We had a new agent join the firm. They brought with them around 1,800 contacts, curated over a few years from lead programs they had been paying for. That is a decent list of past clients, leads that never resulted in anything, friends, family, etc. It is not, by itself, a working business. They knew they had to start somewhere. That is the crux of all of this.
The database was never the problem. Reading it was.
The Myth
The current operating model is that a database goes cold and has to be warmed. More touches. Better templates. A tighter cadence. Spray and pray, which has been the industry’s mentality for a long time. Add a newsletter, add a video, add a drip.
This is the wrong diagnosis, and the tools built on it have made the situation worse rather than better. Volume tools let a real estate agent contact everyone badly instead of a few people well. A market update sent to 1,800 people is not eighteen hundred touches. It is one piece of content and zero conversations. The agent knows this, which is why many feel a newsletter is more of an obligation rather than a strategy.
Reality Check
The constraint in this business is not contacts. It is flow.
Ninja Selling teaches this. Fifty live interviews a week, face to face or voice to voice. Two live real estate reviews a week. Daily focus on a hot list, people who want to transact in the next 90 days, and a warm list, about 20 percent of the sphere. For a 200-household database, that warm list is about 40 names. That is the list you stay in flow with. Not the 1,800.
We sat with the new agent and ran the list through Claude for Business. Last update. Urgency to act. A value on each relationship. Not a draft message. A ranking. 1,800 became approximately 75 names. That is a hot and warm list. That is where you start. This is a working list. The CRM’s job, once you have it, is to keep the context of those relationships so you do not lose the value you just found.
A system that records what happened is an archive. A system that tells you what changed is something else entirely.
Luis Poggi, who spent a decade on product and engineering at Zillow and now builds in this category, put it plainly on Rajeev Sajja’s Real Estate AI Flash: “a real estate agent cannot do long-term nurture by hand.” The conversation happens in a living room or a driveway, and then it dies in the car. The CRM never gets the two-bedroom, the yard they hated, the Saturday at ten. Before the next showing, nobody can say “last time you mentioned this.” Whatever you think about different prop-tech solutions - this much is true: The database did not go cold. It never got the conversation.
Framework
Three classes of signal matter, and they are not equally reliable. Treating them as interchangeable inputs to one score is where most of this goes wrong.
Life-event signals. Job changes, relocations, a growing family, ownership tenure crossing the local median, equity. These are the most predictive and the most inferential. They tell you something changed in a person’s life that historically precedes a move. They do not tell you that this person is moving. The gap between those two statements is where the true mission lives.
Engagement signals. What the contact actually did inside your own systems. Opened your monthly market update email three months running. Clicked through to two listings in a town they do not live in. Replied to something. These are behavior, not inference. The strongest version is what they told you last time, and it usually dies in the parking lot.
Decay signals. Time since the last real contact, weighted by relationship depth. This is the cheapest class to compute, requires no vendor, and is almost universally ignored. The past client from 2019 who referred two people and has not heard from you in fourteen months is not a prospect in any model. They are something more valuable and more fragile, and they are being lost to nothing more dramatic than the passage of time.
The operating rule across all three: the model ranks, the agent decides, and the output is not a message. The output is a name and a reason. A system that produces a draft message has automated the wrong end of the problem.
The Surveillance Line
There is a version of this that goes badly, and it is worth naming.
The newest products in this category work on public social and professional signals across an agent’s existing sphere, turning posts about relocations, job changes, and home projects into a ranked daily queue. Sidekick Radar is a clean example of the design constraint. The question it asks is whether a signal creates a useful reason to help right now, and whether the agent can reach out without making the client feel watched.
The fact that the category has to say that out loud tells you the objection is real.
Can the agent state the reason for reaching out in one true sentence, out loud, to the person they are reaching out to?
“I saw you posted about the new job. Congratulations. Are you thinking about the commute?” passes. It is true, it is public, and the client would not be surprised to learn the agent saw it.
“Our system flagged you as a high-probability seller” does not pass, and neither does the polite version that hides the same thing behind a market update. If the reason cannot survive being said out loud, it is not insight. It is noise.
The Constraint That Decides the Architecture
None of this is theoretical once the system starts producing outbound contact.
I am not going to recap TCPA, FCC, or state telemarketing rules here. They are moving too fast, they vary by channel and by state, and a paragraph in a newsletter is not diligence. If you are scoring lists and then calling, texting, or letting a platform send on an agent’s behalf, you need to be up to speed.
A scored list of strangers in a farm area is a list of people with whom you have no relationship. Every channel you might use to reach them is either expensive, tightly governed, or both.
A scored list drawn from your own sphere is a list of past clients, past inquiries, and people who already know the agent’s name. The relationship is already there. That does not make outbound automatic. It does mean you are not manufacturing a relationship from zero in the hardest channel available.
A scored list that feeds automated outbound is a different object than a scored list an agent works by hand. If a brokerage cannot answer the question “does this queue result in a person dialing, or a platform sending,” it does not have a design. It has a process that needs to be looked at.
Actionable Insight
Three tests, in order, before evaluating a single vendor in this category.
The flow test. Ask your system to name this week’s working list. Ninja’s warm list at 200 households is about 40 names. We got to 75. Not 1,800. Not a segment. Names, ranked, each with a stated reason. If it cannot produce that, it is an archive.
The one true sentence test. Same test as above, applied to every signal class. If it cannot be said out loud, do not surface it.
The channel test. For each signal, decide in advance what it produces. A call. A note. Nothing but a flag on the record. Then confirm the system enforces that decision rather than leaving it to the agent at nine o’clock at night with a dialer open.
The relationship is not the thing being automated here, and any vendor telling you otherwise is selling you a liability. What is being automated is the noticing. An agent with 1,800 contacts cannot hold 1,800 people in their head, and never could. They have always been choosing 75. They have just been choosing them by whoever happened to cross their mind that week.
The opportunity is not to replace an agent’s judgment. It is to stop making them exercise it blind.