A woman who stays in the program is a woman who receives the information that saves lives.
Each day, nearly 800 women die from complications that need not be fatal — a quiet catastrophe unfolding at the margins of information access. In India, where the Kilkari program has already reached tens of millions of pregnant women through voice messages, the challenge has never been vision but friction: the wrong moment, the shared phone, the missed call. An AI system called CHAHAK now enters that gap, learning the rhythms of individual lives to ensure that knowledge — about labor, about newborns, about survival — arrives when a woman can actually receive it.
- Half of Kilkari's 4.3 million active users stop engaging within six months, not from indifference but from the stubborn logistics of poverty — shared phones, unpredictable schedules, calls that land at the wrong hour.
- Every dropout is a quiet severance from information that could prevent a death, making disengagement not a program statistic but a public health emergency.
- CHAHAK responds by replacing uniform scheduling with machine-learned personalization, cutting weekly call attempts nearly in half and flagging the women most likely to fall away before they do.
- Community health workers dispatched to at-risk women boosted listenership by 16 percent; automated reminders added another 5 — together preventing a third more dropouts than standard outreach.
- ARMMAN now aims to carry this model to 70 million beneficiaries within seven years, with ambitions to extend the same AI logic into chronic disease, elder care, and beyond.
Nearly 800 women die each day from preventable complications — a number that frames everything India's Kilkari program is trying to do. Launched by the Ministry of Health and Family Welfare alongside the nonprofit ARMMAN, Kilkari has grown into the world's largest maternal mobile health initiative, delivering weekly voice messages in regional languages to pregnant women and new mothers across 28 states. Its cumulative reach has touched 53 million women and children.
But reach is not the same as impact. About half of participants stop listening within six months. The causes are ordinary and persistent: calls arrive when women are working, phones are shared among family members, and the sheer logistics of mass outreach create constant friction. A message meant to save a life may never reach its intended ears.
CHAHAK — whose name means "chirping of birds" in Hindi — was developed by researchers from Harvard, Google Research, and ARMMAN to address exactly this gap. Using machine learning, the system analyzes call logs to identify the best moment to reach each individual beneficiary, replacing random scheduling with personalized timing. It also flags women at highest risk of disengaging, enabling targeted follow-up: a visit from a community health worker, or an automated reminder.
A pilot in Odisha involving 4,000 beneficiaries confirmed the approach. Weekly call attempts dropped from 4.86 to 2.5 per person. When health workers visited at-risk women identified by the system, listenership rose 16 percent; automated reminders added another 5. Together, these interventions reduced dropout rates by 33 percent — meaning a third more women continued receiving information about labor, newborn care, and immunization.
ARMMAN is already planning to scale Kilkari to 70 million beneficiaries within seven years, with CHAHAK at the center of that expansion. The same model, the organization believes, could eventually reshape public health delivery far beyond maternal care — wherever the distance between information and the people who need it is measured not in miles, but in missed calls.
Nearly 800 women die each day from complications that could have been prevented. That statistic, from the World Health Organization, sits behind everything India's Kilkari program is trying to do. Since its launch by the Ministry of Health and Family Welfare and the nonprofit ARMMAN, Kilkari has become the world's largest maternal mobile health initiative, sending weekly voice messages in regional languages to pregnant women and new mothers with information about prenatal care, immunizations, and family planning. The numbers are staggering: over 4.3 million active beneficiaries across 28 states and union territories, with a cumulative reach touching 53 million women and children.
But scale alone doesn't guarantee impact. About half of Kilkari's participants stop listening within six months of enrollment. The reasons are mundane and stubborn: calls arrive at inconvenient times, phones are shared among family members, the sheer logistics of reaching millions of people at once create friction. A woman might miss a call because she's working. A husband might answer the phone. A message meant for a pregnant woman might never reach her ears. This disengagement is more than a program management problem—it's a public health failure, because the information those messages carry could literally save her life.
Enter CHAHAK, an AI system developed by researchers from Harvard University, Google Research, and ARMMAN. The name means "chirping of birds" in Hindi, and the system does something deceptively simple: it learns when individual beneficiaries are actually available to take calls, and it identifies which women are at highest risk of dropping out. The technical machinery involves machine learning algorithms that analyze call logs to find optimal timing windows for each person, replacing Kilkari's random scheduling with personalized delivery. When the system flags someone as likely to disengage, it can trigger targeted interventions—a visit from an Accredited Social Health Activist, or an automated reminder call.
The proof came from a rigorous test in Odisha involving 4,000 beneficiaries. CHAHAK reduced the average number of call attempts needed to reach someone from 4.86 per week down to 2.5. That's not just efficiency; it's the difference between a woman hearing critical health information and never hearing it at all. When ASHAs—community health workers—visited women identified by the system as at-risk, listenership jumped by 16 percent. Automated reminders added another 5 percent. Together, these interventions prevented 33 percent more dropouts compared to control groups, meaning a third more women stayed engaged with the program and continued receiving the information they needed.
What makes this significant is not just the numbers, but what they represent. A woman who stays in the program for nine months instead of dropping out at six is a woman who receives information about labor complications, newborn care, immunization schedules—the kind of knowledge that can mean the difference between a healthy birth and a tragedy. The system works because it meets people where they actually are, not where a program assumes they should be.
ARMAN is already thinking bigger. The organization wants to expand Kilkari to 70 million beneficiaries within seven years, and CHAHAK is positioned to be central to that scaling. The team envisions a phased rollout starting with a single state before moving nationwide. Beyond Kilkari, the same approach—using AI to optimize timing and target resources toward the people most likely to need help—could reshape how other public health programs work, from chronic disease management to elder care. The blueprint exists. What happens next depends on whether the system can work at the scale India's maternal health crisis demands.
Notable Quotes
Nearly 800 women die each day from preventable pregnancy and childbirth complications— World Health Organization
The Hearth Conversation Another angle on the story
Why does timing matter so much? Couldn't the program just send messages whenever and let people listen when they want?
Because a message you never hear is useless. If a call comes when someone's at work or when a family member answers, the pregnant woman never gets the information. CHAHAK learns each person's actual availability and calls then instead.
How does the system know who's about to drop out?
It watches patterns in who answers calls, who listens to messages, how engagement changes over time. If someone's answering less frequently, the system flags them early and arranges an intervention before they disappear entirely.
What's the difference between a visit from an ASHA and an automated reminder?
An ASHA is a real person from the community who can talk through concerns, answer questions, build trust. A reminder is just a prompt. Together they work—the reminder keeps someone engaged, the ASHA deepens it.
Does this actually save lives, or is it just better engagement metrics?
Better engagement means more women hear information about warning signs during pregnancy, about when to seek care. That directly translates to earlier intervention when complications arise. The engagement is the mechanism.
What happens if the system scales to 70 million people?
That's the real test. Right now it's been proven on 4,000 beneficiaries in one state. Scaling means making sure it works across different regions, languages, economic conditions, phone access patterns. The blueprint is there, but execution at that scale is entirely different.