Case Study
Pest Control AI Agent | SMS Sales, CRM & Payments
Most US pest control companies run their entire business inside FieldRoutes or PestPac. Those systems hold the customer, the route and the billing. If a sales tool cannot write back into them, somebody re-types every deal by hand, and that is usually where the deal dies.
So the agent had to do more than talk. It runs the whole deal over SMS: pitches the service, answers questions, sends a 12-month agreement for signature, validates the payment method, books the technician, then writes all of it back into the CRM. I built the CRM integrations, the contract and payment flow, and the campaign logic that tracks each step through to a closed deal.
How it works
Where the actual work was
Models have been able to hold a friendly exchange about a mosquito treatment for a while now. What took the time was everything after that: sending a real 12-month agreement, validating a card, and landing all of it cleanly in software that predates the phrase "API-first".
FieldRoutes and PestPac are what those companies actually run on, so that is what I integrated with. Each one wants a company subdomain, an authentication key and an authentication token, and past that point the documentation stops being much help. Those integrations are the reason the platform could close its bigger customers, and competitors still have not copied them.
How one deal moves through it
Every conversation is a campaign with a goal attached, and the timeline records each step as it happens. Agreement signed. Payment method validated. Appointment scheduled. Campaign goal completed. When a customer finishes, the thread shows exactly which step closed it and when.
When something goes wrong in production you need to see which tool call fired, what the model got back, and where the CRM write landed, without digging through logs. That is what the timeline is for.
Keeping it safe to run at scale
Tool calling drives every step. Structured outputs keep the CRM writes valid, so a malformed field never reaches the customer record. Anything sensitive waits behind a human approval queue rather than going out on its own.
Python, PostgreSQL and Redis underneath, multi-tenant, in production. The platform this runs on publicly reports $620k in new sales and 1,100 new accounts for one operator, and over $1M in a single month for another.
More screenshots
