Platform · Living AI
It answers in your numbers, or it does not answer.
Ask the portfolio a question in plain language and get a reply built from the same figures the page is showing, with the trace behind it. The value is not that it writes fluently. It is that it will not invent.
Which assets drove the Scope 2 increase last quarter?
Bankside and Fitzrovia account for 81% of the increase. Bankside sits 14.2% above its trailing three month mean1, calculated on the 2024 factor set2, in a run you can reopen3.
edf-2024-q3.csv
DEFRA GHG 2024
run_2291
What was our Scope 3 category 11 figure for 2023?
There are no category 11 activity records for 2023, so there is nothing to cite. No figure is offered. In a regulated disclosure, a confident wrong number is worse than no number.
none
none
none
Living AI · scheduled briefing
To: sustainability@example.co.uk
Scope 2 rose 14% at Bankside last month.
- Bankside electricity is 14% above the trailing three-month mean.
- Fitzrovia gas has no readings for March, so the period is incomplete.
- Portfolio measured share improved from 79% to 82%.
Recommended: confirm the Bankside chiller schedule before the quarter closes.
The objection everybody has, answered first.
Nobody responsible for a reported figure wants a language model anywhere near it, and they are right. A confident sentence containing an invented number is worse than no sentence, because it will be believed, repeated, and eventually filed.
So the assistant here does not generate figures. It reads the same governed dataset the dashboard reads, and every number in its answer is a number you can open. Where the data does not support an answer, it says so rather than producing a plausible one. The in-product disclaimer says the same thing in plainer words: responses are guidance, and critical data should be verified.
What it actually does
Answers portfolio questions in plain language.
What are our total emissions by scope, and the trend. What is our recycling rate, and which tenant produces the most waste. Any waste compliance gaps or transfer notes expiring soon. Questions a person would ask, answered without building a report first.
Reads the period for you.
A written read on the current period sits on the dashboard: what the footprint is, which scope moved, which building is the one to watch. The lede is computed from the page totals, not written about them.
Surfaces what needs a decision.
A building pulse and a flagged queue carry incidents, active issues and the supplier and tenant submissions waiting on your approval, so the things that need a person reach one.
Briefs you on a schedule.
An AI briefing sits alongside the report set, and reports can be generated and delivered on a schedule rather than when somebody remembers.
Every claim opens.
A figure in an answer is the same figure in the lineage view, so the assistant is a faster way into the data rather than a second version of it.
It refuses.
A failed fetch returns a refusal, never a zero. The platform holds that line everywhere: a broken query is not an empty portfolio, and a silence is safer than a confident wrong number.
What it does not do.
It does not act on your behalf. It does not decide that a variance is material, choose who ought to hear about it, and send them an email without being asked. Those are decisions with consequences, and a system that takes them autonomously has to be right every time rather than most of the time.
Where a variance needs to reach somebody, it is surfaced for a person to act on. If that changes, this page will change with it.
Ask it something you already know the answer to.
That is the only honest way to judge one of these. Bring a figure you have checked by hand and see whether it agrees, and whether it can show you why.