Improved private markets emissions estimation

Today we introduced an improved private-asset emissions estimation method. Private markets analysis is often difficult because of limited data availability. This new approach provides improvements to how we handle problems that can arise from limited data availability. We augment the existing Financial Only Metamodel, with a quality-control (QC) outlier detector and a robust fallback to simple emission-factor models.

While the B model performs well on average, it can produce extreme (10–100×) outliers for certain entities, especially with sparse or unusual inputs. Previously, there was no mechanism to detect these failures. The new QC model learns empirical emissions-intensity bounds from reported data by country and sector (with hierarchical fallback and minimum sample thresholds) and flags implausible predictions at inference time, routing them instead to industry-median SimpleEF estimates based on GPPE, revenue, or total assets.

This hybrid “B + QC + SimpleEF” approach preserves ML accuracy where reliable while preventing catastrophic errors, improving MdAPE across all scopes, most notably Scope 3 (+2.6%), with QC fallback rates of ~20%, and integrates cleanly into the existing pipeline with full backward compatibility.