Skip to main content

Developing a sustainability action plan

Instruction and application
Complete

High-level slogans do not change emissions. A sustainability action plan turns goals into SMART outcomes, metrics, operating practices, governance and stakeholder communications.

SMART goals (examples)​

  • Specific: reduce training CO₂e for flagship models by focusing on region mix and architecture choices.
  • Measurable: track kWh per training job and CO₂e estimates monthly.
  • Achievable: pilot on one product line before enterprise rollout.
  • Relevant: aligns with corporate net-zero commitments.
  • Time-bound: deliver first baseline report in one quarter, 10% intensity reduction in two quarters.

Metrics and operational strategies​

Track energy,carbon,fairness audit outcomes,latency per cost unit anddata storage efficiency. Operationally: reduce redundant retrains, adopt autoscaling, follow Green Software Foundation practices, and keep datasets minimal but well versioned.

Governance​

Stand up an AI ethics or sustainability committee, assign asustainability owner paired with engineering and compliance leads, and publish how decisions are escalated.

Stakeholder engagement​

Employees need training; customers need transparent reporting; investors and regulators need consistent metrics; external groups such as GSF or RAI Institute provide benchmarks.

Case study snapshot: EnerSys and GenAI for ESG​

EnerSys used generative AI to consolidate ESG data, accelerate reporting and spot operational improvements—see the Thomson Reuters ESG case study. Themes: break data silos, automate draft reporting, highlight optimisation opportunities, strengthen supplier visibility.

Pause and think​

Which two metrics would you put on an executive dashboard first for AI sustainability in your organisation?

Action item: Quiz​

Question 1 of 2
What makes a sustainability objective “measurable” in an AI programme?
  • A. It uses the word “green” in the title
  • B. It references a vendor logo
  • C. It defines quantitative indicators (for example kWh, CO₂e, disparity metrics) and a collection method
  • D. It promises “best-in-class AI” without definitions
Correct Answer: C

Feedback: SMART measurement needs numbers, sources and owners.

In the EnerSys-style pattern, what problem did GenAI most directly help address?
  • A. Fragmented ESG data and heavy manual reporting workflows
  • B. Lack of Python installed on laptops
  • C. GPU driver incompatibility
  • D. Eliminating the need for human review of all ESG claims
Correct Answer: A

Feedback: The case emphasises integration, reporting efficiency and insight generation—not generic tooling issues.