Where they started

The programme was running a helpline in English and Hindi, on a single 10-digit number, staffed by four operators. It worked for urban callers who spoke Hindi. It didn't work for anyone else — which meant it didn't work for most of the population it was supposed to serve.

Field ASHAs were spending their evenings on personal phones, triaging calls their own supervisors couldn't handle. There was no way to route an urgent case to the nearest clinic. There was no data on who was calling, or what they were calling about, or what happened next.

What we built

  • An 11-language IVRS flow, with audio recorded by frontline workers from each linguistic region. Not professional voice actors — actual ASHAs from the districts. That single choice tripled the completion rate.
  • A risk-flag triage. When a caller answered questions in a pattern that indicated a possible emergency (excessive bleeding, no foetal movement in 12 hours, severe headache with vision changes), the system routed her live to the nearest PHC on-call number. Every routing was logged and reviewed weekly.
  • A supervisor dashboard. Grafana, three panels, colour-coded. District officers could tell at 8am how many calls came in overnight and how many were escalated. No training required.
  • A callback queue. If a caller hung up mid-flow, the system called back once within 24 hours — quietly, at a different time of day, from a different number.

What it moved

Year one: 1.2 million calls answered. 11 languages live. 38% of callers came back for a second call. Average time-to-connect: under 8 seconds. 97% of urgent-flagged cases reached a PHC on-call.

The number that mattered most to the programme director, though, wasn't in any dashboard. It was that ASHAs had stopped triaging calls on their personal phones. The system was picking those up now.

"The IVRS line reached people in three states our outreach workers had never been able to call. Honestly the most useful partnership we've had in five years."

What we'd do differently

The first version had 22 branches at the top of the flow. It was too many. We cut it to 5 in month three and lost nothing in coverage — but recovered 40% of the calls that had been dropping in the first 30 seconds. Start smaller than you think you need to.