Kenya: AI supervision assistant for community health

The problem

Community Health Assistants (CHAs) are the backbone of Community Health Units (CHUs), they are instrumental in turning community-level work into a functioning primary health care system. They supervise and mentor Community Health Promoters (CHPs), validate and troubleshoot reporting, track performance, support referrals and follow-ups, and coordinate with facilities.

Kenya’s Electronic Community Health Information System (eCHIS) has been a major step forward in enabling this work. As a national digital platform for community health, it supports household enrolment and tracking, service delivery workflows, surveillance, and routine reporting, helping shift community health from paper-based processes to more standardized, digital information flows.

As eCHIS expands, it also increases the volume and detail of information available for supervision. This creates new opportunities for CHAs to lead with stronger visibility, but it also means that turning raw system outputs into clear priorities often requires additional context and synthesis. In practice, CHAs face large volumes of information and high-level counts without enough detail to quickly determine what needs attention most. They spend significant time reconciling inconsistencies and translating rows of data into supervision priorities. As a result, supervision time can get pulled toward data correction, and coaching becomes less timely, less targeted, and less specific. This limits CHAs’ ability to spot emerging trends and flag underperformance early, when small course corrections can still make a difference. It leaves less time for proactive patient care and follow-up. CHAs are frequently overwhelmed by manual data verification and administrative reporting, leaving little time for the high-impact clinical coaching that drives community health outcomes.

The solution

To bridge this visibility gap, the Ministry of Health (MoH), Amref, Dalberg Data Insights, and Dalberg Design, co-created the CHA AI Assistant, a WhatsApp-based tool designed as a “quiet enabler”. The tool meets CHAs where they already are, translating raw eCHIS data into timely performance nudges and emerging health trend alerts.

From the various AI use cases we explored, we selected an AI-powered supervision assistant to help CHAs do one thing exceptionally well: turn routine eCHIS data into clear, CHU-level insights they can use for supervision, quickly, consistently, and without adding a new layer of complexity to an already demanding role.

How it works

• CHU Performance Analysis: The assistant reviews data submitted by CHPs and translates it into clear performance insights at the Community Health Unit (CHU) level. It surfaces data gaps and flags inconsistencies for review ahead of monthly meetings and reports, improving data quality from the start.
• Monthly Health Snapshots: In parallel, it produces a monthly synthesis of community health trends. By translating household-visit data into a simple snapshot, it identifies what is rising, persistent, or newly emerging, such as a spike in fever cases or a drop in prenatal visits, so CHAs can guide their teams accordingly.
• Conversational Capacity: Rather than forcing CHAs to hunt through multiple dashboard views to answer specific questions, the assistant supports open-ended, conversational querying. CHAs can ask custom, CHU-level questions in plain language and receive targeted answers on demand.

The gap between having digitized data and being able to use it quickly, confidently, and consistently is the supervision bottleneck this project set out to address. Using AI, not to replace human judgment, but to do the heavy lifting of summarizing patterns, surfacing trends, and highlighting potential issues, we wanted to automate the synthesis work that currently slows CHAs down. The technology acts as an enabler - freeing them up to spend more time deciding what matters and taking action where it’s needed most.

Just as important as what the assistant does is when it shows up. The solution was designed to mirror the existing rhythm of CHAs workflows. Trend alerts are delivered near the end of the month to help CHAs identify emerging community issues and set priorities for upcoming meetings. CHAs preferred WhatsApp over a new app to minimize friction and ensure the tool fit into their daily lives. Importantly, the assistant uses aggregated data, never attempts patient diagnosis, and strictly adheres to national protocols.

The impact

1. CHAs spent less time preparing for supervision and more time coaching
Before the proof-of-concept, CHAs' preparation for monthly CHP meetings was time-intensive. Many spent hours pulling information across eCHIS views, reconciling discrepancies, and manually translating raw data into supervision priorities. With the introduction of the assistant, that manual synthesis burden began to shift, with 87% of CHAs reporting they saved at least 30 minutes preparing for their monthly meetings. By the end of the proof-of-concept, 88% received trends and insights, and 83% were actively using the assistant's health trends and performance insights to prepare for or guide their monthly CHP meetings. 98% of CHAs found the trends and insights easy to understand, and 100% wanted to continue using the assistant.

2. CHAs shifted from reactive supervision to proactive, evidence-based leadership
At baseline, nearly all CHAs (96%) had to manually review each CHP's data as it came in to check for issues or inconsistencies and to decide where follow-up was needed.As a result, many CHAs learned about emerging health trends or CHU performance issues late, sometimes only during monthly meetings or at the end of reporting cycles, with several CHAs sharing that supervision conversations often felt reactive and focused on catching up and resolving gaps rather than preventing issues early and planning targeted follow-up. By endline, 85% of CHAs reported a significantly improved ability to prioritise urgent health issues. This shift also strengthened the quality of coaching conversations: 82% reported an improved ability to explain health issues to CHPs, and 65% said they better understood underlying disease patterns in their CHUs. 100% reported increased confidence in their roles because their guidance was now anchored in evidence rather than general reminders.

3. CHPs took corrective action based on clearer, evidence-based feedback
Before the assistant, supervision often began with searching for problems in raw counts and reconciling inconsistencies, and meetings could drift toward correcting records instead of agreeing on specific follow-up actions. Previously, community health data mostly flowed upward, from households to CHPs, to county and national reporting, without consistently returning to the frontline as guidance that could shape day-to-day decisions. With the assistant, the same routine data was converted into usable direction for coaching, and the data did not just move up, but it also came back down as supervision priorities. By endline, 100% of CHAs reported translating the assistant's insights into direct, actionable feedback for their CHPs. CHPs responded to that clarity, with 91% of CHAs reporting that CHPs took corrective action, such as syncing missing records, correcting data gaps, or following up on high-risk households, specifically because of feedback informed by the assistant.

Lessons learned

• Start with lived experiences, not assumptions. The project started with a simple but important choice: begin with the people doing the work, not with a technology stack or technical roadmap. We carried out an immersion in Machakos County, spending time across Kangundo and Matungulu sub-counties with the people who use and feel the limits of eCHIS every day. Over the immersion, 30 stakeholders were engaged including CHAs, CHPs, focal persons, county health officials and facility staff, through a blend of in-depth interviews, small-group discussions, workflow shadowing, and hands-on platform reviews. As development progressed, we engaged CHA 'AI champions' who tested early versions, reacted to real message flows, and told us plainly what worked and what didn't. Their feedback helped shape the AI Assistant in concrete ways: delivering insights via WhatsApp (a channel they already used daily), framing trends as shared problems to solve together rather than as judgments of CHP’s individual performance and allowing them to view all their CHUs from a single account.
• Safety guardrails were built in from the start. The assistant uses aggregated data, never attempts patient diagnosis, and strictly adheres to national protocols. The same co-design process also helped build understanding and trust around AI. We explained what the assistant would and would not do, how it used eCHIS data, and why it would not replace the CHA’s judgment. Their questions helped us refine guardrails and communication, which later became central to training materials for the remaining CHAs in Machakos County.
• The next step is deliberate, county-by-county expansion. There are currently plans to scale county-by-county to test different conditions (rural/urban, connectivity, device constraints), with county health teams and the Ministry's eCHIS custodians leading onboarding so ownership sits within the system. We see a clear path toward extending the pattern to other workflows beyond supervision, including clinical decision support, administrative transformation, patient access tools, and high-level health system insights for national planning and resourcing.

Launch year: 2025

Attribution

Case study provided to the AI Navigator courtesy of Dalberg Data Insights, Apolitical's content partner. "From Data to Guided Action: How AI helps Community Health Assistants turn data into confident, evidence-based supervision", Dalberg Data Insights, https://drive.google.com/file/d/1_JEVyZtbxBWFVFKtjeaQu03fWhluQIB5/view.