Staff answer the phone. That part stays human. Everything that happened after the call is where the time went.
The problem: every call looked like a job
The documented intake problems were fivefold:
- Unnecessary leads - spam, greetings, internal conversations, wrong numbers and simple follow-ups all creating job records.
- Manual call review - someone having to understand the problem, write a summary and choose the correct service category.
- Manual data entry - name, phone, email, address and the request itself copied across by hand.
- Repeated record creation - a qualified call needing both a lead and a job, so the same details were entered twice.
- Manual follow-up - requesting photos or video so the technician could prepare, done by hand each time.
What was deployed
The workflow picks up after the call ends. It checks that the conversation actually contains useful customer dialogue before doing anything further. It then summarizes the request, selects the closest service category from the locksmith and door-service list, and extracts customer details.
A decision gate follows: if it is not a genuine service request, the workflow stops. A usable request also has to carry a real customer name and phone number. Everything else - spam, wrong numbers, internal chatter - dies at that gate instead of becoming a record somebody has to clean up later.
Qualified requests create the customer, a lead marked in progress, and the related job with notes, location and tags. Where the customer supplied an email, an automated message asks for photos or video of the work area so the technician arrives prepared.
The compounding benefit
Filtering non-jobs matters more than it first appears. Every false lead is not just a wasted record - it is staff time spent later discovering it was never a customer. Pushing that decision upstream is where the real saving lives, because the quality of the pipeline improves at the very first step.
