A course that runs beautifully on office wifi can be unusable on an entry-level Android phone on a two-bar signal outside a taluka town. Low-bandwidth learning is not a special case in India. For a large share of your learners it is the normal case, and it is where completion quietly dies.
The failures are specific. A video buffers for forty seconds and the learner leaves. A page carries megabytes of images before a word of text appears. A learner answers eighteen questions, the signal drops on the nineteenth, and the attempt is lost.
This post works through each of those: video size and adaptive streaming, offline downloads, image weight, saving quiz answers when the network drops, and how to test on a real mid-range phone rather than on your laptop with the network throttled.
What low-bandwidth learning means in practice
Start with the official picture. The Telecom Regulatory Authority of India reported that total internet subscribers in the country rose from 1,028.61 million at the end of December 2025 to 1,092.79 million at the end of March 2026, and that average wireless data usage stood at 26.70 GB per wireless data subscriber per month during the January to March 2026 quarter. TRAI defines broadband as a data connection with a minimum download speed of 2 Mbps to an individual subscriber from the service provider’s point of presence.
Two readings of that. Data volume is not the scarce thing it was: 26.70 GB a month is a real allowance. What is scarce is a steady connection at the moment the learner sits down, and a phone with room to hold anything. So design for a connection that is intermittent rather than absent, on a device with very little free storage.
Video is most of the problem
Video is usually most of a course’s weight by a wide margin. The fix is not to remove it but to stop shipping one large file to everyone.
Adaptive streaming, plainly
Adaptive streaming means the video is cut into small segments and encoded at several qualities. The player watches the transfer rate and, as Mozilla’s developer documentation puts it, if it looks like it is not keeping up, the player drops down to a lower bandwidth and lower quality stream. HTTP Live Streaming, specified in RFC 8216, works the same way: a master playlist lists variant streams at different bit rates and resolutions, and clients switch between them to adapt to network conditions.
Practically, encode each lesson at three or four rungs, say 240p, 360p, 480p and 720p, and let the player choose. A learner on a weak signal gets a watchable 240p instead of a spinning circle.
What a lesson actually costs
The arithmetic below is not a measurement, just the bit rate multiplied by the running time, using decimal megabytes. It is worth putting in front of whoever signs off the course.
| Video bit rate | 10-minute lesson | 20 lessons | Share of 26.70 GB month |
|---|---|---|---|
| 200 kbps (audio plus slides) | 15 MB | 300 MB | About 1 per cent |
| 400 kbps (240p) | 30 MB | 600 MB | About 2 per cent |
| 800 kbps (360p) | 60 MB | 1.2 GB | About 4 per cent |
| 1,500 kbps (480p) | 112.5 MB | 2.25 GB | About 8 per cent |
| 2,500 kbps (720p) | 187.5 MB | 3.75 GB | About 14 per cent |
A 720p-only course of twenty lessons eats roughly a seventh of the average subscriber’s monthly data. The same course at 360p costs less than a twentieth. For a talking head over slides, nobody learns less at 360p.
Cheap wins before you re-encode anything
- Cut lessons to 6 to 10 minutes. Shorter files start faster and a dropped connection loses less.
- Do not autoplay. Let the learner press play when ready.
- Show a still image as the poster frame rather than preloading the video.
- Publish a text transcript below every video. It is searchable, it is tiny, and it is the whole lesson for someone who has no bandwidth at all.
- For a slides-and-voice lesson, consider audio plus images instead of video.
Downloads for offline study
The single most useful feature for a low-bandwidth learner is the ability to take the lesson home on wifi and study it on the bus. Design the download deliberately.
- Offer download per lesson, not per course. A 2 GB course download will fail on a phone with 1 GB free.
- Let the learner choose quality at download time, and default to the lower rung.
- Show the size before the download starts, in MB, next to the button.
- Resume broken downloads instead of restarting them.
- Keep the transcript and any PDF handouts in the offline bundle.
- Record progress locally while offline and sync it the next time the learner is online.
- Let the learner delete a downloaded lesson from inside the app when storage runs out.
Image weight
Images are the quiet cost. A course author drops a 4,000 pixel wide photograph from a camera into a lesson, and every learner downloads all of it to view it 360 pixels wide.
- Resize to the size the page actually displays, then compress. A lesson image rarely needs to be wider than 1,200 pixels.
- Serve modern formats such as WebP or AVIF, with a JPEG fallback for older devices.
- Lazy-load images below the first screen so the text renders immediately.
- Replace screenshots of text with real text. It is smaller, it reflows on a narrow screen and it can be read aloud.
- Set a weight budget per lesson page and have the authoring team check it, the way a print team checks page count.
Saving quiz answers when the network drops
This is the failure that makes learners give up entirely, and it is the most avoidable. An assessment should never depend on the connection holding for thirty minutes.
- Save each answer as it is given, not the whole paper at the end.
- Keep a local copy on the device so answers survive a refresh, a call or the browser being killed.
- Retry silently when a save fails, and show a small, honest indicator of what is saved and what is pending.
- Let the learner resume the same attempt from the same question after a disconnection, with the timer handled by a stated rule rather than silently.
- Download the question set at the start where the design allows it, so navigation between questions needs no network.
- Never discard a submitted attempt because the final call failed. Queue it and reconcile.
Randomised question order and timers still work under this design, and they matter more when the exam is unproctored. We have written separately about building an assessment engine that measures learning rather than memory.
Test on a real mid-range phone
Throttling Chrome on a developer’s laptop tells you about the network. It tells you nothing about a phone with a slow processor, a small screen and almost no free storage.
- Buy two or three budget Android phones of the kind your learners actually use, and keep them in the office.
- Test on mobile data outside the building, not on office wifi, and preferably away from the city.
- Fill the phone’s storage to nearly full before testing downloads.
- Time three things with a stopwatch: login to dashboard, lesson page to first readable text, and play button to first video frame.
- Switch flight mode on in the middle of a quiz, wait, switch it off, and confirm nothing is lost.
- Rotate the screen mid-lesson and mid-quiz. Layouts break there more often than anywhere else.
- Run the same tests after every release, as a fixed checklist, not when somebody complains.
Frequently asked questions
Should we just make courses text-only for low-bandwidth learners?
No. Give everyone the same course and let delivery adapt: adaptive video, a transcript, downloadable lessons. A stripped-down version for rural learners is a second-class course, and they notice.
Is 4G enough for video lessons?
A good 4G connection is comfortable for 360p and often for 480p. The trouble is variation during the day and at the edge of coverage, which is exactly what adaptive streaming handles and a single fixed-quality file does not.
How much data should one hour of learning cost?
Set a target and hold the authoring team to it. At 800 kbps, an hour of video is about 360 MB by simple arithmetic; at 400 kbps it is about 180 MB. Publish whichever number you choose so course authors design against it.
Do offline downloads break our completion records?
Only if progress is recorded solely on the server at the moment it happens. Record it on the device with a timestamp and sync on reconnection, and the record is as sound as an online one.
Building for the connection your learners have
A low-bandwidth LMS is mostly a set of decisions taken early: lesson length, video rungs, image budget, download design and how an assessment saves. Quipu LMS is mobile-friendly and self-hosted on your own servers, with an assessment engine, progress analytics and SCORM import, so these choices stay in your hands rather than a vendor’s.
