A research system for a two-person sales team

This is a project for a property tax consulting firm of about twenty people. It helps businesses lower their property tax bills, and it finds almost all of its clients by outbound. A sales team of two covers roughly 38 states, and the selling season runs from autumn to spring. I am not naming the client.

The problem

To know whether a company was worth calling, the team had to open each county tax website, one county at a time, and find the bill. Some of the sites ask you to prove you are not a robot. It was the biggest time burn in the week, and it was the part that decided who got a call.

The CRM held thousands of records, and the data was going stale. Contacts change jobs, fields were blank, and nobody was sure which numbers to trust. The work is all outbound and seasonal, so there is one short window a year to use what the team knows.

What we built

We started with a free proof. Before anything was signed, we found real tax bills in one county and showed the team what a ready-made research file would look like.

Then the research system. Claude Code reads county tax portals for the team, and it only counts real bills, never estimates. It covers 1,937 counties in 39 states, and 595 of them are read automatically by a plain script.

For each company with a real bill, the system finds the people who handle the tax and writes them into the CRM. It fills empty fields only, so nothing the team typed is overwritten.

We connected Claude to the CRM during a working session and built a dashboard on the call. It shows which accounts have more than one reason to call right now, and the team uses it to decide who to call first.

Each account now carries its tax bills over the years, so the team opens an account and sees the history instead of searching for it.

We also ran read-only checks on the data they already had: duplicate accounts, accounts with too few contacts, and email addresses that no longer work. We flagged what we found and deleted nothing. Last, we wrote a hand-off guide so the team can run the system without us.

What has happened so far

• 1,937 counties mapped, 595 read automatically (our count, 21 October 2026).
• About 14,100 real tax bills stored on 979 accounts (our count, 21 October 2026).
• 2,172 email addresses checked, with the bad ones flagged (16 October 2026).

The firm began its autumn outbound season on 14 October 2026, and most of its deals are signed between March and May. We do not have sales results yet. We will add them to this page when we do.

What this shows

When the slow part of selling is finding out who is worth a call, a system that does that research once, and keeps the CRM clean, gives a small team the reach of a bigger one. See sales systems.