Capturing the sales calls the CRM could not review
A working Android prototype that linked calls to leads, masked phone numbers and uploaded recordings for the CRM’s AI agents. It was demonstrated but never rolled out.
Working POC · not launched
My role. Built the prototype and tested whether its hardware and support requirements could work at sales-team scale.
The recording gap
- Conversation across a typical sale
- ~90 min
- Recorded conversation available before the POC
- 15–20 min
- Dependent on call transcripts in the CRM
- 5 AI agents
The AI only had part of the conversation
A typical sale involved about 90 minutes of conversation across several calls, but only 15–20 minutes was available as recordings. Without the rest, the CRM’s AI agents had little context for reviewing the sale.
Salespeople could disconnect the phone, delete a recording and reconnect, preventing the upload. Calls made from a personal phone left no record in the CRM.
Earlier cloud-calling tests had lower pickup rates than calls from individual phone numbers. That made it worth testing whether a company handset could keep the calling experience while reliably capturing the conversation.
Connect the whole workflow on one device
I modified a rooted Android phone so a server could instruct it to dial when a salesperson clicked Call in the CRM. Root access let me change behaviour that an ordinary app could not control.
Lead numbers were masked in outgoing and incoming calls, along with matching WhatsApp contacts and messages. Access to connectivity settings and stored audio was restricted, closing the demonstrated disconnect-and-delete workaround.
The demo ran end to end on a couple of compatible Android devices: select a lead, place the call, record it and upload it against that lead.
The operational cost stopped the rollout
The team’s existing phones were too old to modify reliably. A rollout would have required around 300 compatible handsets at roughly ₹28,000 each, plus a dedicated team to maintain them. That support team did not exist, and the program was not a priority.
The founders and CTO saw the demo, but it was not approved for production. I agreed with that decision: proving a workflow on a few phones did not make it practical to maintain across the sales floor.
Scaler moved to Exotel for production calling a year later. That choice avoided modified hardware and supplied recordings for the CRM’s AI agents.