ESG data extraction accuracy: 99%+.
Human review cut to under 5%.

Every ESG report started with a mountain of mismatched paperwork.

ESG data across the business was scattered and hard to act on. It came in every format imaginable, utility bills in different languages, handwritten purchase orders, scanned supplier certifications, with no consistent way to pull it together.

The stakes were high. ESG reports are public-facing and scrutinised by regulators and stakeholders alike, so every KPI needed to be accurate and traceable back to source. Manual extraction and review couldn’t keep pace, and it was delaying the first consolidated report.

Three constraints were non-negotiable:

• Data scattered across 400+ stores, 2,000+ suppliers and 10,000+ products
• Formats ranging from multilingual utility bills to handwritten purchase orders
• Public reports demanding full accuracy and traceability

A pipeline for every document type, one brain to route them.

The Fractional Head of AI built a suite of multi-modal ingestion pipelines, each specialised for a document type, utility bills, supplier certificates, purchase orders, handwritten notes, combining LLMs for text recognition with vision models for image understanding. An orchestration LLM inspects every incoming document and routes it to the right pipeline automatically.

Each pipeline validates its own output, flagging outliers, and a separate critique LLM independently assesses extraction accuracy. Anything below the confidence threshold gets raised for human review inside an approval workflow, so the ESG team signs off before any KPI goes live. End-to-end data lineage traces every KPI back to its source documents, extraction confidence and reviewer decisions.

Business users get the insights through a dashboard built around different reporting hierarchies, business units, countries, regions, stores, and a chat interface that answers complex questions directly, returning tables and graphs for queries like which store uses the least renewable energy or which supplier sources the most organic silk.

One platform, 99% accuracy, a fraction of the manual review.

The ESG team now works from a single pane of glass instead of scattered spreadsheets and documents. The first ESG report came together faster than any before it, and the underlying data, lineage-backed and independently critiqued, was solid enough to prioritise real sustainability initiatives, from energy hotspots to supplier sourcing decisions.