Inside the pricing model: how GalimAI calculates true equity position

A companion piece, not a replacement

Our guide to below market value property in the UK covers what BMV actually means, where real discounts come from, and how to source them without getting burned. This piece assumes you have read that, or do not need the beginner version, and goes one level deeper into the mechanics: how GalimAI actually estimates an owner's equity position, the technical inputs involved, and the honest limits of what that estimate can tell you.

Step one: price-paid data against a current valuation estimate

The starting point for any equity estimate is HM Land Registry's Price Paid Data, which records what a property actually sold for and when. That figure on its own tells you almost nothing about current equity, a property bought a decade ago has likely moved in value since, sometimes substantially. The second input is a current automated valuation estimate, built from recent comparable sales in the same area, property type, and size band.

Comparing the two gives a raw uplift or, occasionally, a decline: what the property was worth then versus what a comparable sale suggests it is worth now. That comparison is the base layer of the equity estimate, but it is only the base layer, because it says nothing yet about what the owner actually owes against the property.

Step two: reading the charges register

This is where Companies House becomes relevant, for any property held through a company. The charges register shows registered charges against a company, typically including the lender's identity and the date the charge was created, and whether it has since been satisfied. That is genuinely useful signal, but it is worth being precise about its limits.

The charges register very often does not show the outstanding balance of a loan, only that a charge exists or existed. A charge created five years ago against a company that has made steady repayments since could represent a small remaining balance. The same charge against a company showing other signs of financial pressure, late filings, multiple charges stacking up, could represent something closer to the original amount still outstanding. GalimAI's approach treats charge data as directional evidence toward an equity estimate, combined with hold period and repayment-behaviour signals, not as a direct readout of what is owed. Getting this distinction right is the difference between a genuinely useful estimate and a false sense of precision.

Step three: confidence, not certainty

Every input in this process, the current valuation estimate, the inferred remaining mortgage balance, carries its own uncertainty, and those uncertainties compound. An equity position estimate built from an AVM and inferred charge data is a range with a most-likely midpoint, not a single confirmed figure. We treat it that way internally, and any figure shown to a client should carry that same honesty.

This is consistent with how we talk about our other prediction models. Our sell-timing model deliberately does not publish a headline accuracy percentage, for the same reason: an estimate presented with false precision is more misleading than one presented honestly as a range. The equity position estimate follows the same principle. It is a well-informed range, useful for prioritising which owners are worth approaching, not a number to quote back to an owner as fact.

Step four: adjusting for condition

A raw comparable-based valuation assumes an averagely maintained property. Real properties vary, and Energy Performance Certificate data is one of the few condition signals available at scale and for free. An EPC rating well below what is typical for a property's age and type suggests deferred maintenance or an outdated specification, which typically means the raw valuation needs adjusting downward before it is a realistic estimate. A recent EPC showing a strong rating, especially where the certificate reflects recent works, can support the opposite adjustment. EPC data is a proxy for condition, not a full survey, so it moves the estimate in the right direction without replacing an actual inspection.

Putting it together

The full sequence looks like this: start from the price actually paid, compare it against a current comparable-based estimate to get a raw value, adjust that raw value using EPC and other available condition signals, then separately estimate likely remaining debt using charges register data, hold period, and repayment-behaviour signals. The gap between the adjusted value and the estimated remaining debt is the equity position estimate, expressed as a range rather than a single number, because at every step of that chain, honesty about uncertainty matters more than an artificially precise headline figure.

This mirrors the approach described in our piece on why machine learning beats rule-based filtering: a hybrid of hard data, model-based estimation, and honest treatment of what the model does and does not know, rather than a black box number presented without explanation.

A hypothetical illustration

To make the mechanics concrete, consider a purely illustrative example, not a real property. Say Land Registry shows a property purchased for 180,000 pounds in 2014. A current automated valuation estimate, based on recent comparables in the same postcode and property type, suggests a range centred around 260,000 pounds today. That is the raw uplift before any adjustment. Companies House shows a charge registered against the owning company in 2015, still unsatisfied, from a mainstream buy-to-let lender, a profile consistent with a standard mortgage rather than short-term or bridging finance. Combined with an ordinary repayment history and no other distress signals, this points toward a meaningfully reduced but not fully repaid balance. An EPC certificate rated D, in line with what is typical for a property of that age, suggests no material adjustment to the raw valuation either way. Put together, the resulting equity estimate is presented as a range, not a single figure, reflecting the AVM's own uncertainty and the fact that the exact remaining loan balance is inferred rather than directly observed.

The bottom line

An equity position estimate is only as honest as its treatment of uncertainty. GalimAI builds it from price-paid data, current comparable valuations, charges register signals, and condition data from EPCs, and presents the result as a range with a most-likely estimate, not a certainty. That is a deliberately more modest claim than some competitors make, and it is the one we think is actually defensible.

Where this fits with our other technical research

This piece sits alongside two related explanations of GalimAI's technology: how the sell-timing model works, which predicts when an owner is likely to sell, and why machine learning beats rule-based filtering, which explains why we moved away from simple rule-based scoring in the first place. Together the three cover prediction, prioritisation, and pricing, three separate technical questions that get combined into a single owner score in the portal, but that deserve to be explained on their own terms rather than compressed into one page.

Common questions

Is this an exact valuation of a property?

No. It is a comparable-based estimate adjusted for available condition signals, presented as a range rather than a single confirmed figure. A formal valuation from a qualified surveyor is a different, more precise exercise.

What does the Companies House charges register actually show about equity?

It shows that a charge exists or existed against a company, typically with the lender's identity and dates, but usually not the outstanding balance. GalimAI treats this as directional evidence combined with other signals, not a direct readout of what is owed.

Does GalimAI publish a headline accuracy percentage for this estimate?

No, consistent with how we treat our other prediction models. We think a precise-sounding accuracy number invites more confidence than any estimate built this way deserves, so we describe the estimate as a range instead.

How does EPC data affect the estimate?

An EPC well below what is typical for a property's age and type suggests the raw comparable-based valuation should be adjusted downward, and a strong recent rating can support the opposite adjustment. It is a proxy for condition, not a substitute for an inspection.

This is general information about GalimAI's methodology, not a valuation, and not financial or investment advice. Any specific equity estimate is directional and should be confirmed with independent professional advice before being relied on.