Learning Layer
Judgment that compounds.
Lore learns how your desk actually underwrites, with machine learning trained only on your book, and turns it into models and rules you own.
01 · Observe
Every disagreement is a signal.
Approvals, edits, overrides and rejections at the review gate are captured as structured signals, pinned to the playbook version that was live when the desk decided.
02 · Learn
Patterns this desk would miss.
Learning stays inside your tenancy. It finds consistent overrides and revealed appetite in your own history. Your judgment never trains another MGA's model.
03 · Propose
A playbook change, with evidence.
Lore Suggests drafts a rule update you can read: the band that should move, how many chief underwriter overrides support it, and what shadow-replay would do to your book.
04 · Govern
Nothing goes out unless an underwriter says so.
Shadow-test against the book, name any stricter flips, then publish. A new playbook version goes live. Never a silent overnight change.
What training unlocks
Five products, earned from your own book.
The more your desk decides inside Lore, the more it gives back. Each layer trains on your firm alone. Each one proposes. None of them moves the playbook or the gate.
Playbook Suggests
Rule changes drafted from your own overrides, pre-tested against the book before you release.
“Move the telematics edge 50 → 45: 14 of 16 chief underwriter overrides agree. Shadow: 3 softened, 0 hardened.”
Rule changes drafted from your own overrides, pre-tested against the book before you release.
Action predictor
On a live case, the likely call from your own precedent, before you even open it.
“72% chance your desk approves this despite the edge. Four similar fleets already bound.”
On a live case, the likely call from your own precedent, before you even open it.
Queue prioritisation
The work most worth your attention rises to the top of the queue.
“Today, ranked by likely appetite, broker responsiveness and time to decision.”
The work most worth your attention rises to the top of the queue.
Intake memory
Extraction learns your corrections and stops repeating the same misses.
“Auto-corrected the fields your desk always fixes, like vehicle-type mapping.”
Extraction learns your corrections and stops repeating the same misses.
Outcome-linked appetite
Once real outcomes land, appetite questions raised from actual performance, not a hunch.
“Fleets you approved in this band ran 18% worse loss ratio. Harden the rule?”
Once real outcomes land, appetite questions raised from actual performance, not a hunch.