Aveni: building Britain's finance AI — and the watchdog for it

Most British AI stories point at London or the big labs. The more interesting one this month is in Edinburgh, where a University of Edinburgh spinout called Aveni is doing something unusually deliberate: building both a large language model purpose-built for UK financial services, and the layer that watches AI agents to keep them compliant. In a regulated industry, those two things together are the whole game — and Aveni has convinced the banks to pay for both.

The people

Aveni was founded in 2018 by a blend you don’t often see: career finance operators plus a serious academic. CEO Joseph Twigg spent 15 years in the investment industry, latterly as Head of Strategy at Aberdeen Standard Investments; COO Jamie Hunter came from the same firm. The technical half is Professor Lexi Birch, who holds a personal chair in natural-language processing in Edinburgh’s School of Informatics and serves as chief scientist. That pairing — people who know exactly where compliance hurts in a wealth manager, alongside one of the UK’s stronger NLP research groups — is the company’s real moat.

The product

Aveni didn’t start with a model; it started with products that solve expensive, boring problems for regulated firms. Aveni Detect monitors adviser–client conversations for compliance and conduct risk; Aveni Assist handles the productivity side — summarising calls, drafting the admin, freeing advisers from note-taking. Both are deployed across UK banks, wealth managers and advice firms. That order matters: Aveni earned distribution and real financial-services data before training a model, which is the opposite of the usual model-first hype.

In 2026 it built on that foundation with a suite aimed squarely at the new problem — keeping AI agents in line. Its Agent Assure, Agent Approve and Unified Assurance Platform extend the same compliance DNA from monitoring humans to monitoring autonomous agents, which is where the £12M is pointed.

The technology

The headline build is FinLLM, developed by Aveni Labs as a large language model trained specifically on UK financial-services data, rather than a general model prompted to behave. The bet is that in a domain this regulated and jargon-dense, a vertical model — one that understands suitability, conduct rules and the language of advice — beats a bigger generalist. The FinLLM team is based at the Edinburgh Futures Institute, tightening the university link.

The cleverer move is the second layer. As firms rush to deploy customer-facing AI agents, Aveni has pivoted its compliance heritage into agent assurance — monitoring what an AI agent says and does in real time, against the rules. It demonstrated real-time agent assurance in the FCA’s inaugural “Supercharged Sandbox” — positioning itself not as another agent, but as the thing that lets a bank trust one.

The money

The funding tells its own story. A £11M Series A in 2024 was backed by Puma Private Equity alongside Lloyds Banking Group and Nationwide — and in June 2026 Aveni raised a further £12M, led by PXN Ventures with Puma, Lloyds, Nationwide and Scottish Enterprise returning. The same banks are co-developing FinLLM. When your customers, your investors and your training-data partners are the same institutions, you have something rarer than capital: a distribution and credibility loop most AI startups never get.

What we can learn

A few things travel well beyond Edinburgh:

• In regulated industries, the assurance layer is the product. Everyone can bolt on an agent; far fewer can prove to a regulator that the agent behaved. Aveni is selling trust, not just capability — which rhymes with our own view that the durable advantage in AI is verifiability, not raw model size.
• Vertical beats general where the language is specialised. FinLLM is a bet that a smaller, domain-trained model is more useful — and more defensible — than prompting a frontier generalist. For UK SMEs, the read-across is that the most valuable models for your sector may not be the biggest ones.
• University spinouts are a genuine British strength. Aveni is what happens when a real NLP research group meets operators who know the pain. It’s a template the UK produces well and undersells.

Aveni isn’t a household name, and it doesn’t need to be. It’s quietly building the unglamorous infrastructure — a domain model plus a compliance brain — that decides whether agentic AI is allowed anywhere near your mortgage or your pension. That it’s being built in Britain, out of a university, funded by British banks, is exactly the kind of story this section exists to tell.