How we're different: The three ages of AI and sustainability

In the triple dog years of AI, evolution is happening at unprecedented speed across business and society. And in sustainability, while it’s been a little slow to get going, everyone is now selling AI capabilities. Every new entrant with a model behind a login is calling itself AI-native. It’s getting hard to tell what is a real jump in capability, and just how different the newer propositions are from ‘legacy plus AI’.

We wrote this because the wrong choice is not just a minor inconvenience. You will build years of work on top of the choices you make. If the foundation only ever knows a shadow of your business, everything you build on it inherits that limit.

So this is the guide we hope helps you tell the difference.

While many will be using co-pilot or one of the flagship LLM’s, the limits to using the tech in isolation for professional purposes are becoming obvious to us all. You need a volume and quality of data - and much more besides. Here we focus on how the ESG and carbon accounting platforms that you use are evolving to harness the technology, and the new categories that are emerging. Almost everything for sustainability on the market today falls into one of three groups.

Type 1: AI added to a system of record. This is reporting platforms (e.g. Watershed, Persefoni) with tools built on top. The data and how that data is recorded has not changed. They never captured the full context. There are two distinctions here: 1. The system still depends on someone entering the right numbers in the right place, and the AI can only ever reason over what survived that process, e.g. an incomplete spreadsheet, a supplier who never replied, a figure typed in from memory three weeks late. 2. It doesn’t understand ‘why’: You can add the most capable model in the world to some reporting software and it will still only know that your Scope 3 figure is what it is. It will not know why, what you left out, or what you could do about it. It reasons over a shadow of your actual footprint, not the footprint itself. Taken together this limits the value of the underlying data and the connections you can make across datasets.

Type 2: AI that automates tasks but has no model of your business. A newer kind of tool (e.g. Briink) which mainly uses AI to do task work. It reads a PDF, extracts an emission factor, and drafts a section of a report. This is real and worth having. The hours it saves are hours you can put to more valuable work. But pulling a number out of a document is not the same as reasoning across your whole value chain to find where an emission reduction opportunity sits. Drafting a disclosure section is not the same as knowing why you keep failing the same CDP question year after year. This is AI as a very fast assistant. It is not AI that understands the work.

Type 3: AI built on a model of sustainability itself. The third kind starts somewhere different. It starts with a structured, machine readable framework that maps, connects and represents institutional knowledge in the context of your business and its sustainability performance. It takes that business specific knowledge and connects it to the wider frameworks, concepts, regulations and real world data that define the world of sustainability. Together this forms a unique bank of knowledge for your business and sustainability, a real, deep context that drives AI outputs that are specific and valuable to your business. As this bank gains more insight and more data, through every task it is used for, it can make connections across your data, both to uncover opportunities and risks it would have been impossible to find before.

This is the only one of the three that can move you from reporting on the past to building the value ahead of you, because it is the only one that knows enough to reason rather than retrieve. That third kind is what we are building at Annwn.

Whatever you are shown in a demo, run it through these five questions. They are ordered. Each one is harder to fake than the last.

Test 1: The capture test. How much does this system really understand about your business? Does it understand your real supply chain, the data that lives in other people’s inboxes, the context that explains the numbers. A system that only knows the cleaned-up output of that process misses the interesting part. Failure mode: it knows your reported number but not how you got there.

Test 2: The connection test. Can it reason across everything at once, or only one thing at a time? Sustainability sits across the whole business (supply chains, energy, capital, regulation, reputation) which is why there is so much value in making new connections. Ask whether the tool can hold CSRD, TCFD and a wide range of your own data in the same thought. Failure mode: it summarises one disclosure beautifully but cannot draw on multiple sources.

Test 3: The why test. When something is true, can it tell you why? A system of record will tell you an energy audit happened. It won’t tell you why you used more than last year and how an efficiency programme might be worth funding. Failure mode: every answer is a record of what happened, never an account of why it happened.

Test 4: The query test. Can you ask it anything and get an answer grounded in evidence? Can it answer a real question, for example “where is our biggest unmanaged transition risk, and what would it cost to close it?”. And can it answer referencing your data, the regulations and wider context (e.g. other Transition plans, tailored sectoral advice), with the reasoning and sources shown. Failure mode: it answers questions about data already in a dropdown, and nothing else.

Test 5: The action test. When it understands something, will it act on it? Can it move from generating an insight to actions (e.g a board paper, disclosure amends, a new business case, an email to suppliers), drafted with the full context, ready for you to review and refine rather than build from nothing. Failure mode: it surfaces a good insight and then asks you to go and act on it in without helping

Because Annwn will understand the domain and your business, not just your reported numbers, it will do three things in one place. It will orchestrate the work: find the missing data, chase the people who own it, draft the disclosure, so you review rather than assemble. It will uncover the wins: patterns across your data, the regulations and the wider context that point to a cost to cut, a risk to close, a commercial opening you had not seen. And it will translate: the same finding, put in the language of the CFO, the board, an investor.

Where an ESG platform records the energy audit, Annwn will give you the case for the programme: the commercial framing, the peer benchmarks, the supplier options, the financing routes, the people to bring with you. One is a record of the past. The other is a way to build the future.

And it compounds. Every engagement teaches Annwn more: which sources you trust, the methods you use, the decisions you made and why. That knowledge usually walks out of the door when people move on. Here it stays, and it makes the next piece of work better than the last.

The questions worth asking are about what the thing actually knows. Does it know the full context of your footprint calcs? Does it know why things are true? Can you ask it anything, the way you would ask a colleague who had read every disclosure? Is the understanding based on a broader view of the business and best practice? Can it make credible strategic recommendations?

If the answer to any of those is no, you are just looking at a slightly better version of the past. The work in front of you is not the past. You need to choose a system that can look forward. If you work in sustainability and any of this resonated, we would welcome a conversation. We are building this for you, and the more we understand the work, the better we build it. Thanks for your time. Ben, Adam, Rory and Ben