Delegated Authority Has a Second Underwriter Now: Your Data

Every MGA capability statement now has a line about data and AI. It's become a homogenous claim, said so often it's stopped meaning very much. Underneath it, something more specific is actually happening in the UK MGA market and it has nothing to do with how sophisticated anyone's AI roadmap is.

Delegated authority is no longer a niche distribution channel. Somewhere between 40% and 45% of all Lloyd's GWP now flows through it, and the FCA has a dedicated review of MGA and coverholder governance due early 2027. At the same time, the market is soft: rate alone isn't going to carry growth for anyone. Put those two facts together and you get a straightforward but underappreciated shift: capacity providers and acquirers are no longer treating an MGA's data as background information. They're pricing it.

Two very different claims, collapsed into one

Most of the market's "data and AI" energy goes into data as a growth capability: faster triage, AI-assisted underwriting, fewer hours lost to re-keying. It's a real and worthwhile pursuit. It's also, per Capgemini's 2026 World Property & Casualty Insurance Report, where 60% of insurers are stuck at exploration or proof-of-concept, with 42% not even tracking whether the investment is working!

The claim capacity providers and acquirers are quietly acting on is different: data as an underwritten asset. Once authority is delegated, an MGA's data is the only real window a capacity provider has into how that risk is actually being managed day to day. Profit commission is calculated directly from MGA-reported numbers, so bad data doesn't just obscure a problem, it can misstate who gets paid what. And on the M&A side, buyers are now explicit about it: audits of tech and data happen before the deal, not after and the diligence outcome moves the price. A modern, clean system on the buyer's platform can add 0.1 to 0.2x to the multiple. A legacy system with no API access can cost 0.2 to 0.4x. Documented, working AI deployment adds a further 0.1 to 0.15x.

These are not the same claim and treating them as interchangeable is where the gap between ambition and evidence actually lives.

Why the distinction matters in practice

An MGA can have genuinely sound underwriting judgement and still lose on capacity terms or get discounted at exit, because what's being priced isn't the loss ratio in isolation. It's whether that loss ratio can be verified quickly, by someone who isn't you.

Chaucer and Ceto AI's Lloyd's marine MGA is a useful illustration, and not for the reason it usually gets cited. The interesting part isn't that they use AI. It's that they built a data pipeline, continuous vessel machinery and performance data, that most competitors structurally don't have. The AI is downstream of a data architecture decision, not a substitute for one. Good AI sitting on a mediocre data foundation is a feature. A data source competitors can't replicate is a moat.

A test worth running

If you lead underwriting, claims or data at an MGA, the useful question isn't whether you have an AI strategy. It's narrower than that:

• Could a capacity provider trace a bordereaux entry back to the underlying decision this week, without you compiling anything specially for them?
• Could you show accumulation drift on a live book today, not at quarter-end?
• If a buyer's diligence team asked for five years of granular, line-level loss data tomorrow, could you hand it over? Or would that become a project?

If the honest answer to any of those is "we'd need some time," that's not a data problem waiting to be solved eventually. It's the thing currently being priced into your capacity terms and your valuation, whether or not anyone has said so directly.

The MGAs doing well out of this market aren't the ones with the most AI. They're the ones whose data was built to be underwritten from day one, not defended after the fact.