Build an AI Employee that Prioritizes Past Clients to Reconnect With

The "sphere prioritizer" that ranks your database.

Jul 9, 2026 · 5 min read

Somewhere in your CRM is your next closing.

You already met them. You already earned their trust. You already handed them keys and watched them cry a little in the driveway. Then you lost track of them.

The data on this is not subtle. Per NAR's 2024 Profile of Home Buyers and Sellers, 88% of buyers say they would use their agent again or recommend them. Only 21% actually use that same agent the next time they buy.

That gap is not a loyalty problem. It is a follow-up problem.

People forget the name of an agent who goes quiet for three years. Then they hire whoever showed up in their feed last month.

NAR's 2026 Member Profile puts weight behind this. For the typical agent, 28% of business comes from repeat clients and another 22% comes from past-client referrals. About half your pipeline is supposed to come from people who already know you.

So the highest-value list in your business is one you already own. The problem is reading it.

The manual version

Say you sit down to rank that list by hand this week.

You export your contacts. You scroll every row and try to hold it all in your head: who bought in 2019, who referred their sister, who mentioned upsizing when the baby came, who you have not spoken to since the closing.

A few hundred rows in, they blur. You end up working whoever you happen to remember, which is not the same as working whoever matters most. Most agents give up and close the tab.

What you are about to build

You can build an AI employee that does the reading for you. It scores every contact and ranks them by who to reconnect with first, with one line of reasoning for each.

That ranked list is the whole job for this build. Pulling in local market data and drafting the actual reconnection message come next in this series. First you need to know who to start with.

Set aside about an hour for your first pass. Most of that hour is pulling a clean export and running a batch or two until the ranking looks right. The AI part is the fast part.

Before it ranks anyone, it weighs four things:

1. How long since you last spoke
2. How strong the original relationship was
3. Any note that hints at a move coming
4. Whether they are near a typical move cycle

The full build, the exact system prompt, and the Fair Housing guardrail that keeps it inside the lines are below.

The build

Step 1. Export your contacts

Pull a CSV out of your CRM. You want one row per contact and, at minimum, these columns:

Name, Last Contact Date, Relationship Type, Transaction History, Notes

Do not clean it obsessively. The point is that the AI employee does the reading. If a column is missing, that is fine, the prompt handles gaps.

One rule before you go further: strip out any column that records race, religion, national origin, family status, disability, or anything close. You are not allowed to prioritize on those, and you do not want that data anywhere near this workflow.

Step 2. Understand the scoring logic

Your employee ranks contacts by reconnection priority, highest first. It weighs four signals:

Recency. Long gaps lower the odds of a warm reply. A very long gap with a strong past relationship can still rank high, framed as a reconnect rather than a check-in.

Relationship strength. A past client outranks a referral source, which outranks a lead who never closed, which outranks a cold contact.

Move signals in the notes. Growing family, job change, lease ending, "thinking about selling," equity questions. This is the strongest single indicator, and it only works if you write notes. Start writing notes.

Transaction timing. A buyer from 2019 is closer to a typical move cycle than a buyer from last spring.

Step 3. The system prompt

This is the employee. Paste it into Claude or ChatGPT, then paste your CSV rows underneath it. Copy it exactly.

You are a real estate database analyst helping a licensed agent decide who in their sphere to reconnect with first. You will receive a list of contacts in CSV form.

For each contact, produce:

1. A priority score from 1 to 100. Higher means reconnect sooner.

2. A one-line reason for the score, referencing a real detail from the row when one exists.

Score using ONLY these signals, in this order of weight:

- Time since last contact. Longer gaps lower the odds of a warm reply, but a very long gap paired with a strong past relationship can still rank high as a reconnect.

- Strength of the prior relationship: past transaction > referral source > lead who never closed > cold contact.

- Notes that suggest a likely move: growing family, job change, downsizing, lease ending, "thinking about selling," equity questions.

- Transaction timing: a past buyer approaching a typical move cycle ranks higher.

Do not write outreach messages in this step. Rank only.

Compliance rules. Do not break these:

- Do not use, infer, or reference race, color, religion, national origin, sex, familial status, disability, sexual orientation, or any protected characteristic in scoring.

- Do not steer a contact toward or away from a neighborhood based on demographics.

- If a row contains protected-class information, ignore it and never repeat it.

Return the result as a table sorted by priority score, highest first:

Name | Score | Reason

This is a working list for the agent to review. Rank only, no messages.

Step 4. Run it

Two paths. Start with the first.

Path A, do it this week. Paste the prompt, then paste 50 to 100 rows of your CSV, and let it return the table. Copy the table into a spreadsheet. Repeat for the next batch. No accounts, no keys, no cost beyond the chat subscription you already have. The trade-off is that you run each batch by hand.

Path B, make it recurring. You can wire this into a Make.com scenario so it re-ranks on a schedule, reading from a Google Sheet and writing the results back. That version needs a Make account, an API key that bills per use, and real setup and testing time. If Make and API keys are new to you, treat it as an afternoon project, not a coffee-break one. The full automation walkthrough is its own piece later in this series.

For now, Path A gives you the list. That is what you came for.

Step 5. Read your list, top down

Your employee ranks. You decide.

Read the top 20. Sanity-check them against what you know, and move anyone the data misjudged. That order is your reconnection queue for the coming weeks.

What you do with the queue, how you warm the channel on social and what you actually say, is the next stage of the series. For now you have gone from a database you avoid to a list you can act on.

What this is worth

Run your own numbers. If your average sale is $400,000 and you earn 2.5%, one reconnected past client is roughly $10,000 in commission.

One closing pays for years of anything you would use to surface that list. It also pays back the afternoons you used to lose scrolling a database you never finished.

You just built the first piece of your own Engagement Manager: a prioritizer that turns a dead database into a ranked reconnection queue, inside Fair Housing standards.

Next in this series, you teach it to pull in each contact's local market data, so the reason to reach out gets specific. After that, you build the layer that drafts the reconnection message itself.

If you would rather not rebuild and re-run this every month, the whole job lives on Avenue Growth as Engagement Manager. It keeps your list ranked and current on its own, and every stage of this series comes already built.

You can build it yourself, or hire it. Both roads end with your database working for you.