Education Technology

Policy Priorities for EdTech in LMICs

• Avoid hardware-first approaches. These have consistently failed and are expensive.
• Prioritise interventions with strong evidence of cost-effective impact:
  • Personalised computer-aided learning (CAL) (especially when supplementary).
  • Structured video instruction integrated with teacher support.
  • Parent information systems.
  • Remote tutoring (emerging evidence).
• Protect instructional quality. Many failures arise when EdTech disrupts teaching or eliminates teacher-student interaction.
• Build assessment and data systems that strengthen governance.
• Invest in iterative improvement. The most scalable EdTech models (SMS, WhatsApp, CAL) show large gains only after adaptation cycles.
• Treat EdTech as complementary – not a substitute – for teachers, governance reforms, and curriculum design.

EdTech: Presentation of key takeaways

At our launch event, Abhijeet Singh outlined everything you need to know about EdTech.

Education technology (EdTech) has expanded rapidly across low- and middle-income countries (LMICs), yet the evidence from the past two decades shows that technology itself is not typically an education input. Rather, EdTech is a delivery mechanism whose effectiveness depends on what is delivered, how, and what it displaces. Across the major categories of interventions – hardware access, instructional technology, parent/teacher information tools, and governance reforms – several consistent lessons emerge.

1. Provision of Hardware Alone Is Very Unlikely to Improve Learning

Across multiple countries and large-scale programmes, laptop or tablet distribution on its own produces no learning gains and, in several cases, negative effects. The OLPC experiments in Peru show no improvements in test scores in both the short and long run, despite substantial usage. Similar null or negative effects appear in Romania, Uruguay, Honduras, Costa Rica, and China. Hardware-heavy programmes – smart classrooms, laptops per child, computers without structured pedagogy – remain attractive to policymakers but are costly and have consistently underperformed.

Policy implication: Avoid hardware-led EdTech investments unless they are tied to instructional models with proven impact. Hardware should not be treated as a stand-alone learning input.

2. Internet Access Helps Only When Use Is Structured

The evidence on connecting schools or students to the internet is mixed. In Peru, school internet access generates meaningful learning gains, but only after several years, as teachers and students learn how to integrate it into schooling. Supervised, academically targeted use (e.g. access only to Wikipedia) improves learning, but general 3G expansion across countries is associated with declines in PISA scores – likely due to increased distraction.

Policy implication: Invest in structured, curriculum-aligned internet use rather than general connectivity. Gains depend not on access but on how the internet is integrated into teaching and learning.

3. Instructional EdTech Works When It Personalises Learning or Supports Teachers – Not When It Disrupts Classrooms

Supplementary, personalised computer-aided learning (CAL)

When technology supplements class time and provides tailored instruction, effects can be large and consistent across studies, especially for low-achieving students. 

In-class CAL that displaces teachers

Replacing regular instruction with CAL has mixed results. When poorly integrated, it can reduce instructional quality.When carefully implemented, personalized CAL integrated into school schedules generates substantial gains even when replacing 25–50% of class time. But gains fall sharply when adult supervision is reduced. 

Policy implication: Personalisation is powerful, but implementation quality is decisive. CAL should be adopted only with:

• Robust supervision and teacher support.
• Careful integration into instructional schedules.
• Protection against crowding out effective teacher-student interaction.

4. Remote Tutoring Appears Highly Promising – But Evidence Is Still Early

Phone-based tutoring during COVID-19 shows positive average effects across multiple countries, at very low cost. Yet effects vary widely and many trials overlap with atypical conditions (school closures, phone-based testing). Understanding how sustainable these effects are under typical conditions should be a high priority for programming. 

Policy implication: Remote tutoring is a potentially scalable, low-cost intervention, but needs longer-term, post-pandemic evaluations. Policymakers should treat it as promising but not yet fully mature.

5. Technology That Supports Parents Can Improve Engagement – But Not All Designs Improve Learning

SMS/WhatsApp nudges that provide information about children’s performance reliably improve parental monitoring and student outcomes (e.g. GPA, attendance). However, EdTech substitutes for high-touch parenting programmes (like IVR or message-only early childhood interventions) show uneven effects: usage may increase, but child outcomes often do not improve unless human support is added. Iterative refinement (e.g. enhanced personalisation, peer support for preschool workers) meaningfully improves efficacy.

Policy implication: Parent-targeted EdTech should be information-focused (attendance, grades, options) or paired with light-touch human interaction. Purely automated parenting interventions should be used cautiously.

6. Technology Can Dramatically Reduce Information Frictions in School Choice and Improve Decision-Making

Well-designed online choice platforms and targeted information – such as personalised guidance about nearby high-quality schools – shift parents toward better options and generate sizeable learning gains over time. Evidence from Chile shows families lack basic information about options; light-touch digital advisories meaningfully improve choices.

Policy implication: Integrate real-time, personalised information into school application platforms, especially for low-SES households. LMICs with emerging centralised admissions systems stand to benefit enormously.

7. Digital Assessments Can Strengthen the Integrity and Usefulness of Testing

Tablet- or computer-based tests reduce cheating, provide more precise measurement, and enable adaptive testing. Evidence from India and Indonesia shows that digital testing sharply reduces inflated test scores and improves the reliability of assessments used for policy. High-quality digital assessment systems also help power effective teacher incentives, report-card programmes, and performance-based management.

Policy implication: Invest in digital assessment infrastructure as a complement to improvements in instruction and governance. Reliable measurement is foundational for scaling nearly all high-return education reforms.

8. EdTech to Improve Teacher Practice Shows Potential but Requires More Evidence

Programmes that combine scripted pedagogy with technology (e.g. tablets used to deliver lesson plans) can generate very large effects, but evidence is still limited and often confounded with broader school reforms. Virtual teacher training is cheaper but risks dilution in quality compared to in-person models.

Policy implication: Use EdTech to augment, not replace, proven teacher-support models. Evidence on large-scale impact remains thin, so pilot and evaluate before scaling.

9. AI may substantially boost productivity of current EdTech models, but evidence is still lacking

It is likely that AI will boost the inner working of several software products (e.g. learning software through better personalisation). The evidence for this is best considered within the bucket of what the intervention actually does (e.g. individual computer-based tutoring or improved implementation follow-up) than as a separate class of ‘AI interventions’.Evidence on the use of LLMs is mixed and still at early stages.