What happens to a teacher’s month when the paperwork gets lighter? In 2026, an independent team tested exactly that with 273 teachers in Machakos, Kenya. The clearest result was in the quality of teachers’ lesson plans. But the change teachers talked about most was how they felt.
In Kenya, the shift to the Competency-Based Curriculum (CBC) asks teachers to prepare well-structured Schemes of Work and lesson plans, formulate clear learning outcomes and Key Inquiry Questions, and use assessment approaches that support learning. Many do this in classrooms of 50 or more students, with uneven connectivity, uneven access to devices and limited professional support. In practice, much of that preparation lands in breaks, in the evening and at home.
Amal was built for this reality. Through WhatsApp and a lightweight web platform, teachers can draft lesson plans, schemes of work and assessments in minutes, in line with the curriculum. Today more than 30,000 active teachers use Amal Labs’ platforms across East Africa and the wider Global South, and more than 7,000 teachers have been reached in Kenya via WhatsApp. The question for this study was a harder one than reach: does Amal actually change teachers’ professional practice and experience?
The evaluation was led by Education Design Unlimited (EDU) under Jennifer Cotter Otieno, its Founder and CEO. It used a school-level cluster-randomised controlled trial: schools were randomly assigned either to receive Amal Labs with training, or to continue preparing as usual. 300 teachers took part at baseline and 273 were retained at endline, a 9% attrition rate. Impact was estimated with a difference-in-differences approach and checked against classroom observations, interviews and focus groups. Teacher quality was assessed directly from the documents teachers submitted, not only from self-report.
Four results from a single month
+1.56 points: Lesson plan quality on a 0 to 4 rubric (p < 0.001), the largest effect of any outcome measured. Every rubric dimension improved significantly.
−39.2%: Time spent preparing summative assessments (p < 0.001). Scheme of work preparation time fell by 27.1% (p = 0.009).
+1.04 points: Teacher wellbeing on a five-point index (p < 0.001), a substantial improvement linked to reduced preparation pressure.
+0.21 points: AI skills adoption index (p = 0.042), driven by gains in lesson planning and assessment.
Alongside better lesson plans came a substantial gain in how teachers felt. Teachers described spending less time on preparation and more time with their families and their learners. As one teacher put it in a focus group:
“It takes me only five minutes to generate a lesson plan, unlike before when I used to spend about 15 minutes preparing one lesson. It has given me more time with my family because I no longer have to take as much work home.”
Teacher, focus group discussion
The evaluation points to a specific mechanism. The wellbeing gain was linked mainly to reduced preparation-related pressure, not to a lighter overall workload. Marking and administration stayed largely unchanged, so Amal should be understood as easing one recurring source of pressure, the routine burden of lesson preparation, rather than as a solution to every demand on a teacher’s time.
Amal is designed as a teacher-in-the-loop tool. 91% of treatment teachers used the platform, and 96.5% said they were likely or very likely to keep using it. But most did not use outputs exactly as they came: 69% made minor changes before class, 4% made major changes and only 15% used the draft as generated. Teachers treated Amal as a starting point for their professional judgement, not as a finished product.
The results also show where that judgement matters most. Lesson plans are written daily and follow a clear structure, and quality rose by 1.56 points. Schemes of work are written once a term and showed a smaller gain that was not significant overall, though one dimension improved significantly. Assessments depend on what was actually taught and on each learner’s ability, and showed no significant change within the month. The more context-sensitive the task, the more important teacher review becomes.
We want to be as clear about the open questions as about the results. Only 42% of teachers said they often or always verified AI-generated content, so building verification into the product is a priority. Poor internet connectivity was the most common barrier, reported by 77.9% of teachers, which is why low-bandwidth and WhatsApp-based access are core design requirements rather than extras. Peer learning carried much of the adoption: 65.5% of teachers turned to colleagues for in-use support, but peer support did not replace formal training, and some teachers still needed structured help. And because the study covered one month, it cannot yet show whether these gains last or reach learners.
What it does show is a result worth building on. Within a single month, a well-designed AI tool, paired with training and peer support, measurably improved the quality of the work teachers do most frequently, and made a recurring part of their work faster and less stressful. A more useful question than whether Amal works is under which conditions AI-supported teacher tools deliver lasting improvements: curriculum-grounded tools, continued professional development, peer learning, active teacher verification, supportive school leadership and reliable connectivity.
Over the coming months, Amal Labs will refine the platform for 50,000 teachers, strengthen teacher verification and low-bandwidth access, and work with content and distribution partners across Sub-Saharan Africa. If you are a university, foundation, or NGO working on teacher development, we would like to build this evidence base with you.
