Polarized Engagement in Low-Tech Learning: Evidence from WhatsApp-Based Video Learning with The Apprentice Project

The Apprentice Project
Nov 27
3 min read

Across India, millions of students from low-income backgrounds depend on basic smartphones to learn. To make this learning more accessible, The Apprentice Project (TAP) uses TAP Buddy, a WhatsApp-based chatbot that shares short, co-curricular video lessons in coding, science, financial literacy, and art.

In the 2024–25 academic year, 100,000+ students used TAP Buddy. Within this larger batch, this study is based on TAP Buddy’s usage of 34,000 students from 13 states. From this usage, 58,880 records of student–video interactions were analyzed. These interactions provide a reliable window into how students choose to learn when technology is simple, data access is limited, and time is scarce.

Given this scale, the next logical question is:

What does student engagement actually look like in such environments?

Problem and Belief Addressed

Many EdTech models are built on a familiar idea: learners begin with interest and gradually lose attention as content progresses. This “linear decline” assumption influences how videos are created, sequenced, and measured.

However, the data from low-tech usage offers an opportunity to test whether this belief represents reality. Instead of assuming a universal pattern, the study examines a core question:

Do students generally lose interest over time, or do they show different, stable patterns of engagement?

To answer this, the study shifts focus from technology-driven design to learners’ actual choices.

A Lens of Agency, The Learner’s Choice.

This shift makes it necessary to view engagement as an expression of agency, not just a metric of performance. Students decide to start a video, pause midway, sample briefly, or watch fully depending on relevance, curiosity, and constraints like shared devices or data limits.

By framing engagement as a decision rather than a failure to persist, the study evaluates students as active participants shaping their own learning experience.

Recognizing this helps refine the research approach.

What This Study Examined?

The study aims to determine whether student engagement follows a single downward trajectory or whether multiple engagement modes exist. In simple terms, it asks:

Is engagement one straight line?
Or does it split into distinct behavioural categories?

To explore this, the research relies strictly on how much of a learning video students actually choose to watch.

Design and Method

A secondary analysis of TAP Buddy platform data was conducted using watch percentage as the core measure. Since this percentage reflects how much of the totala video a student viewed - not necessarily from start to end - completed, it was treated as a measure of depth of engagement, not just time spent.

To identify possible engagement groups, the study used:

Descriptive statistics to understand overall patterns
Gaussian Mixture Models (GMM) to detect clusters
Bimodality Coefficient (BC) to assess the strength of multiple modes

These statistical methods helped determine whether viewing behaviour forms one trend or multiple stable engagement types.

Key Findings

The results were clear. Engagement did not decline in a uniform pattern. Instead, viewing behaviour was multimodal (BC = 0.68), showing Three statistical clusters emerged, with two of them — low watchers and full watchers — being especially strong. The middle group is more spread out, showing partial viewing behavior that overlaps both ends.three distinct and stable engagement groups:

Engagement Group	Average Watch %	Behaviour
Low Engagement	~6%	Brief sampling or early exit
Mid Engagement	~53%	Partial viewing with early pause
Full Engagement	~99%	Near-complete or full viewing

The most striking observation is polarization: many students either watch almost nothing or complete the entire video. Fewer remain in the middle. This indicates that students are not gradually disengaging; they are choosing how deeply they want to participate.

This choice-based pattern signals important implications for design.

What does this mean?

The findings show that low-income learners using low-tech platforms are not passively dropping off; they are selectively engaging based on interest, time, and access constraints. This behaviour has clear design consequences:

The opening minute is decisive Engagement splits early. Strong, concise introductions can influence whether a student becomes a “Skimmer” or a “Finisher.”
Micro-modules can increase completion Shorter segments with clear takeaways help students finish learning quickly and feel rewarded without heavy data use or long viewing commitments.
Analytics should move beyond ‘watch time’ Instead of simply tracking duration, platforms should monitor:
- Voluntary replays
- Return behaviour across topics
- Completion linked to interest and choice

In summary

This study shows that when learning happens through simple tools like WhatsApp, students don’t just drop off, they make choices. Their engagement isn’t a slow fade; it splits sharply between quick sampling and deep, focused completion. Instead of assuming low-tech learners are passive or easily distracted, the evidence encourages us to view them as active decision-makers navigating real constraints. Designing for that agency, through stronger hooks, shorter modules, and smarter analytics, can turn small screens into meaningful learning spaces.