Evidence before precision.
Enter a registration, describe the loss, and get a valuation with the comparable set, the adjustments and the limits shown in full.
The four steps
Enter the registration
One field unlocks the vehicle identity. The model decodes make, model, year, variant and fuel type automatically where a registration lookup is available for the market. Where it is not, the same fields are entered manually — the engine treats a decoded identity and a declared identity differently, because a typed variant is an assertion and a decoded one is a record.
Add condition and damage
Mileage at the point of loss, pre-loss condition grade, write-off category, primary damage area, nature of loss and location. These are not form-filling. Each one selects or excludes a different slice of comparable evidence, and each one carries a different weight depending on the vehicle.
The engine builds a comparable set
The model assembles the closest observed transactions it holds, discards the ones whose price type or age makes them unusable, and applies adjustment layers in a fixed order: identity, condition, damage, loss type, location, then timing. Every adjustment is recorded as a separate line, not folded into one number.
See the result and its limits
You get a central estimate, a range, a confidence level, the evidence behind it, each adjustment applied and the specific limitations that apply to that vehicle. If the comparable set is too thin or too stale to support a figure, the engine says so rather than producing one.
What the engine is actually doing
A valuation is a chain of decisions, and almost all of the disagreement in this market happens in the first link, not the last. These are the four principles the engine is built on.
Comparable selection comes before arithmetic
The hardest part of a damaged-vehicle valuation is not the maths, it is deciding which observations are allowed to count. A same-model car with front damage and one with flood damage are not the same market. Neither are two Cat S saloons three hundred miles apart when one sits near an export port and the other does not. The engine filters first, then adjusts, and reports how many observations survived that filter.
Adjustments are layered, ordered and visible
Condition, damage severity, loss type, location and time each move the figure independently. Applying them in a single blended multiplier hides which assumption is doing the work — and it is usually one assumption doing most of it. Layering them in a fixed order means you can see the size of each move, disagree with one, and know exactly how much of the result depends on it.
Confidence is a property of the evidence, not the model
Confidence reflects how many usable observations the engine found, how recent they are, how tightly they cluster and how close they sit to the subject vehicle. A tight cluster of recent completed sales earns high confidence. Four loosely related listings from last year do not, no matter how confident the number looks on screen.
A refusal is a valid output
Most valuation tools are built so that something always comes back. That is a commercial decision, not a methodological one. When the evidence will not support a figure for a given vehicle, the engine returns an explicit insufficient-evidence result with the reason. A missing number you can act on beats a confident number you cannot check.
What goes wrong elsewhere
These are not criticisms of any particular provider, and we do not name one. They are structural problems in how damaged-vehicle value is commonly established — the habits, shortcuts and inherited assumptions that survive because the answer is rarely checked against a completed sale.
Category codes are treated as price bands
Write-off categories record how damage was classified at the moment of a claim decision. They were never designed to describe severity, repair cost or residual value, and they are applied with real variation between assessors and insurers. Pricing straight off the letter produces a tidy table and a wrong answer. The category is an input to the model, not a shortcut past the evidence.
Asking prices get laundered into sold prices
A great deal of the price information that circulates in the trade is what somebody hoped to get, not what somebody paid. Once an asking price is copied into a spreadsheet the distinction usually disappears, and every downstream figure inherits an optimism bias nobody can see. Price type has to travel with the observation permanently, or the dataset quietly becomes a wishlist.
Percentage-of-clean-retail rules of thumb
The habit of valuing a damaged vehicle as a fixed percentage of its undamaged retail price is the most common structural error in the sector. It assumes damage scales evenly with value, that parts demand tracks whole-vehicle demand, and that the discount is stable across models and time. None of those hold, and the error is largest precisely where the money is: high-value vehicles and unusual damage.
Thin comparables produce confident nonsense
For common models with common damage, there is plenty of evidence. For a low-volume variant with an unusual loss type, there may be almost none — and that is exactly where a model will happily interpolate from whatever it can reach. Without a visible count of surviving comparables, the buyer cannot tell the two situations apart.
Photo-graded condition is treated as measurement
Pre-loss condition is usually inferred from images and a short description. That is a judgement with a known spread, not a reading off an instrument. Treating grades as precise inputs imports somebody else's subjectivity as fact. Grades belong in the model, but their uncertainty has to survive into the output range.
Repair-cost workflows are mistaken for valuation
Estimating what repair would cost and estimating what a damaged vehicle is worth are different questions with different evidence bases. Tooling built for the first is routinely used to answer the second. They diverge most sharply on older vehicles, on parts-value cases and wherever the economic outcome was never repair in the first place.
Electric vehicles sit in a data vacuum
Battery state of health, remaining warranty, real usable range and pack accessibility drive most of the value of a damaged EV, and there is very little settled transaction evidence to anchor them. Models built on combustion history quietly extend their assumptions across, which is how the same pack can be treated as an asset and as a liability in two valuations of the same car.
Staleness is invisible
Salvage demand moves with scrap and metals prices, parts availability, currency and export routes. An observation from eighteen months ago is not a weaker version of a recent one, it is a reading from a different market. Where the age of the underlying evidence is not shown, nobody can tell whether they are looking at today's market or a memory of one.
More on how observations are graded and labelled on the evidence page.