One morning in 2016, while working as the director of the obstetrics and gynecology department at a nonprofit hospital in rural Cambodia, I received a phone call from the maternity ward stating that a patient was in trouble. Although the woman had successfully delivered the baby, she had an extremely swollen labia major 90 minutes later, suggesting a second child was coming. I lived near the hospital and arrived within minutes. What I found was not a second baby but a massive expanding vulvar hematoma. My colleague had tried to suture the bleeding vessel but was unsuccessful for more than two hours before calling me. I attempted to perform a ligation, which also failed to control the bleeding, and then had to pack the vagina with two large surgical gauzes—the kind normally reserved for open abdominal procedures. The patient survived, but it was almost too late.

Data from national maternal death audits in Cambodia between 2010 and 2017 highlights an institutional shift from access to quality as the main challenge affecting maternal survival. Although reported maternal death cases dropped from 176 to 59 over this period, the women who died in 2017 were significantly more likely to have received antenatal care (90.0% vs. 68.2%) and to have given birth inside a health-care facility (81.6% vs. 55.3%) compared to 2010. This reveals that basic health-care access is no longer the primary bottleneck; instead, survival hinges entirely on the quality of emergency care, swift protocol execution, and health-system readiness at the moment a crisis occurs.

Today, low- and middle-income countries are increasingly gaining access to an array of digital health innovations such as portable handheld ultrasound devices driven by artificial intelligence (AI) for identifying placental abnormalities or automated digital early-warning systems for tracking postpartum vital signs. Analysis of cross-country panel data suggests that AI adoption is correlated with reductions in maternal mortality, showing the most pronounced statistical gains in regions having the highest baseline mortality. However, without proper workforce training, integrated clinical workflows, and institutional readiness to handle emergencies, even the most advanced digital tool will not make a tangible difference.

In the rural hospital where my patient had the emergency that morning, no triage algorithm was going to help, because the limiting factor was not detection but response capacity. The facility had a functioning chain of human accountability, although that had already begun to break down before I walked through the door. Some AI tools carry hidden assumptions when diagnosing medical conditions, including clear clinical responsibility and staff who know their roles in an emergency. They assume that handovers happen, that documentation is complete, and that someone will act on the algorithm's output.

In some low-resource hospitals, these assumptions are wrong. The midwife who misread the patient's condition that morning was not negligent; she worked in a system that had never clearly defined her escalation responsibilities. No triage AI tool would have changed what she did. A postpartum-risk tool might have flagged the patient as a high priority, but if no one with authority is watching the output, the flag would just fade away.

AI Brings Promise for Maternal Health

At the end of February 2026, I participated in the Design for Implementation conference in Chicago, organized by the Coalition for National Trauma Research with support from the Agency for Healthcare Research and Quality. The conference gathered health-care professionals to discuss strategies to refine health-system structure by coordinating partners, processes, and tools to better serve the needs of all patients.

The central question that sessions explored at this meeting cut across every clinical discipline: how do innovations reach the bedside? I have been trying to find an answer to this question for more than 10 years while working in maternal health systems in low-resource settings.

Session after session returned to the same uncomfortable reality: clinician engagement and understanding of AI tools cannot be assumed; it has to be deliberately designed, accounting for cognitive load, trust, and institutional culture. There was also a clear consensus emerging that AI should augment, rather than replace, clinician input, because clinical judgment responsibility cannot be delegated to the tool.

Applied to maternal health, this distinction becomes clear: an AI tool that flags preeclampsia risk during an antenatal visit is only useful if the nurse reading it is confident in the tool's assessment and knows to alert the physician. Practical approaches discussed at the conference centered on exactly this gap: embedding dedicated clinical champions—physicians or senior nurses who understand the tool and can support colleagues in adopting it—within departments, and establishing structured review sessions so that staff understand what the algorithm flags and why through the tool's own features or departmental case review.

Evidence from maternal health settings shows what becomes possible when these conditions are met. An AI-driven early-warning system evaluated across district-level hospitals in sub-Saharan Africa in 2024 generated real-time risk scores for postpartum hemorrhage and eclampsia at 15-minute intervals, delivering automatic alerts to clinical staff and showing early results suggesting meaningful improvement in the timely detection of both conditions. Machine learning models have shown that postpartum hemorrhage—the leading direct cause of maternal death globally, including in Cambodia—can be predicted with meaningful accuracy before it occurs, giving clinical teams time to prepare rather than react. Triage support tools can also help junior staff recognize warning signs, like swelling, that they might otherwise miss, and clinical decision support systems can ensure that staff consistently use lifesaving interventions. Risk prediction models can flag high-risk pregnancies earlier in the antenatal pathway, and research has confirmed this capacity.

In parallel, digital health interventions have demonstrated real results in low-resource settings when embedded within structured care pathways. The QUALMAT project, which introduced an AI-assisted electronic clinical decision support system for antenatal and intrapartum care in rural health facilities in Ghana and Tanzania, achieved adoption rates of more than 70% of all antenatal care visits within two years. The system guided health workers through WHO-recommended care steps in real time, reducing practice variation without requiring additional specialist staffing.