A safer real estate lead scoring model scores the work your team owes—not a person's worth, eligibility, or probability of buying or selling. Use a short, explainable priority rubric based only on an explicit request or an open commitment. Keep every inquiry in the regular response process; a low or missing score must never suppress service.
The rubric below is a proposed operations aid, not an empirically validated predictor, a housing-law checklist, or a feature claim about Callion. Its thresholds are starting choices for a team to test locally, with qualified brokerage review before use.
First decide what the score is allowed to do
A queue-priority score can help a team see which promised action needs attention next. It should not estimate how likely a person is to transact, decide who deserves a response, determine access to housing, or replace the brokerage's normal response standard. Do not call it a “hot lead” grade: the number describes pending work, not the person.
That boundary matters because scoring can make weak assumptions look objective. The Federal Trade Commission's 2016 report on big-data analytics discusses risks from inaccuracies and bias, including possible exclusion of underserved groups. The report is broad, predates current systems, and is not a study of real-estate lead scoring; use it as a reason to test assumptions, not proof that this proposed rubric predicts harm. Read the FTC report.
HUD states that the Fair Housing Act prohibits discrimination in housing sales and related activity based on race, color, national origin, religion, sex, familial status, and disability. That is a summary of federal protections, not a complete statement of every state or local law or how a specific workflow applies. HUD's Fair Housing rights and obligations.
A proposed follow-up-priority rubric
Score only a concrete action visible in the record. Count each row at most once. These weights are examples for internal testing, not validated cutoffs.
| Recorded work signal | Example points | Evidence required in the record | What the points mean |
|---|---|---|---|
| A customer explicitly requested a next step that is still open | 1 | The request in the person's own words, the action, and the assigned owner | A follow-up task exists; it is not a prediction of intent |
| The customer gave a specific time for that action and it falls before the next routine queue review | 1 | The stated date/time and time zone, not an agent-invented deadline | The task may need earlier handling |
| The team has already missed a commitment it made | 1 | The original promised action and its due time | A service recovery task is waiting |
| Total | 0–3 | Each point must be traceable to a recorded event | An ordering hint only; never a service eligibility decision |
Example (hypothetical): A seller asks for a callback tomorrow at 10 a.m. and an agent promised to send a property-preparation guide first. If that guide has not been sent, the record could receive one point for the open requested step and one for the stated time, if it is due before the team's next queue review. The score does not say the seller is “hot,” that a listing is likely, or that other inquiries can wait beyond the team's baseline standard.
Use a simple handling rule: three points prompts a human to inspect the task first; one or two points keeps it assigned with its documented due time; zero means no rubric signal is recorded, not that the inquiry is low value. If facts are missing or unclear, mark the score unknown and route it through the ordinary queue instead of converting missing data into zero. Recalculate when the work changes; do not carry a permanent score from one conversation to the next.
What must stay outside the score
This proposal deliberately excludes protected characteristics and guesses about them, inferred demographics, names, accent, language, disability, family status, neighborhood demographics, ZIP-code proxies, perceived income, source price, or predicted willingness to pay. It also excludes tone, sentiment, and model-generated “motivation” unless a human can point to a specific customer request that creates an actual task. This is a conservative design choice, not a legal determination that any particular input is unlawful in every context.
The Department of Justice describes a Fair Housing Act case involving an algorithmic tenant-screening score: plaintiffs alleged discriminatory effects, and a court found their claims plausible when denying a motion to dismiss. That was a rental-screening case—not a brokerage follow-up queue, not a final liability finding, and not evidence that the rubric here has the same effect. It illustrates why a number used around housing-related decisions deserves careful scope and review. DOJ case summary: Louis v. SafeRent.
Test it before anyone relies on it
NIST's AI Risk Management Framework is a voluntary, general framework—not a real-estate standard. It organizes risk work around Govern, Map, Measure, and Manage, and calls for documenting context, testing, and evaluating fairness and bias where relevant. Those principles can help a brokerage ask better questions, but they do not certify this rubric or satisfy legal review. NIST AI RMF Core.
Before a pilot, write down the intended use, owner, inputs, exclusions, baseline response requirement, and who can override a score. Then test the workflow with synthetic cases: an explicit request with a due time, an inquiry with no stated timing, missing context, a missed commitment, and two records with different wording but the same requested action. Confirm that missing data does not lower service, every item still has an owner, and a person can see why a point was assigned. Do not claim these tests prove fairness or forecast sales.
During any approved pilot, review task completion, overdue commitments, overrides, and service delays. If a score cannot be explained from the underlying record, remove it from use until corrected. Consult the brokerage's qualified Fair Housing and compliance reviewer about permitted inputs, appropriate monitoring, records, and local obligations before operational deployment. Avoid collecting sensitive attributes solely to run an audit unless that collection has its own approved legal and privacy basis.
When a score is the wrong tool
If the team cannot define a fair baseline for every inquiry, fix ownership and response expectations first. If the rubric changes who receives housing-related service, uses inferred personal traits, or is marketed as predicting appointments or closings, stop and seek qualified review rather than treating this example as approval. A transparent score can still be poorly designed; transparency does not establish accuracy, fairness, compliance, or business value.
For intake questions that prepare a human conversation, see real estate lead qualification questions. For named ownership and exception handling, use the lead response plan; for defining response-time measures, see the lead response SLA guide. Explore related material in the lead qualification hub.
FAQs
Does a score of zero mean the lead is unqualified?
No. In this proposed rubric, zero means none of its three documented work signals is present. It does not measure the person's readiness or value, and the regular response process still applies. Missing information should be marked unknown, not treated as a zero.
Should a lead score decide whether an agent contacts someone?
No. This proposal only helps order outstanding tasks. Whether, when, and through which channel to contact someone must follow the brokerage's separately approved process; the score grants no permission and does not replace the baseline response standard.