استخراج أرقام الاستدامة من التقارير

المشكلة

Analysts needed comparable environmental, social, and governance information out of long, inconsistently structured reports and scanned documents. Comparability is the hard part. Every issuer reports on its own template, with metrics in tables that differ in units, scope, and reporting period, and some of it exists only as a scan. Most of an analyst's review is therefore spent establishing what a given number actually refers to, before any comparison is possible.

ما بنيناه

We implemented a pipeline that ingested PDFs and images, extracted text and metadata, identified ESG concepts and metrics, linked each piece of evidence to its source page, and presented both document-level and portfolio-level comparisons.

ما توقّف عليه

The evidence boundary held throughout - raw values and source pages were never overwritten by normalised outputs, and a missing disclosure was never silently converted into a zero or an inferred commitment. That specific failure is what makes most ESG analytics untrustworthy, and avoiding it is what makes the throughput figure meaningful.