Speed tests have become the thermometer everyone reaches for. Something feels off, you open the app, you get a number, and the number either reassures you or gives you something to shout at your ISP about. But a video call that stutters and a download that takes too long are two very different faults, and only one of them has anything to do with megabits.
Megabits measure width, not reaction time
Your line speed tells you how much data can move through the connection every second. That's it. It's the width of the road. A 200Mbps line is a wide road, and a wide road is genuinely useful when you're pulling down a big file or when four people are streaming at once.
Latency is a different measurement entirely. It's how long a single piece of data takes to get from your laptop to the server it's talking to and back again, measured in milliseconds. You can have a ten-lane highway and still take forty minutes to get across town if there's a robot at every second block. Width and travel time are separate problems.
For a video call, travel time is what matters. Your voice has to reach the other person and their reply has to come back, continuously, while you're both trying to have a normal conversation. Once the round trip creeps past about 150 milliseconds, people start talking over each other because the natural rhythm of turn-taking breaks down. Past 250 and it becomes genuinely tiring.
Some of that is geography and you can't argue with it. A server sitting in Johannesburg or Cape Town will typically answer you in under 20 milliseconds. A server in Frankfurt is around 150. Virginia is closer to 240, before anything has even gone wrong. Most of the big meeting platforms now run servers on this continent, which helps enormously — but only if your traffic is actually being routed to them, and that's your ISP's decision, not yours.
The wobble is what really ruins a call
Latency being high is annoying. Latency being inconsistent is what makes someone's face freeze mid-sentence while their audio keeps going.
The technical word is jitter. If your packets arrive 40ms apart, then 41, then 39, the call software copes without you noticing. If they arrive 40ms apart, then 200, then 15, then 180, the software has to guess what to do with the gaps. It either stalls and waits, which is the freeze, or it throws away what arrived late, which is the robot voice.
This is why Netflix is fine on the same connection that mangles your meetings. Netflix downloads the next thirty seconds of your show before you get there, so a wobble in the delivery is invisible — the buffer absorbs it. A live conversation has nothing to buffer with. Whatever arrives late is already useless.
Packet loss does something similar. Lose one packet in a hundred and you probably won't notice. Lose three and the picture goes blocky and the audio starts clipping words. Neither jitter nor loss will ever show up on a standard speed test, because a speed test is deliberately measuring the best your line can do in a short burst, not how steadily it behaves over ten minutes.
Nobody ever checks the upload
Here's the one I run into most often in South African homes. Plenty of fibre packages are asymmetric — 100Mbps down and 25 up, or worse on some LTE products. All the attention goes to the download number because that's the one on the marketing.
But on a video call, you are a broadcaster. Your camera feed, your microphone, your shared screen — all of that is going up, not down. And uploads are where the queue forms. One person in the house backing up photos to the cloud, or a laptop deciding it's a fine moment to sync a few gigabytes to OneDrive, can saturate a small upload pipe entirely. Everyone else's calls fall apart, and every speed test in the building still reads beautifully, because the download side is untouched.
What actually helps
Stop diagnosing calls with speed tests. If a meeting is misbehaving, open a terminal or command prompt and run a continuous ping to something like 8.8.8.8, then leave it running through the call. You'll see the spikes as they happen, and you'll have something concrete to show your ISP rather than a vague complaint about things feeling slow.
Get onto 5GHz, or plug in. The 2.4GHz band travels further through walls but it's the crowded one — every microwave, baby monitor and neighbour's old router is sitting in it. If you're taking calls all day from the same desk, a network cable to that desk is the single cheapest upgrade available to you.
Prioritise properly rather than policing your household. Telling everyone to stop using the internet while you're on a call is not a network strategy. A system that understands what each device is doing and gives conversation traffic right of way is a much easier life — it's a large part of why we run Plume, because that prioritisation happens on its own without anybody being told to log off.
And if calls are how you earn a living, one line is a gamble. Bonding fibre with an LTE link through BondAI means a wobble on one path doesn't take the meeting with it — traffic moves across both at once rather than waiting for a failure before switching.
If your calls keep breaking and nobody has been able to tell you why, get in touch. It's usually one of about four things, and it's usually not the number on your speed test.