Sustainable AI, measured rather than promised
Every model we deploy uses compute, and compute uses energy. We choose the smallest model that does the job, measure what runs, and report it to you.
Three principles
- 01
Smallest model that works
We test a smaller or open model before reaching for the largest one. Most classification and extraction tasks don't need it.
- 02
Measured, not assumed
Production systems log estimated energy per 1,000 requests, so you see the cost of a feature, not just its accuracy.
- 03
Region by energy mix
Where latency and data-residency rules allow, we host in cloud regions with a lower-carbon grid.
What you get in every project
- The reasoning behind the model choice, in the design document
- A monthly report of usage and estimated energy
- A switch-off plan for pilots that don't go live
Our commitments
- From 2026 Report estimated energy use for every production system we operate.
- By 2027 Publish an annual statement of estimated emissions from the systems we run.
- By 2027 Start every new deployment in a low-carbon region where data rules allow.
Sustainability programmes for your operations
Beyond the AI we build, we run sustainability programmes for companies that want a smaller footprint and an operation that lasts.
-
Carbon footprint
Tracking and reporting for Scope 1, 2 and 3 emissions, so you can measure, analyse and cut them toward net zero.
-
Energy efficiency
AI reads usage patterns to manage energy. Reductions of up to 30% are possible (estimate; it depends on the site).
-
Circular operations
Waste reduction, better recycling and resource efficiency, with material flows tracked end to end.
-
Sustainability reporting
Automated reporting so you can share ESG performance with stakeholders.
Aligned with GRI, SASB, TCFD, ISO 14001 and ISO 50001.
Want this built into your AI plan?
The assessment includes a governance section that covers model choice and running costs.
