All work

What 35,000 customer conversations showed me about where leads die

AI & Automation, 2 min read

Warm leads found in 35,000 messages
200+
The problem
Interested leads from a cold text campaign sat without a reply for two weeks. I needed to know how many leads we had dropped across about 35,000 messages in our CRM.
My role
I pulled and sorted the data with Claude Code, built the follow-up list and changed the next campaign based on what I found.
Result
Found 200+ warm leads that never got a follow-up. Three of them became customers.

Skills AI agents, pricing, process design

Context

Every year I pull the public short-term rental permit list for our area and text Airbnb hosts, offering us as a backup cleaning team. Over two years that has added up to about 7,000 contacts and 35,000 messages in GoHighLevel. Fast communication is one of our main selling points, so leads going unanswered was a real problem.

What I did

  • Pulled every conversation. I exported the full message history instead of a sample, because I knew I would want the same view for our realtor campaign later.
  • Sorted it into three groups. Warm (replied with interest but never booked), cold (never really responded) and do not contact (asked us to stop or are already clients). Each contact got the date of the last message and who sent it.
  • Checked the results. I read a sample from the warm and cold groups to confirm the tags were right. I also had the sales agent flag any mistakes while he worked the list. He didn’t find any.
  • Followed up with warm leads first. The sales agent sent a personal follow-up to all 200+ warm leads before we texted the cold list again.
  • Changed the offer for the cold list. Most of these contacts never replied the first time, so I made the offer stronger: first turnover free, 25% off the second month, a 12-month price lock and a limited number of spots.

Results

  • 200+ warm leads found that had never been followed up with
  • 3 became customers, which was more than I expected after two weeks of silence
  • The cold campaign is running again with the new offer. I’ll update this when it finishes.
  • The whole analysis took about one afternoon

What I’d do differently

  • Run this pull every month instead of after a problem shows up.
  • Give the sales role a response-time target from day one.