A buyer submits an inquiry on a Tuesday afternoon. Your AI follow-up sequence fires off an immediate text, then an email, then a voicemail drop 48 hours later. But another buyer who submitted at the same time gets twelve touchpoints over the next two weeks. The difference? The second buyer's price range. Your sequence was tiered by deal size — and an attorney is now arguing that tiering correlates with a protected class characteristic in your market.
This is not a hypothetical edge case. It is the kind of fact pattern that HUD and state fair-housing agencies look for when reviewing AI-assisted brokerage operations.
How tiered sequences create disparate-treatment risk
Disparate treatment does not require intent. If your AI system applies different communication intensity, response speed, or nurture duration to leads based on a factor that correlates with race, national origin, or another protected class, the intent behind the design is largely irrelevant to a complaint examiner.
Say a team runs 40 leads a week. The automation is set to give "high-priority" leads — those above a certain price point — eight follow-up touches in the first week, while lower-price leads receive three. On paper that is a business decision. In a market where price-point geography maps tightly onto neighborhood demographics, that business decision can look like steering. The burden of proving otherwise falls on the brokerage.
The same logic applies to other segmentation variables: response language, communication channel preference, lead source, or behavioral score. Any of these can proxy for protected-class characteristics depending on how your market is structured. Before you build a branching sequence, someone needs to ask that question explicitly.
What a complaint examiner will request
If a complaint is filed, investigators can request documentation of how the AI system works. That typically means: the logic that determines which leads receive which touchpoints, the training data or rules that drive lead scoring, a full message log for the complainant and a sample of similarly situated leads, and records showing who reviewed the sequence design before deployment.
Brokerages that cannot produce clean records are at a disadvantage regardless of intent. If the sequence was built inside a third-party platform and the brokerage never exported or retained the underlying message logs, producing those records becomes difficult. The data-ownership problem compounds the compliance problem.
This is one reason Real Estate AI Group documents the logic of every sequence it builds and recommends that brokerages retain message-level logs in a system they control — not solely inside the automation platform.
Three sequence design decisions that reduce risk
The first is to audit your segmentation criteria before building. Every variable that drives a branching decision should be reviewed against the question: could this variable correlate with a protected class in this market? Price range, zip code prefix, lead source name, and even the time zone of a phone number can all carry demographic signal. This is not a reason to avoid segmentation — it is a reason to document your rationale.
The second is to treat all leads to a minimum baseline of service. A sequence that gives every lead at least the same first response within a defined window, and the same number of follow-up attempts over a defined period, is easier to defend than one that provides meaningfully different service paths from the first touch.
The third is to build in a human review step before sequences go live. That review does not need to be a legal audit of every campaign. It needs to be a documented decision by a responsible person — ideally a broker or designated compliance reviewer — confirming the segmentation criteria were examined for fair-housing implications.
None of this is legal advice, and the risk profile varies by state and market. Your broker and, for significant deployments, a fair-housing attorney should review sequence designs before launch.
Where to start
Pull up one active follow-up sequence and list every variable that determines which branch a lead takes. For each variable, write one sentence explaining why that variable was chosen and confirming it was not selected because of its potential correlation with demographic characteristics. If you cannot write that sentence for any variable without hesitation, that sequence needs a redesign conversation before it runs another lead through it.