Home / Platform / AI Policy Vault
The household’s entire insurance picture, including the part you cannot see today.
An insurer knows the policies it writes. It does not know what else the household holds, what it is paying, where the gaps are, or when any of it renews. The vault closes that gap in a single upload flow.
In build Motor extraction in pilot. Travel, home and health extraction in build.
From a photographed document to a structured policy
Upload
Multiple documents, multiple formats, photographed or attached. The household is guided by category rather than asked to classify insurance paperwork it does not understand.
Extract
Insurer, product, cover, sums insured, excesses, exclusions, limits, conditions, premium and renewal date. Structured, normalised, and held against the household record.
Evidence
Every extracted field carries a view-source link back to the exact place in the document it came from. Nothing the AI asserts is unsourced.
Assess
Coverage snapshot, exclusions and limits analysis, gaps against the household profile, what-if scenarios, and the Protection Score.
Two kinds of confidence, never conflated
Extraction confidence answers a narrow question: did the system read the document correctly? Interpretation confidence answers a harder one: is the conclusion drawn from that reading sound?
A system that reports one number for both is hiding the distinction that matters. A perfectly-read clause can still be ambiguously worded. The vault reports them separately, and the concierge behaves differently depending on which one is low.
When the AI is unsure
Poor scans, unsupported document types and ambiguous wording are surfaced to the household as an explicit request rather than silently guessed. Low interpretation confidence routes the item to human review before it can influence a recommendation.
The Conditions tab on policy detail exposes the wording in full. A household — or an agent — can always read the clause the conclusion rests on.
The Protection Score
A 0–1000 household measure of how well the family is actually covered, built from the vault contents against a household profile.
What it is. A structured comparison of held cover against identified household exposure, weighted by severity of the gap. It is designed to be explained — every point of the score traces back to a specific finding.
What it is not. It is not an underwriting score, not a credit proxy, and not a substitute for an insurer’s own risk assessment. It should not be used to decline or price a risk.
The commercial consequence
Once the vault is populated, the partner can see the household’s full premium spend, its gaps, its renewal calendar, and every line it holds elsewhere. Cross-sell stops being speculative and becomes specific: a named gap, a known date, a stated premium to beat.
Watch the vault read a real motor policy.
Upload, extraction with visible confidence and source evidence, assessment, and a renewal trigger.