Lenders
Area scoring your model risk team can defend.
Every response is stamped with the engine version that produced it. Pin a version and every call across your book returns the same numbers, deploy after deploy. Built for regulated underwriting.
Built for regulated underwriting
Four things model risk signs off.
Pin the methodology, score the book in bulk, watch it drift, and keep every percentile scoped to its own nation.
Pinned methodology
Set the engine version once, owner-only. Every key in the org honours it and every response carries it.
Bulk-score the book
Score the whole portfolio in one run. Scoring is not metered against your monthly quota.
Track drift on the book
Prove you knew on the day a tracked area moved. Sample-size gated, signed on delivery.
Country-scoped percentiles
England, Wales and Scotland use different indices. We never manufacture a cross-border rank.
Why it survives audit
Made to be defended, line by line.
Your auditors ask how every number was made and whether it still holds. The answer is in the response.
Versioned and reproducible
Pin the engine version and two calls return the same numbers, deploy after deploy.
Plan-replayable AI
Every query echoes the executed plan. Replay it and the same rows come back, no model in the loop.
Sample-size honest
Change detection is gated on transaction count. Thin evidence never fires a false alert.
Frequently asked
Questions your model risk team asks first.
Is the engine version pinnable at the org level?
Yes, owner-only. Set the engine version once and every product surface honours it, stamping it on the X-Engine-Version response header while the body reports what actually ran. Two calls under the same pin return the same numbers across deploys.
Can we replay an AI-assisted query for audit?
Yes, this is the core audit contract. Every query returns the executed plan alongside a plan_source that says whether the planner translated a natural-language question or the caller sent the plan directly. Store the plan with the decision, paste it back later, and the same deterministic executor returns the same rows. The model is never in the replay path.
How do you help with FCA and PRA model risk requirements?
We do not certify your model, that is your model risk team's job. We give you the inputs to do it well: a versioned methodology with a public registry, engine_version on every response, a plan-replayable AI seam, sample-size gating, country-scoped percentiles, and source attribution on every signal. The full methodology is public and drops into your model documentation pack.
What gets stored about our portfolios?
Just the postcode or area code and an optional label you supply to map back to your own records. No PII, no borrower data, no loan amounts. Change reports and signed webhooks carry only the areas and the moves.
Close the model-governance gap your auditors keep flagging.
Pin the engine version, bulk-score the book, track drift, and replay any AI query as a deterministic plan. The audit trail is the response.