How to clean a real estate CRM in a weekend

Most agents we speak to already pay for a CRM. It has three versions of the same person in it, tags from a system they left in 2023, and a pipeline where forty deals sit in a stage called "nurture" that nobody has defined. So they work from memory and their phone, and the database quietly rots while the subscription renews.

You can fix most of that in a weekend if you do it in the right order. The order matters more than the tooling.

Before anything: export everything

Full CSV export of contacts, notes and deals, saved somewhere that is not the CRM. Do this even if you are only planning to look. Every irreversible mistake in this process happens in the first hour, usually to someone who was "just tidying up".

Step one: duplicates

This is the highest-value hour of the weekend, because everything downstream depends on it. Automations sent to a duplicate look careless to the client and inflate every number you look at afterwards.

Decide the survivorship rule before you start merging, and write it down:

• Match on email first. Emails are near-unique and rarely mistyped in a way that collides.
• Then match on phone, stripped to digits. "+1 (512) 555-0134" and "5125550134" are the same person and most CRMs will not spot it.
• Name matching last, and only with a second field agreeing. There are a lot of real Dave Smiths.
• When merging, keep the oldest created date and the newest activity. The oldest date is where the relationship actually started.

Merge, do not delete. Every serious CRM has a merge that preserves notes from both records. If yours does not, that is genuinely a reason to move.

Step two: validate what is left

Bounced emails and dead numbers are not neutral. A sequence firing into a 30% bounce rate damages your sending reputation, which means your real emails start landing in spam. That is how a database you ignored starts costing you deals you would otherwise have had.

Run the list through a validation service, archive the hard bounces, and check your consent and opt-out flags survived whatever migration got you here. Someone who unsubscribed in 2024 must still be unsubscribed in 2026 — under CASL and TCPA that is not a nicety, and "the data moved systems" is not a defence.

Step three: cut the pipeline down

Almost every unused pipeline has the same disease: too many stages, none of them defined. Eleven stages feels thorough and is unusable, because two of them overlap and nobody can say which one a given deal belongs in. So deals stop moving and the pipeline becomes decoration.

Five or six stages is plenty for most agents. The test for whether a stage is real is not whether it describes something — it is whether you can state the single event that moves a deal out of it.

• New — exits when someone has had a two-way conversation with them.
• Qualified — exits when timeline and financing are known and recorded.
• Active — exits when an offer is made or they go quiet for 30 days.
• Under contract — exits at close or collapse.
• Closed, and Nurture for everything that went quiet.

If you cannot write the exit event for a stage, that stage is a feeling rather than a stage. Delete it and let those deals sit in Nurture, which is at least honest.

Step four: one taxonomy, written down

Tags rot faster than anything else because everyone invents them on the fly. Pick a small set, write them in a document, and archive the rest.

A workable set is source (where they came from), type (buyer, seller, both, referral partner), and area (the neighbourhoods you farm). Three dimensions, each with a fixed vocabulary. Anything you cannot fit into those is usually a note, not a tag.

Step five, and only now: automate

Automation on top of dirty data multiplies the mess — you send three emails to the same person and two to someone who unsubscribed. This is why it is last, not first, however tempting the feature list is.

Start with three, not thirty: a task when a deal enters Qualified, a check-in on anything that has been silent for 30 days, and a closing anniversary touch for past clients. That last one is where most referral business actually comes from, and it is the one nobody sets up.

What Monday should look like

One record per person. Five stages you can define out loud. A tag list that fits on a sticky note. Three automations running. And a saved view that answers "who do I call today" without you scrolling through everyone you have ever met.

That is a weekend. If your database is fifteen years across three systems, it is not a weekend, and anyone telling you otherwise has not looked at it. That is the point at which it is worth having someone audit it first and give you a real number.