Skip to content
01Services

Everything we build sits on trusted data.

AI projects often stall because the data under them can't be trusted. We work from the bottom up, and each layer only goes in once the one under it holds.

Book a data mapWhere it starts
Above ground · what your team uses
  1. 05AI that takes stepsLater
  2. 04AI on your documents
  3. 03Models in production
Below ground · the part nobody sees
  1. 02Enriched data
  2. 01Trusted data
The data map starts below ground, and tells you which layer the real problem is in.
02

Trusted data

Policy, claims and bordereaux data that reconciles, with keys that join across systems and checks that run every day.

At an MGA, that looks like
  • Recognising the same insured across policy admin, claims and the broker's bordereau.
  • Bordereaux that reconcile before they're sent, not after a capacity provider queries them.
  • A pricing dataset that rebuilds itself overnight instead of by hand each month.
Measured byRecord rejection rate and bordereaux turnaround.

Expect most of the work to be here.

03

Enriched data

We look for outside data that would improve your rating and test it against your own claims history. If it earns its place, it goes into your records with the source marked on every field.

At an MGA, that looks like
  • Trying candidate rating data against your own claims history before anyone pays for it.
  • Insureds matched to their Companies House record: legal name, registered address, status and filings.
  • Addresses matched to a single property reference, with public flood risk data attached.
  • Vehicle registrations matched to make, model, and tax and MOT status.
Measured byHow much each new source improves the rating model, and the match rate behind it.
04

Models in production

Pricing and claims models that run on that data every day, with monitoring that tells you when one starts to drift.

At an MGA, that looks like
  • A new rating model live within days of sign-off.
  • Claims likelihood scored on every new policy as it's written.
Measured byDays to reprice.
05

AI on your documents

Language models reading what your team reads today: broker submissions, slips and claims notes.

At an MGA, that looks like
  • A broker submission turned into a structured risk record for the underwriter to check.
  • Pricing-change notes drafted for review rather than written from scratch.
Measured byHours freed and error rates, against a baseline.
06

AI that takes steps

Where this leads

Systems that handle a whole task, start to finish, with an underwriter signing off. This comes last, once everything below it holds up.

At an MGA, that looks like
  • Working through bordereaux exceptions and sending only the hard ones to a person.
Measured byThe same numbers as the layer below.

Not sure which layer you're stuck on?

That's what the data map is for. Two weeks, for a fixed fee agreed before we start.

Book a data map