Sustainability

Explanation

Sustainability tracks the environmental footprint of running AI workloads so a bank can manage and reduce the energy and carbon cost of its models — relevant as chat, RAG, and document-extraction services scale to millions of calls. It matters because unchecked compute consumption raises both operating cost and ESG/climate exposure that regulators and boards increasingly scrutinise. It is measured by the Token-Energy / Carbon Proxy metric, which estimates energy use and associated carbon from processing volume (such as tokens consumed) as a stand-in for direct measurement. To implement it, instrument the inference and extraction layers to capture per-request token and compute usage, multiply by an energy-per-token and grid-carbon factor to derive the proxy, and aggregate into dashboards by model, use case, and time period; log usage records and methodology assumptions as evidence. Run collection continuously as telemetry and roll it up for periodic ESG reporting. The threshold is to track and report the proxy and reduce it 5% year over year; failing to meet the 5% YoY reduction triggers review of model right-sizing, caching, batching, or hardware/region choices and an action plan, supporting RBI Safety, Resilience and Sustainability and NIST MANAGE 2.2 resource-efficiency goals.