Not configured yet. Open index.html and paste your Supabase URL and anon key into the CONFIG block at the top of the script, then redeploy. See README-DEPLOY.md.
Data collection · AI environmental disclosure
Map your organisation’s AI estate — together.
One link per department. Each team records the AI it actually uses; the project lead watches submissions arrive live, verifies them, and exports one clean dataset for the four-lens assessment.
Project lead
Start an engagement
For the Sustainability / IT / Finance coordinator. Creates your private console and the share link you’ll send to departments.
Adviser
Portfolio view
For advisers and PE: compare every engagement side by side — footprint, shadow-AI risk, defensibility — on one screen.
If your project lead sent you a link, just open it — no login needed. You’ll record the AI tools your team uses (it’s guided, and your draft saves automatically in this browser).
Lost the link? Ask your project lead to re-send it.
How the flow works
1 · Lead creates the engagement → 2 · departments submit via the share link → 3 · lead verifies & locks each submission → 4 · one consolidated dataset feeds the four-lens engine (carbon · water · renewables · governance).
Department submission
AI estate — your team’s tools
What to include — think beyond chatbots
Standalone AI platforms, AI features inside software you already use (CRM, design, productivity), and AI in products you sell. Empty categories are worth a second look — AI features are often switched on quietly. Just tell us where it runs; water and grid are worked out for you. Personal/free “shadow AI” is collected separately through an anonymous all-staff survey — you don’t report it here.
Not sure where a tool runs? How to find out ▾
Where a tool’s compute runs sets its grid-carbon, water risk and green-claim check — so it’s worth finding. How findable it is depends on who runs it:
Cloud you operate (Azure OpenAI, AWS Bedrock, Vertex): you chose the region — read it from the cloud console, your Terraform / infrastructure-as-code, or the billing export. Known.
Third-party AI (ChatGPT, Claude, Copilot, Gemini, AI inside SaaS you buy): the provider runs it. Check the enterprise admin for a region / data-residency setting; then ask for the provider’s sub-processor list with locations (UK/EU GDPR Art. 28) and put region disclosure in the contract.
Hardware / on-prem: your own site — Known.
Can you pin the region? Azure OpenAI, AWS Bedrock and Google Vertex — yes (Data Zone / geographic profile / regional endpoint). OpenAI — yes (per-project, or ChatGPT Enterprise residency). Anthropic first-party — US / global only (EU/UK via Bedrock or Vertex). M365 Copilot — EU users stay in the EU boundary, others may not. The default tier on every platform processes globally — pinning is always opt-in, so “we use Azure/AWS/Google” doesn’t by itself say where it runs.
A network traceroute usually shows the CDN edge you connect to, not where the model runs — treat it as a hint, not proof. If you genuinely can’t establish it, leave the region blank: it’s recorded as “undisclosed” (a finding, not a failure) and becomes a request to put to the provider.
Want provider-measured numbers instead of an estimate? Where to get them ▾
This tool estimates a footprint from usage; where a provider publishes measured data we use that instead — it upgrades a tool to the highest-confidence tier (“provider-measured”). What’s available depends on who runs the compute:
Cloud you operate — each hyperscaler has a native carbon dashboard:
AWS — the AWS Sustainability service (the former Customer Carbon Footprint Tool). Since late 2025 it includes Scope 3 — server manufacturing, energy, transport — on both market- and location-based methods. Monthly, with a lag; export the period as CSV.
Microsoft Azure — the Emissions Impact Dashboard (and Sustainability Manager in Fabric): Scope 1–3 by subscription and service.
Google Cloud — Cloud Carbon Footprint. From January 2026 data it also allocates AI-inference emissions down to the services that generated them, so Vertex / Gemini-on-Cloud now shows up.
Two cautions: these default to the market-based number (which leans on the provider’s renewable certificates — checked separately in the renewables lens), and none of them reports water.
Third-party AI you buy (ChatGPT, Claude, Copilot, Gemini, AI inside SaaS): providers don’t yet give you a per-customer footprint. Two published reference points exist — Google’s median Gemini prompt (~0.24 Wh, 0.03 gCO₂e, 0.26 mL water) and Mistral’s externally audited model LCA (~1.14 gCO₂e, ~45 mL per ~400-token reply) — but they are provider averages, not your usage, and aren’t comparable to each other. The useful ask to the vendor: your usage counts (tokens or prompts) and their region, so a benchmark can be applied.
Open models you self-host: you can measure directly — CodeCarbon logs the actual GPU energy of your inference, and the Hugging Face AI Energy Score rates model efficiency (1–5 stars, reasoning included) to guide right-sizing.
What to hand your adviser: the dashboard export for the reporting period, whether it is market- or location-based, and usage counts for anything bought as a service. What these tools miss: purchased AI SaaS and shadow AI never appear in a cloud dashboard — which is exactly what this inventory captures, so the two together give the full picture.
Submit to your project lead
Self-certification (Level 1). To the best of my knowledge this is a complete and accurate record of the AI tools my team uses in its work. (Personal/free “shadow” AI is covered by the separate anonymous survey, not here.)
Your data goes only to your project lead’s console for this engagement. You can update and resubmit until the lead locks your submission.
0 tools
Anonymous survey · 60 seconds
Which AI tools do you use for work — that weren’t provided by the company?
Completely anonymous · measurement, not enforcement
We’re measuring the environmental footprint of all AI used for work — including free and personal tools. The survey stores only the tools you tick, how often you use them, whether they touch company data, and (optionally) your broad area — no name, no email, no department. Nobody gets in trouble; honest answers are the whole point.
Tick any you use for work tasks
When you tick a tool, a small “how often” choice appears next to it — set it roughly (daily / weekly / occasionally).
Still anonymous — this one answer is what lets us judge the risk honestly, not just count tools.
Helps target training — skip it freely.
Nothing else is collected. If you use none of these, you can still submit — a “none” answer is useful too.
Thank you. Your anonymous response has been recorded. You can close this page.
Project-lead console
Engagement console
sets the coverage bar
Coverage
■ approved & locked ■ verified ■ submitted
The briefing is a self-contained page (print to PDF) built from live collection data — status, data quality, shadow-AI risk & cost, and the decisions to put in front of senior management. Clearly labelled provisional: the assessed four-lens results follow after lock & assessment.
Live picture live
What this means for each lead
The same live data, read four ways — for the people who have to act on it. Everything here is provisional and drawn from the lens views below.
Top actions this cycle
Provisional four-lens picture
Live signals from what departments have submitted so far (all submissions, verified or not). Provisional — the assessed results, including the CO₂e figure with its honest range, come from the four-lens assessment once the dataset is locked.
Scope: this is AI’s environmental footprint — one slice of the wider ICT footprint and the full materiality picture. It measures AI’s cost, not its handprint (where AI enables efficiency or avoided emissions — assessed separately). Read them together.
Collection
Department submissions
Waiting for the first submission… departments appear here the moment they submit.
Shadow AI
Anonymous all-staff survey
Shadow AI — anonymous all-staff survey
Shadow AI is the most sensitive data to collect, so it is not asked via departments. Send this anonymous survey company-wide instead — no name, department or device is recorded, which is what makes answers honest.
0
anonymous responses
0
distinct tools surfaced
Honesty note: responses are anonymous and unauthenticated — anyone with the survey link can respond, and nothing stops multiple responses per person. Treat counts as directional (that is also why the estate entry is flagged “directional”), and be sceptical of response counts that look implausible against headcount.
Cost to bring reported use into governed accounts: —
Low = consolidation: staff reporting × £/user/mo for one governed assistant (ChatGPT Business / Claude Team ≈ £16 list; edit to your quote). High = like-for-like: each reported tool at its governed seat’s list price (shown per tool below; verified Jul 2026, indicative). All figures annualised, and a floor — anonymous surveys undercount.
Adds one “Company-wide (anonymous survey)” submission with the aggregated shadow tools — flagged directional (self-selected sample). Re-fold any time before you lock it.
Distribution
Share links
Share with departments
Send this link to each department (add &team=Marketing to prefill the team name). Keep your own console link private — it contains your lead key.
Output
Consolidate & export
Consolidate
Only approved & locked submissions are included. Produces the single business-wide dataset in the exact format the four-lens engine reads.
Cycle snapshots & trend
Save an immutable point-in-time record each collection cycle — the honest way to build a trend (snapshots can’t be edited or deleted via the app).
No snapshots yet.
Start the next cycle
Each collection cycle is its own engagement (e.g. “Acme — Q4 2026”), so full tool-level history is kept per cycle. One click creates the next engagement and carries the current tools forward as unlocked drafts — departments update what changed instead of re-entering everything.
Compare with a previous cycle
Paste the previous cycle’s console or share link. Because each cycle keeps full tool detail, this shows the four-lens movement and a per-tool diff — deeper than snapshots.
Regulatory & advocacy landscape
Policy radar
What’s moving on AI-emissions transparency — regulation, standards, enforcement and the lobbying pushing disclosure your way. Every item dated, sourced and verified.
Loading…
Want to go deeper?
The radar is curated, not machine-generated — live web summaries can’t be verified to disclosure standard. For a bespoke deep-dive on any policy area (your sector, your jurisdictions, your providers), ask your adviser:
Score a prospective AI tool before it enters the estate. Same screening factors as everything else — so environmental impact becomes part of the buying decision, not an audit surprise later.
Pick a data type above
Radical transparency
Methodology & factors
Every factor this console uses, with its source — on screen, not hidden. Screening-grade and flagged indicative; the assessment verifies each against its current primary source before anything is disclosed.
Loading…
Adviser · multi-engagement
Portfolio view
Every engagement side by side. Add an engagement by pasting its console link (the one with ?e=…&k=…) — stored only in this browser.