 Climate Change: GHG Scope 1, 2 and 3 emissions, transition plan, physical and transition risk vulnerability, internal carbon pricing, EU Taxonomy alignment.
The specificity vs financial reporting: a large share of data comes from non-accounting sources — IoT sensors for energy consumption, carrier BoLs for upstream Scope 3, fleet fuel invoices, HR data for S1 (workforce), IT CMDB for Scope 2 electricity. This imposes a specific data architecture.
Target architecture: three layers
A proven architecture has three layers: (1) Ingest layer — connectors to ERP (SAP, Oracle, MS Dynamics) for invoiced consumption, to CMMS (IBM Maximo, SAP PM) for meter readings, to TMS for Scope 3 transport, to HR (Workday, SAP SuccessFactors) for ESRS S1, to partner EDI for primary upstream Scope 3 data. Recommended pivot format: JSON with versioned schema stored on object lake (S3, Azure Blob, GCS) with object lock for immutability.
(2) Compute layer — aggregation and GHG conversion engine (multiplication by emission factors, currency handling, unit conversion). Typical architectures: data warehouse (Snowflake, BigQuery, Databricks) with dbt for versioned transformations, and a workflow orchestrator (Airflow, Dagster) for pipelines with error handling and retries. Emission factors are a dedicated library (Base Carbone ADEME, EPA eGRID, IEA, DEFRA) that must be versioned — annual revaluation of factors changes all historical reports if recomputed.
(3) Reporting layer — CSRD report generation in XHTML+iXBRL format (inline XBRL) per the ESRS XBRL taxonomy published by EFRAG (first version July 2024, updated May 2026). Production assisted by specialised tools (Workiva, Diligent, Tagetik, IBM Envizi, SAP Sustainability, Microsoft Sustainability Manager) or by custom generation via XBRL libraries (Arelle, FreedomXBRL).
Scope 3: the central challenge
Scope 1 (direct combustion) and Scope 2 (purchased electricity) are relatively well-tooled since the GHG Protocol Corporate Standard (2004). Scope 3 — the 15 categories of indirect upstream and downstream emissions — remains the major challenge. For large industrial companies, Scope 3 typically represents 70-90% of total. The most critical categories: cat. 1 Purchased goods & services, cat. 4 Upstream transportation, cat. 11 Use of sold products, cat. 12 End-of-life treatment.
Two approaches: (a) spend-based (multiply purchase in EUR by a sector monetary factor — easy but coarse), (b) activity-based (collect primary data from the supplier via EDI or shared platforms like Watershed, Persefoni, Sweep, Greenly). Mature 2026 organisations target activity-based on ≥ 50% of critical purchases.
To retrieve primary supplier data, EDI emerges as a natural channel: specialised messages (CARBONIC in EDIFACT, UBL extension with custom EmissionsReport, or CSV format via PEPPOL) are starting to appear. No universal standard yet in 2026 — convergence expected 2027-2028.
The ESRS XBRL taxonomy
The ESRS XBRL taxonomy published by EFRAG is the technical pivot of the CSRD report. It defines ~1 144 quantitative datapoints and ~3 000 qualitative datapoints. The report is published in xHTML with inline XBRL (iXBRL technique), allowing human reading AND machine extraction. The auditor verifies compliance (limited assurance since 2025, reasonable assurance expected later).
For data teams, the key work is the mapping of internal datapoints to ESRS taxonomy datapoints — typically 4-8 months of effort for a large group. Workiva or Tagetik tools facilitate but do not replace this step which requires accounting + ESG + IT expertise.
Audit trail: the invisible essential
Specificity of CSRD reporting vs financial reporting: sources are heterogeneous (sensors, supplier invoices, HR declarations, ERP ETL) and numerous (typically 50-200 sources for a large group). The auditor will trace each published number back to its primary source.
The data architecture must therefore maintain an immutable data lineage: every transformation versioned (dbt with its snapshots, or storing each Airflow run), every emission factor used tracked by vintage, every human reprocessing logged. Typical tools: OpenLineage, DataHub, Marquez. Without these, the auditor cannot verify and refuses certification.
Conclusion: a data subject above all
CSRD/ESRS is often perceived as an ESG or accounting topic, and partly it is. But operationally, it is a data project: multi-source ingestion, transformation, traceability, governance. EDI teams have a natural role on the upstream Scope 3 side, which will increasingly go through standardised EDI channels. The architectural choices made in 2025-2026 will structure large groups' reporting capacity for the decade.