Seller leads can be identified inside an existing contact database by matching each contact against property data such as equity position, loan status, and length of ownership, then ranking the matches by how closely they resemble a homeowner who is ready to move. The result is a scored list drawn entirely from people already known to the agent, rather than a cold list bought from elsewhere.

What Is Seller Lead Scoring?

Seller lead scoring is the practice of ranking known contacts by their statistical likelihood of selling, using property and financial signals rather than expressed intent. It differs from ordinary lead capture in that nobody in the list has raised a hand: the ranking is inferred from circumstances, so every name on it is a hypothesis rather than an enquiry.

The workflow in this article combines two exports, a contact list from your CRM and property data from a source such as RPR, the Realtors Property Resource available to National Association of Realtors members. An AI agent, in my case Dubb Agent, matches one against the other, scores the result, and drafts outreach that references the specific signals behind each score.

This One Needs a Sphere

Let me put the limitation first, because it decides whether the rest of this is useful to you.

This method runs on an existing database. If you have a CRM with contacts in it, people you have worked with, met, or collected over the years, this will work. If you are starting from nothing, it will not, because there is nothing to score.

If that is where you are, the cold equivalent is a different workflow entirely: pulling homeowners who have already declared intent, meaning for sale by owner listings and expired listings. We cover that in the post on generating real estate leads with AI, and it is the right place to start if your database is thin.

Everything below assumes you have a sphere and have not done much with it lately.

Why Your Database Beats a Cold List

Database activation is not a new idea in real estate. What is new is being able to rank seller leads by signal rather than working the list alphabetically.

Most agents work their database on a schedule. Everyone gets the market update, everyone gets the holiday card, and the ones who happen to be thinking about moving occasionally reply. That is a broadcast strategy and it treats a contact who bought eleven years ago with eighty per cent equity exactly the same as one who closed last spring.

Scoring inverts that. You are not contacting everyone less often, you are contacting a small number of people who look like movers, with a reason to be calling them specifically. And because they are in your database, you have something no cold list can give you: they already know who you are.

That is the actual advantage here. The signals are available to anyone. The relationship is not.

What You Need Before You Start

Two exports.

Your CRM contacts. Names and property addresses at minimum. If your CRM carries the date you last spoke, bring that too, because recency changes how you open.

Property data for those addresses. RPR is the usual source for agents, since it is a National Association of Realtors member resource and carries property, mortgage, and ownership detail. Your MLS may offer something comparable. What you are after is equity position, loan status, ownership length, and last sale price.

The matching is the work, and it is the part that used to make this impractical. Doing it by hand for four hundred contacts is an afternoon nobody has. Doing it for four thousand is why most agents never did it at all.

Check the Terms Before You Export Anything

This section is not in the video and I am putting it before the workflow rather than after it, because it gates everything that follows.

You are about to take property and mortgage information about named private individuals and put it into an AI tool. Three things to settle first.

The data source's terms. RPR, your MLS, and your association each set rules about how their data may be exported, stored, and used, and those rules are not uniform. Bulk export and use in third-party tools is exactly the category most likely to be restricted. Read the terms of use for whichever source you are pulling from, and if it is ambiguous, ask your broker or association rather than guessing.

Where the AI tool sends the data. A general-purpose consumer AI tool is a different proposition from one your brokerage has approved. Ask what happens to what you paste in, whether it is retained, and whether it trains anything. If your brokerage has an AI policy, this is what it is for.

What you actually need to send. You very often do not need names attached to the property data to get a ranking. Scoring addresses and rejoining names locally afterwards gets you the same list with less exposure. It is slightly more effort and it is the version I would default to.

None of this is a reason not to do the work. It is a reason to spend ten minutes on it once, rather than discovering the constraint after you have exported four thousand records.

The Signals That Predict a Sale

The prompt is only as good as the signals you ask it to weigh. These are the ones that carry real information, and what each one does not tell you.

Notice that the strongest signals are all financial and none of them is about intent. That is the central limitation of this method and the reason the outreach matters as much as the scoring.

The Prompt

The video points people at a prompt sent out through the comments. Here is a version you can use immediately, written to take both exports at once.

Two instructions in there are doing more work than the rest. The confidence flag stops a score built on a ten-year-old loan record from looking as solid as one built on current data. And the final line matters more than it should have to: an AI asked for a complete table will fill gaps rather than leave them, and a fabricated equity figure in a list you are about to act on is worse than no list at all.

Reading the Score Without Believing It

You will get a ranked list of seller leads. Treat it as a reading order, not a verdict.

A score of 94 does not mean that person is selling. It means their numbers look like the numbers of people who sell. The model has no idea that they just renovated the kitchen, or that their daughter starts school next year, or that they told you at a barbecue they are never moving.

So I read the top of the list with my own memory switched on. Roughly a fifth of any list I have looked at gets struck immediately on things I happen to know, and that is a feature of having a sphere rather than a cold list. Nobody can strike names off a purchased list.

The other thing worth doing is reading the bottom of the list. Contacts scoring low on equity are not opportunities today, but they are the people for whom a valuation might be genuinely surprising after a few years of appreciation. That is a different conversation, not an absent one.

Turning the List Into Outreach

A list of scored seller leads is not the point. It is the input to the part that takes time.

What makes this worth doing is that the same signals that produced the score are the raw material for the message. An AI agent can draft personalised outreach for every name on the list, referencing the specific equity position and ownership length that put them there, so each message is about that house rather than about the market generally.

That is the difference between a mail merge and personalisation. A mail merge changes the name. This changes the reason.

Where video earns its place is immediately after. A message saying you have been looking at their street is a claim; a thirty-second video with their block on screen is evidence, and it is much harder to file mentally under advertising. The scoring tells you who to record for, which is the question that usually stops people recording at all.

What to Say to a High-Equity Owner

The opening is where most of these campaigns fail, so it is worth being specific.

Do not lead with an invitation to sell. Someone who has not been thinking about moving experiences that as a stranger telling them what to do with their home, and it is the fastest way to lose a warm contact.

Lead with the information instead. Most owners carrying serious equity do not know the number, because they are anchored to what they paid and have not updated it. Telling someone what their position looks like now is useful whether or not they ever sell, and it is a legitimate reason for an agent to be in touch.

Then leave the decision entirely with them. Something close to: here is where your equity sits, here is what that would make possible if you ever wanted it, no action needed and happy to run the numbers properly if you are curious. That is a message a person can receive without feeling worked on.

Two things to keep out of it. Any figure you cannot stand behind, and any implication that the estimate is an appraisal. It is a model output from property data, and saying so plainly costs you nothing and protects the relationship if the real number lands differently.

Before You Send Anything

The workflow ends with an agent drafting and sending outreach at volume, which is the point at which a few rules start applying that did not apply to the scoring.

Calling. Check Do Not Call status before you dial. An existing business relationship changes what is permitted, but it is time-limited and it does not cover everyone who happens to be in your CRM.

Texting. Consent rules for SMS are separate from and generally stricter than those for calls, and having someone's mobile number in your database is not the same as having permission to text them marketing. Check before you send, not after.

Email. Bulk sending needs a working unsubscribe and an accurate sender identity, and your brokerage may have rules of its own about what goes out under its name.

Everywhere. Your state licensing rules and your brokerage policy sit on top of all of it, and advertising rules differ meaningfully between states.

Rules vary by jurisdiction and they change. Confirm what applies where you practise rather than relying on a blog post, including this one.

Seller Lead Tools Compared

Predictive seller identification is an established category, and the products in it differ mostly in whose list they score.

The honest comparison is that the bought-territory products solve a different problem. They give you volume among strangers. This approach gives you a much shorter list among people who already know your name, which converts differently and costs nothing but the time to run it. If your database is small, the territory products are the better answer.

Common Mistakes and Troubleshooting

Running it without a database. There is nothing to score. Start with declared-intent leads instead.

Exporting before checking the terms. The data source's rules and your brokerage's AI policy both apply, and finding out afterwards is expensive.

Believing the score. It is a reading order. Strike the names you know better than the data does.

Letting the AI fill gaps. Instruct it to report missing fields as missing. A confident invented equity figure is worse than an obvious hole.

Leading with an invitation to sell. Lead with the information. The ask comes later or not at all.

Calling the estimate an appraisal. It is a model output from property data. Say so.

Sending to everyone at once. Run twenty-five, read what comes back, then adjust the message before you touch the rest of the list.

Skipping the compliance step because it is your own database. Familiarity is not consent, particularly for text messages.

Best Practices I Actually Follow

These are the habits that survived a few rounds of doing this, and none of them are specific to one tool.

Score addresses, rejoin names locally. Same list, less exposure, ten extra minutes.

Ask for the signals, not just the number. A score you cannot explain is a score you cannot write an opening from.

Work the top twenty-five before generating more. The constraint is your follow-through, not the list length.

Record video for the top ten. Personalisation is a claim in text and evidence on camera.

Re-run it quarterly rather than monthly. Equity and tenure move slowly. Monthly re-runs produce mostly the same names and train you to ignore them.

Keep a note of who replied and why. After two rounds you will know which signal actually predicts a conversation in your market, which is worth more than the model's weighting.

Proof: Why This Actually Works

The mechanism behind scored seller leads is not mysterious. Equity position and length of ownership are correlated with moving, and those fields are available for most properties, so ranking a known list by them puts likelier movers at the top. What makes it worth doing is that the list is warm.

Two patterns hold consistently across the agents we work with. The first concerns list length. Agents who work a short scored list finish it, and agents who generate several hundred names work the first thirty and abandon the rest, which is the same outcome as never having generated them. The binding constraint is follow-through rather than data.

The second concerns the opening. Outreach that leads with a specific figure about that specific property gets replies at a noticeably different rate from outreach that leads with an offer to help them sell, even when the underlying list is identical. The data selects the audience; the framing decides whether they answer.

What I take from it is that the scoring is the cheap half. The expensive half is still saying something worth reading to twenty-five people who already know you.

Frequently Asked Questions

How do I find seller leads in my existing database?

Export your contacts with their property addresses, then pull property data for those same addresses from a source such as RPR or your MLS. An AI tool matches the two on address and ranks each contact by equity position, length of ownership, loan status, and the gap between purchase price and current estimated value.

The output is a scored list of people you already know, ordered by how closely their numbers resemble those of homeowners who move. Check your data source's terms of use and your brokerage's AI policy before exporting anything.

What data predicts that a homeowner will sell?

Estimated equity as a share of value is the strongest single signal, followed by years since the last sale and the gap between purchase price and current estimate. Loan rate relative to the current market and owner occupancy add useful weight.

None of these measures intent. They measure capacity and circumstance, which is why the list is a set of hypotheses rather than a set of leads, and why the wording of the first message matters as much as the ranking.

Do I need RPR to do this?

You need property data from somewhere. RPR is the common choice for agents because it is a National Association of Realtors member resource carrying property, ownership, and mortgage detail, but your MLS may provide comparable fields.

What matters is that you can get estimated value, equity or loan balance, last sale date and price, and occupancy status for the addresses in your database. Check what the source permits you to export and use in third-party tools before you start.

Is it safe to put client data into an AI tool?

Settle three things first: what the data source's terms allow, what the AI tool does with what you paste in and whether it is retained, and whether your brokerage has an AI policy that applies.

In most cases you can score addresses without attaching names, then rejoin the names locally afterwards. You get the same ranked list with meaningfully less exposure, and it costs about ten extra minutes.

What should I say to a homeowner with high equity?

Lead with the information rather than an invitation to sell. Most owners carrying serious equity are anchored to what they paid and do not know their current position, so telling them is useful whether or not they ever move.

Then leave the decision with them: here is where your equity sits, here is what it would make possible, no action needed. Avoid any figure you cannot stand behind, and never imply that an estimate is an appraisal.

Can I text or call everyone on the scored list?

Not automatically. Check Do Not Call status before dialling, since an existing business relationship changes what is permitted but is time-limited and does not cover everyone in a CRM. Consent rules for marketing text messages are separate and generally stricter, and having someone's mobile number is not permission to text them.

Your state licensing rules and brokerage policy apply on top. Rules vary by jurisdiction and change, so confirm what applies where you practise.

The List You Already Own

The sequence is short. Export your contacts and the property data for their addresses, settle the terms and policy questions before you move anything, run the prompt, read the top twenty-five with your own memory switched on, and open with the number rather than the ask.

What makes it worth the afternoon is that this is not prospecting. Every name is someone who already knows you, ranked by a signal you could never have seen by scrolling your CRM alphabetically. Dubb Agent is where I run the matching and the outreach, because the drafting and sending sit in the same place as the scoring, and the gap between a list and a sent message is where most campaigns die.

If you change one thing after reading this, score the database you already have before buying access to anyone else's. The warmest seller leads you will get this year are probably already in your phone.