In short
Start by mapping where your company uses AI, then make the efficient choice the default: the smallest model that meets the quality bar, short outputs, AI features off unless needed. Ask suppliers for energy per request and where they run, measure per useful task, and put it in a short policy with an owner. Most of these steps cut cost and delay as well as energy.
Steps
- List the AI tools, features and automated workflows in use and roughly how often each runs.
- Test whether a smaller model or a non-AI method meets the quality bar for your highest-volume tasks.
- Set efficient defaults in company tools, such as smaller models, output limits and features off unless needed.
- Ask suppliers for energy per request, where workloads run and how they report it.
- Track energy or tokens per useful task and review the figures every quarter.
- Write the rules into a short policy with a named owner and a review date.
Make the efficient choice the default, then measure whether it is working. For most companies the energy is in running models (inference) many times a day, not in training them, so everyday choices about models, prompts and features matter most.
Two findings show how much room there is. Luccioni, Jernite and Strubell found general-purpose generative models orders of magnitude more expensive to run than task-specific ones for the same tasks. UNESCO and University College London found that smaller, task-specific models could cut energy use by up to 90%, and shorter prompts and responses by more than half.
Where do I start?
- Map your AI use: see how to map where your organisation uses AI.
- Right-size the busiest tasks: try small models first; see choose a smaller AI model.
- Set defaults: switch off features nobody uses; see switch off unneeded AI features.
- Ask suppliers: see sustainability in AI procurement.
- Measure per task: see track AI energy per task.
- Write it down: see write a green AI policy.
What makes the biggest difference
Model choice on high-volume tasks, and turning off AI that runs without anyone needing it. Staff habits matter, but defaults reach everyone without relying on memory.
Common mistakes
- Counting cost per call instead of per successful task, so retries stay hidden.
- Treating a green cloud contract as the whole answer while usage keeps growing.
- Claiming savings before measuring a baseline.
The full approach is in our guide for business.
Sources
- Smarter, smaller, stronger: resource-efficient generative AI and the future of digital transformation, UNESCO and University College London (2025)
- Power Hungry Processing: Watts Driving the Cost of AI Deployment?, Luccioni, Jernite and Strubell, ACM FAccT (2024)
Related
Green AI for business
A practical guide for businesses on cutting the energy, cost and carbon of AI: model choice, suppliers, measurement, procurement and honest claims.
How do I map where my organisation uses AI?
How to build a simple inventory of AI in an organisation: tools, built-in features, automations and suppliers, with volumes, owners and data.
How do I write a green AI policy?
What a short green AI policy should say: when to use AI, which defaults to set, what to ask suppliers, how to measure, and who owns it.
Choose the right model for the job
How to right-size AI: when to use small or specialised models, when a frontier model earns its cost, how routing works, and when not to use AI.