How to Clean Up Your CRM Before You Add an AI Agent

Before adding an AI agent to a CRM, reduce tags and fields to a small set where each one answers a question the business actually asks, merge duplicates, rewrite templates in one voice, and set rules for who can add new tags. An agent treats every field as true. Whatever mess is in the CRM today, the agent will repeat it quickly and with confidence.

The CRM you have is a record of every idea anyone ever had

Nobody designs a messy CRM. It accumulates. One person tags leads by source. Their replacement tags by neighborhood. A webinar in 2021 adds forty tags in an afternoon. Someone imports a spreadsheet and every column becomes a custom field.

People cope with this because people know which tags to ignore. An agent doesn’t. It reads “Hot Lead 2019” as a fact about today.

When we started at BrightWork Realty Advocates, the CRM held 2,642 tags. The agent we were there to build didn’t get switched on until that number was about 60.

Step 1: export everything and count it

Pull every tag, custom field, stage, template, and sequence into a spreadsheet. Add two columns: how many records use it, and when it was last applied.

The count tells you how big the job is. It’s usually less work than the number suggests, because most of the list is dead.

Step 2: find what’s alive

Sort by last applied. Anything untouched for a year is a candidate for removal. Anything attached to fewer than a handful of records probably started as a typo or a one-time campaign.

At this point most teams find that a few dozen tags carry nearly all the real activity.

Step 3: write down the questions the business asks

Do this step without the CRM open.

List what you need to know about a contact to decide what happens next. Where did they come from? What do they want? How soon? Who owns the relationship? Are they a past client? Your list will be short.

Every tag or field you keep has to answer one of those questions. If it doesn’t, it goes, however clever it seemed at the time.

Step 4: rebuild a small, controlled set

Create the new tags with a naming pattern a stranger could read: the question, then the answer. Map old tags to new ones in your spreadsheet, apply the new set in bulk, then archive the old.

Merge duplicate contacts while you’re here. An agent that finds two records for one person will follow up twice.

Step 5: fix what the agent will sound like

Templates and sequences are where your voice lives in the CRM. They were usually written by several people over several years, and they read that way. At BrightWork the same pass covered 16 sequences and 86 templates.

Rewrite them against one written description of how the company talks. That description is part of what we call Groundwork, and it’s what the agent drafts from afterward.

Step 6: set the rules for adding new tags

A clean CRM drifts back within a year unless someone owns it. Decide who can create a tag, and write down the naming pattern. One owner, one page of rules.

Then switch the agent on

With the CRM cleaned, the agent’s judgments have something solid under them. At BrightWork, the Revenue Agent has reached new leads in a median of 7.7 seconds since June 2026, and the founder trusts what it tells him because he knows what every tag means.

This is what we mean by knowledge first, integration second, autonomy last. It’s also most of what AI implementation involves in practice, before anything that looks like AI shows up.

Read the BrightWork case study · What a Revenue Agent does

Questions we get

Do I need to clean my CRM before using AI?

Yes. An AI agent treats CRM data as fact. Outdated tags, duplicate contacts, and inconsistent templates produce wrong prioritization and off-voice drafts. Cleanup comes first.

How many CRM tags should a small business have?

Few enough that every one answers a question the business asks when deciding what to do next with a contact. For most companies that’s dozens. It’s rarely hundreds.

How long does CRM cleanup take?

It depends on the size of the database and how many people have shaped it. The audit in steps one and two tells you within a day or two how big the job is.

Which CRMs does this apply to?

All of them. We’ve done this work in HubSpot and Follow Up Boss, and the same steps hold for Pipedrive and Zoho.

What is Groundwork?

Groundwork is the structured record of your company that every agent we build reads from, including how you talk. A clean CRM plus Groundwork is what an agent needs to be useful.