How a Rooftop Solar Installer Books Site Surveys with a Bengali AI Agent

The installer

A rooftop solar sales and installation company in West Bengal, selling to homeowners and businesses. Its customers mostly speak Bengali. Demand is driven by the central rooftop solar subsidy, so leads arrive in bursts from ads and referrals, often in the evening and at weekends.

The challenge

• Leads outran the team. Every enquiry needed a first call to check whether the home was suitable, and a small team couldn't call them all quickly.
• Speed matters. Homeowners enquire with several installers; the first useful call usually gets the site survey.
• Language. Customers are most comfortable in Bengali, with English technical words like net metering and subsidy.
• Only the site survey matters. The goal of the first call is a booked free survey; the field team does the rest.

What was built

An outbound Bengali AI voice agent that calls every new lead from the installer's CRM automatically, and an inbound agent that answers enquiry calls. After each call, the conversation is turned into structured fields (interest, property type, electricity bill, objections, survey date and time, outcome) and the CRM is updated automatically:

• lead fields, activity log and a summary note;
• pipeline stage moved to Contacted, or Meeting Scheduled when a survey is booked;
• a meeting created and handed to the field rep;
• a follow-up task if the lead wants time, and daily retries if they didn't answer.

A custom analytics dashboard shows interested leads, not-interested leads, meetings scheduled and follow-ups created, so the team sees the funnel without listening to calls.

Results so far

Figures as of September 2026, rounded:

• About 1,000 calls analysed across outbound and inbound agents.
• 300+ free site surveys booked from outbound lead calls, roughly one in three analysed calls.
• About 200 callbacks scheduled for leads who wanted to talk later.
• The inbound agent has booked surveys on roughly one in three of its analysed calls since going live in August.

What happens after the survey (installations and revenue) is handled by the field team and isn't measured in the call data.

What made it work

• One goal per call. The agent's job is the site survey, not the sale.
• Guardrails on facts. The agent never invents prices or subsidy amounts; exact quotes come from the field rep.
• Natural Bengali. It accepts soft answers like haan, hmm and ji as a yes, and switches language when the customer does.
• Automation after the call. Every call updates the CRM, so no lead is forgotten.
• Calling window. Calls only go out during sensible hours.
• Hardened together. Early issues, such as confirming bookings reliably and making sure CRM updates are never silently missed, were found in call reviews and fixed in the first months.

Related

• Download this case study as a one-page PDF
• AI voice agent for solar and energy companies
• Rooftop solar subsidy calculator
• Bengali AI calls for West Bengal businesses
• Book property site visits with AI calls
• Send AI call results to your CRM or Google Sheets