Op-Ed · CRM & Data

What 2,200 messy records taught me about segmentation

Everyone wants better targeting. Almost nobody wants to do the unglamorous work that makes targeting possible.

When people talk about segmentation, they usually jump straight to the fun part: the personas, the clever names, the matrix with four neat quadrants. I get the appeal. But I spent a good chunk of an internship learning that segmentation doesn't start with a framework. It starts with a database you can actually trust.

The CRM I inherited at the Greensboro Convention & Visitors Bureau had grown faster than its own structure. Records had been imported from different sources over years. People's job titles were out of date. The same company existed three times under slightly different names. There was no consistent way to group partners by industry, which meant every campaign started with someone manually rebuilding a list and quietly guessing about who was actually who.

You can't target a segment you can't reliably describe.

So before any strategy, I did the boring thing. I transferred, cleaned, tagged, and standardized 2,200+ contact and company records, resolved the import errors, and validated contacts against their current roles. Then I classified 1,100+ partners across 25+ industry categories so the database could finally be filtered instead of skimmed.

The lesson hiding in the tedium

Here's what surprised me: the cleaning was the segmentation. Every decision about how to categorize a partner (which industry, which tier, which tag) was a small strategic choice about how the team would later think about its audience. By the time the data was clean, the segments had basically defined themselves, because I'd been making segment-shaped decisions the entire way through.

That reframed how I see the work. By the time a CRM is genuinely clean, you've already made most of the calls a segmentation would have made anyway, so it's closer to an early draft of the strategy than the setup for it.

Three things I'd tell past me

• Decide the taxonomy before you tag. Tagging on instinct gives you 40 categories that mean almost the same thing. Agree on the buckets first.
• Validate against reality, not the spreadsheet. A record can be perfectly formatted and completely wrong. Roles change; check them.
• Treat structure as a product. The goal isn't a tidy list, it's a database the team can ask questions of without calling you.

The flashy deliverables, the personas and the campaign recommendations, all came later, and they were better for it. But if you ask me where segmentation actually starts, it isn't on a slide. It starts with the unglamorous decision to make the data tell the truth.

Related: the Greensboro CVB case study