Can GenAI fill banks' emissions data gaps?

This paper presents new evidence highlighting significant limitations in emissions data for major listed banks in the euro area, sourced from four leading providers. Notable issues are data gaps, inconsistencies across providers and high volatility over time in scope 3 emissions, which are often inexplicably lower than scope 2 emissions and do not correlate with the banks' exposure to high-emitting sectors. The paper then examines whether Generative Artificial Intelligence (GenAI), which relies on broader and diverse information sets, can help bridge the existing data gaps by comparing the outcomes from three GenAI tools. We find that GenAI-based emissions data, either estimated or retrieved, are correlated with data from traditional sources and may therefore help to partially fill current gaps and identify anomalies in the available data. Nevertheless, they suffer from similar issues in terms of quality and consistency to those documented for the data supplied by professional providers. Additional concerns regard replicability and transparency, underscoring the need for initial caution in their use. Despite the current limitations, looking forward, GenAI may become a valuable complementary data source as models that are fine-tuned for this task are developed. Future improvements will also depend on the availability of more reliable underlying information, which in turn requires parallel progress in defining simple, clear and actionable measurement standards as well as regulatory developments to promote climate disclosure by non-financial corporations within banks' portfolios.