Roshan Kumar

Roshan Kumar

AI Product Manager

I build agentic products
for enterprise workflows.

Iro AI / Lohono Stays

Night Shift Agent

After-hours villa enquiries with a clear next step.

Role
Founder & Product Lead
Delivery
Seven-week live pilot
Scope
Five agents, roughly 300 calls a day

90% of call scenarios handled end to end. The remaining scenarios moved to human follow-up, with caller details and a callback window preserved. The client approved full handover after reviewing the pilot.

The problem

Guests called from different time zones while no representatives were available overnight. Enquiries reached voicemail instead of progressing toward a booking.

Product decisions

Route by location before property.

Live calls showed guests usually started with a destination, not a villa name. Changed routing to collect location first, reducing early back-and-forth.

Keep specialist handling inside one conversation.

Orchestrated five flows for triage, availability, property details, booking support and non-business calls. Guests experienced one continuous interaction.

Make human follow-up part of the product.

Unresolved enquiries retained contact details and a next-morning callback window. Noise or unclear intent could lead to follow-up rather than a dead end.

Pilot results

Live use

Roughly 300 calls a day over seven weeks, across real guest enquiries and noisy calling conditions.

Resolution

90% of call scenarios handled end to end. The remaining 10% escalated with details preserved for next-day follow-up.

Iteration

Reviewed answer gaps weekly and updated knowledge coverage. Changed routing around observed caller behavior.

Outcome: client approval for full handover. Human follow-up remained the path for escalated enquiries.