Real Estate & PropTech// definition

Distress score: a call order, not a prediction that anyone will sell

In short

A distressed property score is an ordering function over finite calling capacity: its only job is to decide which rows a team works first this week. It is ordinal rather than predictive, says nothing about what a property is worth or whether anyone intends to sell, and must never set an offer price.

Key takeaways

  • The score is ordinal: the gap between rank 1 and rank 2 carries no meaning and is not a probability.
  • Facts of record and inferences belong in separate weights, with separate decay rates per family.
  • Correlated inputs inflate rank; count the distinct source records behind a row before trusting it.
  • A low score in a county that publishes little is an unobserved row, not a low-priority one.

A distressed property score is an ordering function over finite calling capacity. If 3 people can hold 40 useful conversations a day and the list has 12,000 rows, its job is to choose which 40 go on tomorrow's sheet. Every problem with these scores comes from asking them to carry a bigger claim.

It is ordinal. The distance between rank 1 and rank 2 means nothing, and neither does the absolute number. Publish a rank or a band, never a percentage: a figure that looks like a probability gets read as one within a week, usually by somebody deciding what to offer.

Facts of record and inferences cannot share a weight

Signal familyKindWhat it evidences
Filed instruments: defaults, tax liens, judgmentsFact of recordA party took a procedural step on a stated date
Ownership and transfer historyFact of recordLength of hold and how the property was acquired
Assessed value and property attributesFact of record, modelled contentAn assessor's figure on an assessment cycle, not market value
Mailing address mismatch, vacancy indicatorsInferenceA comparison of 2 fields, with the usual false positives
Contact and household attributes from tracingInferenceA vendor's match confidence, not a fact
Signal families, and what each evidences

Store each signal with its own observation date and source, and keep inferences from outweighing facts of record. Decay by family: a filing 3 weeks old and one 3 years old are different evidence, while a long ownership hold does not decay. The commonest inference is the address comparison behind the absentee owner flag.

Four decisions this number must never be allowed to make

  1. It must not set an offer price. Price comes from the property, the comparables, the condition and the title work; a call order says nothing about what anything is worth.
  2. It must not substitute for valuation. Assessed values move on an assessment cycle and exist for taxation, so a score built on them inherits the lag and the purpose.
  3. It must not characterise a person. Motivated seller describes a state of mind inferred from records about a parcel, and the records support a call order only.
  4. It must not decide access to housing or credit. Once the output influences an application rather than a queue, different obligations attach and this design cannot carry them.

A high-ranked row needs a stopping rule when it will never become a contactable person, the problem in the owner of record that resolves to an entity. Ranking an unreachable row above a reachable one wastes the capacity the score allocates.

Geography is the route by which protected characteristics get in

Fair housing and equal credit law — the Fair Housing Act and the Equal Credit Opportunity Act in the US, the Equality Act in the UK, equivalents elsewhere — restricts decisions taken on protected characteristics, and reaches marketing and targeting as well as approvals. Confirm current requirements with counsel.

The mechanism is proxying. A score that never sees a protected characteristic can still encode one through geography: postcode, census tract, school catchment, neighbourhood indices and surname-derived features all carry demographic information. Keep geography out of the ranking except as an operational constraint — drive time, market coverage, licensing — and record what each feature is for.

Set the cut-off from capacity, then prove the order beats random

The threshold is not a quality bar. If the team can work 200 rows a week, the cut-off is wherever 200 rows sit, and it moves as staffing moves. A score tuned to a fixed number produces an empty queue or a backlog nobody works, and both look like data problems when they are capacity decisions.

Then test it. Draw about 5% of each period's calls at random from below the line as a control, because a score never compared with unranked rows cannot be shown to do anything. Judge it on rank order, not absolute conversion, which moves with the market.

One coverage caveat sits under all of it. A row can score low because nothing happened, or because the county publishes nothing readable; the score cannot tell those apart and will de-prioritise whole jurisdictions. Carry a coverage flag per source, and treat county records against licensed feeds as part of the model.

The score allocates attention. Any sentence using it to describe a property's value, or a person's situation, has left what the data supports.

Building the queue, the cut-off, the control sample and the coverage flags as one small internal system is ordinary internal tools and ops work. This page sits in deal sourcing, property data and owner outreach, part of real estate and proptech software.

Frequently asked questions

Short answers to the follow-ups this page tends to raise.

What is a distressed property score?

It is a ranking used to decide which property records a team contacts first, given finite calling capacity. It combines filed instruments, ownership history, property attributes and derived flags into an order. It is not a probability that a sale will happen, not a valuation, and not a description of the owner.

Can a distress score be used to set an offer price?

No. The score ranks who to speak to; price depends on the property, comparable sales, condition and title, none of which the ranking contains. Using it as a price input also creates a feedback loop, because the records that raised the rank get read as leverage.

How do you keep a lead score from encoding protected characteristics?

Keep geography out of the ranking except as an operational constraint, and test for proxying rather than assuming it away. Re-rank with every geographic feature removed and compare the top of both lists; if the order changes materially, those features carry demographic information.

How many signals should a distress score combine?

Fewer than most lists claim, because inputs derived from the same document are one signal rather than several. Count the distinct source records behind a row before trusting its rank, and cap rows supported by a single record. Add a signal only when it changes the order of rows you can work.

  • lead scoring
  • deal sourcing
  • data modelling
  • fair housing
// shipped work

The work behind this page

Builds from our portfolio that this page draws on.

Working on something in this space?

Tell us where you are in a sentence or two. We'll tell you honestly whether we're the right team, and what a sensible first slice of the work looks like.

Start the conversation