Don’t trust our numbers. Test yours.
Upload a CSV of your own historic sales. See how we perform against data you already know the answer to. Free, no commitment.
Per-property confidence scoring. Published PE10 and backtest metrics. API-first delivery for cascade integration. Designed for your model risk team to evaluate, not your sales team to pitch.
Or see the full sample set — strong case, weak case, and a no-value response: Sample Reports
How lenders use Gadsden Valuations
Origination screening
Pre-screen mortgage applications with instant AVM checks before instructing a full survey. Identify cases where the requested loan-to-value may be outside acceptable parameters.
Portfolio monitoring
Revalue your mortgage book periodically using batch CSV uploads or our API. Track portfolio LTV drift, identify emerging risks, and support Basel 3.1 preparation.
Retrospective revaluation
Revalue individual properties at scale for arrears management, forbearance decisions, and loss mitigation. Each valuation includes an FSD confidence band to support risk-proportionate decision-making.
What your model risk team needs to know
Accuracy metrics
Walk-forward backtested against ~295,000 held-out Land Registry transactions. We publish PE5, PE10, PE15, PE20, MdAPE, FSD, and the Fitch A–D band breakdown. Segmented by property type, price band, and region. Updated with every model release. All metrics on the accuracy page.
Per-property confidence scoring
Each property is assigned an individual Forecast Standard Deviation (FSD) based on comparable evidence, feature completeness, and model prediction uncertainty, via a granular lookup (217 cells). The FSD maps to the Fitch A–D classification bands, with a no-value response where data is insufficient. Your lending policy can map bands to risk decisions — for example AVM-only for Band B, desktop review for Band C, physical inspection for Band D.
Cascade and waterfall compatibility
Designed to operate as a node in a multi-provider valuation cascade. Instant API response, structured confidence output, and clear decline signals let you route properties through AVM → desktop → physical inspection based on FSD band and LTV.
API-first delivery
RESTful API with JSON response. Single-property and batch endpoints. Typical response time under 2 seconds. Organisation accounts with shared valuation pools for team access. API documentation.
Supporting your regulatory obligations
PRA SS1/23
Published methodology, accuracy metrics, confidence indicators, and per-valuation audit trail. Designed to support your SS1/23 vendor assessment process (the PRA’s model-risk framework for banks using internal models).
Basel 3.1
Scalable portfolio revaluation via batch CSV or API. Each valuation individually confidence-scored for risk-proportionate capital treatment.
Ratings agency methodology
We publish the PE10 and FSD metrics that rating agencies (Fitch, Moody’s) evaluate in their own AVM vendor reviews; our Fitch bands follow that agency’s published RMBS classification thresholds — an external standard, not our own grading, and no agency endorsement is implied. Methodology documented with reference to IVS 105 concepts and the EAA ESSVM 3rd Edition disclosure standards.
Backtest your own data
We publish our metrics to get attention. But you don’t need to take our word for it. Upload up to 250 recent completions — address, sale date, and sale price. We’ll value each property as of the sale date. You compare against the price you already know.
Why this test is structurally blind
You submit recent completions
Transactions from the last 6 months. Address, sale date, and the price you know.
We can’t have trained on them
Land Registry data takes months to publish. Our model trains on Land Registry data. Recent completions can’t be in our training set. Full stop.
We value as of the sale date
Each property is valued as of the date it sold — a like-for-like comparison. PE5, PE10, PE15, PE20, MdAPE, segmented by property type, region, and price band.
The comparison report
We don’t just return raw valuations — we return the analysis. Your portfolio accuracy: PE10, MdAPE, breakdown by property type, region, and price band. The document your underwriter takes to their risk committee. We generate it for you.
Zero risk
Free backtesting. No credit card required. 250 properties is enough for statistically robust metrics segmented by type and region. If we perform, you know it from your own data. If we don’t, you’ve lost nothing.
Already using an AVM?
Run Gadsden as an independent cross-check alongside your existing provider. At 20p per valuation, it’s cheap insurance against over- or under-valuations slipping through unchallenged.
When both AVMs agree
You have more confidence in the valuation. Two independent models reaching the same conclusion is stronger evidence than either one alone.
When they disagree
You’ve flagged a property worth a second look — before it becomes a problem. Divergence between two independent models is valuable information, not a nuisance.
No friction
No integration required — CSV upload or API. No contract, no minimum commitment, no procurement cycle. Your risk team can justify 20p per valuation as a validation layer without committee approval.
The cost of not checking is higher than the cost of checking.
Pricing
CSV upload or API. Organisation accounts with shared valuation pools for team access. View full pricing.
Use cases by risk level
AVM-only (low LTV, standard residential)
When the loan is well-cushioned and the property is standard, the accuracy is more than sufficient. Instant, at a fraction of the cost of a physical inspection.
AVM + surveyor (medium risk)
Two independent data points for the cost of one traditional valuation plus pennies. Agreement between AVM and surveyor means high confidence. Divergence is the signal to investigate.
Portfolio monitoring
Ongoing LTV monitoring across a mortgage book. Bulk valuations via CSV or API. Basel 3.1 portfolio revaluation at scale.
Questions to ask any AVM provider — including us
1. What benchmark is your accuracy measured against — surveyor opinions, or what properties actually sold for?
Ours: actual Land Registry sale prices. The harder test. See the accuracy page.
2. Are your results published and dated — including the segments where the model is weakest?
Ours: published on every model release, full segment tables, weak cases included. See the accuracy page.
3. Had the model already seen the test data?
Ours: our published results are measured only on sales completed after the model’s training cutoff (1 July 2025) — sales it could not have seen. Test us the same way: upload completions from after that date and your test is blind by construction.
4. What does your confidence score actually mean, and what happens when confidence is too low?
Ours: every valuation carries a Forecast Standard Deviation and a Fitch band classification. Where the evidence is too thin, we return no valuation and don’t charge. Confidence scoring · No-value policy.
5. Can I test it on my own completed loans before paying anything?
Ours: yes. Free, self-serve, today — including via CSV upload of your completions. Team accounts let your credit team share one company workspace. Start a backtest.
6. Do I get any free valuations?
Yes — 5, upon first registration.
7. Do you give credit?
Yes. Any bona fide company registering immediately has credit, limited to 1,000 valuations until your first invoice is settled. Need more? Contact us for a quick decision on a higher limit.
8. Do you offer API access?
Yes — free of charge. Auto-enabled when you buy 10+ bundles, or contact us to request it. View API documentation.
Ready to test us?
Create a free account and upload your own data. No sales calls, no demos, no commitment. The results speak for themselves.
Need more than one seat? Organisation accounts let multiple users from your company share one workspace and valuation pool.