Estimated reading time: 4 minutes · Last updated:
PROMPTS is an AI-powered SMS system run by Jacaranda Health that answers pregnancy and newborn questions and routes urgent cases to nurses. As first reported by NPR, the service now handles about 15,000 messages a day from users in 24 Kenyan counties and roughly 700,000 people a year. PROMPTS understands English, Swahili and Sheng, flags about 7% of incoming messages as potentially urgent and sends those directly to a human triage queue. Its developers say the goal is to shorten delays that contribute to Kenya’s roughly 6,000 maternal deaths a year.
If a mom in a rural part of Kenya starts bleeding at 2 a.m., she can't just Google it.
Jay Patel, director of technology, Jacaranda Health
Key takeaways
- Scale: PROMPTS handles about 15,000 SMS messages a day and reaches roughly 700,000 users each year.
- Geography and language: The system is active in 24 Kenyan counties and supports English, Swahili and Sheng, with versions being built for Twi and Hausa.
- Safety routing: About 7% of messages are flagged as potentially urgent and routed straight to a nurse queue.
- Human oversight and cost: Eighteen nurses monitor and audit responses; current nonprofit funding absorbs a maintenance cost of about $2 per user.
Table of contents
- Key takeaways
- How PROMPTS turns texts into triage
- Scale and reported impact on delays
- Safety limits, governance and technical trade-offs
- Expansion, local context and the limits of AI in low-resource settings
- How PROMPTS could scale — and what could stop it
- What to be careful about
- Frequently asked questions
How PROMPTS turns texts into triage
Jacaranda Health launched an automated SMS reminder service that evolved into PROMPTS after patients began replying with questions. The system uses language models tailored to local speech patterns to parse incoming messages in English, Swahili and Sheng and to prioritise them by perceived urgency.
When PROMPTS identifies a message as potentially urgent it bypasses automated replies and places the conversation in a special nurse queue for immediate review. Jacaranda employs a team of 18 nurses who monitor, audit and respond around the clock, and the organisation says that human oversight is central to the model’s safety.
Scale and reported impact on delays
Usage grew from under 100 questions a day when Jacaranda first staffed replies to roughly 15,000 daily messages today, serving an estimated 700,000 users each year. The volume gives clinicians a new signal about gaps in care: aggregated, anonymous queries can show where routine checks are being missed or where a cluster of symptoms is emerging.
Kenya records about 6,000 maternal deaths annually, which Jacaranda and partners link in part to delays in seeking care; nearly a third of those deaths are believed to stem from such delays. PROMPTS’ back-and-forth can steer a mother to a nearby facility sooner than she might otherwise travel, or conversely reassure her when care can wait.
Safety limits, governance and technical trade-offs
PROMPTS is deliberately narrow in scope and trained on pregnancy- and newborn-related concerns to reduce error rates, but errors remain the core risk. Javaid Iqbal Sofi, a researcher on AI policy at Virginian Tech University, asks what happens if the system misclassifies an urgent report as routine and whether secondary checks will reliably catch that mistake.
Jacaranda’s mitigation is the nurse queue and continuous auditing. Still, maintaining that human layer raises operational costs and creates a funding dependency: the nonprofit reports a maintenance cost of about $2 per user that it currently absorbs.
Expansion, local context and the limits of AI in low-resource settings
Jacaranda has adapted PROMPTS for parts of Ghana, Nigeria and Tanzania and is building versions for Twi and Hausa, emphasising that local cultural and nutritional advice cannot be copy-pasted across countries. Laura Down, a Jacaranda spokesperson, says local phrasing and common foods must shape answers or recipients will reject them.
Experts note these interventions work best where a functioning health system exists to receive referrals. Smisha Agarwal, director of global digital health at Johns Hopkins University, says AI can change the direction of clinical conversations by centring patient questions, but only when clinics can act on the referrals.
| Country | Languages | Users / scale | Notes |
|---|---|---|---|
| Kenya | English, Swahili, Sheng | about 700,000 users a year | Active in 24 counties; handles ~15,000 messages/day |
| Ghana | Twi (version built) | Expanded to parts of the country | |
| Nigeria | Hausa (version built) | Expanded to parts of the country | |
| Tanzania | Swahili (adaptation) | Expanded to parts of the country |
How PROMPTS could scale — and what could stop it
The case for
- Wider deployment could give clinicians near real-time feedback from roughly 700,000 users a year, improving service delivery where clinics act on referrals.
- Language-specific models for Twi and Hausa reduce rejection risk and make triage more accurate for non-English speakers.
The case against
- Sustaining human oversight at about $2 per user creates a funding barrier unless donors or governments underwrite ongoing costs.
- In areas with very weak clinical capacity, routing patients sooner will not reduce mortality if no facility can and will accept referrals.
What to be careful about
- Misclassification risk: an urgent symptom might be treated as routine and not sent to the nurse queue.
- Privacy and data-use concerns from collecting SMS conversations, especially if retention or sharing policies are unclear.
- Financial sustainability: the $2 per-user maintenance cost depends on continued nonprofit funding.
- Equity limits: PROMPTS requires a reachable clinic to turn referrals into care, so impact varies by county.
The bottom line
PROMPTS shows how a narrowly focused AI model, paired with human oversight, can extend clinical reach via basic SMS. Its scale — roughly 15,000 messages daily and about 700,000 users yearly — gives health workers new visibility into service gaps and common symptoms. But the model’s benefits depend on reliable triage, transparent data practices and continued funding to keep nurses in the loop at roughly $2 per user. Independent evaluations and clearer funding commitments will determine whether PROMPTS moves from a promising pilot to a durable part of maternal care.
What to watch
- Watch for Jacaranda Health’s announcements on national rollouts to additional Kenyan counties; no date has been set.
- Watch for published evaluations measuring PROMPTS’ effect on care-seeking or clinical outcomes; no date has been set.
Frequently asked questions
How many people use PROMPTS each year?
Jacaranda estimates PROMPTS serves roughly 700,000 users a year, handling about 15,000 messages a day.
How does PROMPTS identify urgent cases?
PROMPTS flags about 7% of messages as potentially urgent and routes them straight to a nurse queue monitored by an 18-person nursing team.
Who runs and funds PROMPTS?
PROMPTS is run by Jacaranda Health; the nonprofit currently absorbs a maintenance cost of about $2 per user.
Related reading
This article is information, not medical advice. Anyone acting on it should speak to a qualified professional.
