Why voice matters

We recently built and tested a voice-based extension to PROMPTS, our two-way SMS health platform, and the early results from our pilot with mothers were encouraging. Many mothers who used the voice option chose to call the service; more than 50% returned more than once, and reported that hearing a voice helped them feel supported when they were worried. 

While PROMPTS reaches many pregnant and postnatal mothers from lower socioeconomic backgrounds through SMS, some of the most vulnerable mothers are still less likely to engage. For women who have difficulty reading, have vision challenges, or feel unsure during moments of concern, typing a message can be a barrier. Voice removes the need to read or write and allows mothers to speak directly and receive guidance in a way that is easier to use when they need help.

Integration of IVR with PROMPTS

When a mother calls PROMPTS Voice AI via a toll-free number, she asks her health question in her preferred language (English or Kiswahili). The system first converts her spoken input into text, which is then processed to understand the intent of the question. A clinically approved response is generated and converted back into natural-sounding speech using text-to-speech technology.

PROMPTS Voice AI

The end-to-end process enables mothers to receive timely, hands-free health information through a simple phone call, without requiring smartphones, data, or any advanced literacy. 

Pilot Overview

The pilot was tested over a short period between late November and December 2025. It focused on assessing technical feasibility, real-world performance, and early user engagement and trust, rather than full clinical impact. During this period, we received over 2,800 calls from more than 2,000 unique mothers, providing practical insights into how voice-based support performs in everyday conditions and how it can complement PROMPTS’ existing SMS model before wider rollout.

Preliminary results

• User engagement: Mothers actively used the voice service and returned multiple times

2,883 calls were recorded from 2,060 unique callers during the trial period, with most mothers frequently returning to the service. While 819 users made a single call, 846 users called two to three times, and 395 users returned four or more times, including one highly engaged user who made more than 80 calls.

• Real-world performance and technical feasibility: Over 90% of the calls were understood, and the system responded to health questions

Once a caller’s speech was successfully transcribed, the system performed with near-perfect reliability. Across the full pilot, 91.7% of all calls were completed successfully. 

Most unsuccessful calls occurred at the speech-to-text stage, largely due to real-world conditions such as background noise, quiet speech, or phone quality. These failures did not reflect weaknesses in clinical content or AI reasoning, but instead highlight the importance of improving speech recognition in everyday, low-resource environments.

• User Feedback: 88.1% of mothers said they would use the service again. 

Feedback received was very positive, with 88.1% of the mothers expressing their clear intent to use the service again. 80% liked the advice, and 82.6% recognized that the voice message was advising a specific health action.

Understanding was strong but not universal: 65.9% clearly understood what to do next, while others reported needing more information, pointing to the need for simpler explanations and stronger use of local language.

Trust and emotional response were particularly notable. 81.8% of mothers said they felt supported or reassured after hearing the voice response, and 83.8% said they trusted the service a lot, reinforcing the value of voice as a more human mode of engagement.

• Behavior change: Voice feature led many mothers to seek care or take health-related action

The pilot provides early evidence that voice-based guidance can influence real-world health behavior. After using the service, 46% of mothers visited a clinic or spoke to a healthcare provider, 34% followed the advice at home, and 9% scheduled or attended an antenatal care (ANC) visit. 

Among mothers who did not take action, the most common reason was needing more information (52.9%), followed by the belief that the issue was not urgent (15.0%), suggesting that clearer and more detailed guidance could further increase follow-through.

Looking ahead

The pilot showed that voice-based support is both feasible and valuable for mothers, especially in moments of uncertainty.  With the results, it was clear that no single channel works for everyone. Some mothers prefer voice, others continue to prefer SMS. Going forward, PROMPTS will continue to use a multi-channel approach, allowing mothers to choose how they engage while ensuring that the most vulnerable are not excluded by design.