Coalition's NBN speed forecasts are on the right track

Bandwidth demand forecasts need to be much more subtle than simply punting on past trends. A fact that some fibre-to-the-premises advocates have conveniently forgotten.

The Vertigan Panel has, on behalf of the Government, recently published a cost benefit analysis of various approaches to the NBN. One input to their work was a forecast of Australian bandwidth needs prepared by my firm. We found that by 2023 the top five per cent of households would require at least 43 Mbps, and the median household would require 15 Mbps.

At first blush these numbers may seem low. But it’s worth remembering that most Australian households have just one or two people. A household where two people were both watching their own HDTV stream, each surfing the web and each having a video call all simultaneously, then (in part thanks to improving video compression) the total bandwidth for this somewhat extreme use case is just over 14Mbps in 2023.

Understanding future speed requirements is vital to planning the NBN, because if you need to provide very high speeds (over a 100 Mbps, say), then, with today’s technology at least, you need to deploy fibre-to-the-premises (FTTP), a much more expensive solution than fibre-to-the-cabinet. Perhaps surprisingly, given the criticality of this issue, ours was the first serious attempt to forecast Australian speed requirements. Billions have already been spent without such a forecast.

Given that our forecast calls into question the rationale for spending those billions, it is perhaps unsurprising that those who were instrumental in leading Australia down the path of an FTTP NBN have attacked our work. For instance, in a recent article Professor  Rodney Tucker claims that “it is simply wrong”.

Punting on past trends

The sole evidence he offers in his article for this claim is a comparison of our numbers to historic bandwidth numbers from Ookla, an internet measurement firm. The Ookla figures are (roughly) for capacity - the capability of the line in question. Our figures are for demand - how much a line is used. The two are completely different things, and to confuse the two is a gross error.

For example, Tucker uses the historic Ookla capacity trends to predict future needs for Australia. But in doing so he effectively says "because broadband speeds have been improving, we must need evermore speed in future". If you applied this logic to roads, you'd say "we've built some roads, therefore we must build some more roads", and before you know it, the entire country is covered with empty tarmac.

Bandwidth demand forecasts need to be much more subtle than this, not least because growth is likely to slow for a number of reasons. Once everyone in a household is online, more people can't create a need for more bandwidth; once people have devoted a certain number of hours per day to the internet, they won't have more time to give; video compression allows us to steadily reduce the bandwidth we need for a given video quality; video resolution can’t usefully outstrip the resolution of the human eye; and so on. Any serious attempt to forecast long term needs must take account of these effects. Simply punting on past trends (as Tucker does) just doesn't work - even if Tucker were looking at the right metric (which he isn't).

Too much faith in Ookla figures?

Tucker also claims that because our average number for 2023 is lower than the Ookla number for today, we are saying that "there is no need for any increase in download speeds". This is simply nonsense. Firstly, the Ookla numbers Tucker places such faith in are just not very accurate - they're based on a self-selected sample of people who happen to show up to do speed tests of their connection. In the UK there is accurate data from Ofcom (the regulator) of broadband speeds to compare to the Ookla figures. Over the last three years, Ookla's UK figures have, on average, overestimated by 54 per cent. In other words, the Ookla figures that Tucker uses for current Australian average bandwidth may be too optimistic, and current capacity much lower. Thus in reality our forecasts certainly would imply a need for an upgrade (though not to expensive FTTP).

Secondly (as Tucker himself acknowledges later in his article) we made it very clear in our forecast that networks shouldn't be built for the median user - by definition you'd be disappointing half the people if you did so. If you were building for (say) 95 per cent of users, our forecasts suggest you'd to roll out a network capable of providing 43 Mbps in 2023. Ookla's average line speed for Australia today is 16 Mbps. How Professor Tucker can look at those two figures and claim we're saying no increase is required is a mystery.

No-one is saying that broadband demand forecasting is easy or certain. As we said in our report, our forecast is "at best a mid-point projection", and we genuinely welcome a thoughtful debate. Sadly we're still waiting for that thoughtful debate to begin.

Robert Kenny is the co-founder of UK-based telecom consultancy Communications Chambers.