Last year when Cisco released the 2012 mobile VNI forecast, I noted that they had been building castles in the air, and needed to put foundations under them. In particular I was concerned about substantial changes in the assumed share of offloaded traffic, which had changed dramatically between the 2011 and 2012 reports. Specifically, in 2011 Cisco had estimated that in 2010 21% of US smartphone and tablet traffic was offloaded (from mobile-connected devices, i.e. apparently excluding WiFi-only tablets) and that would increase to 30% by 2015. In 2012 they estimated that 49% of this traffic was offloaded in 2011 and that would decrease to 46% in 2016. Now in the latest report (actually in the VNI tool stats, not the report itself), Cisco estimate that:
47% of the United States’s mobile data traffic was offloaded in 2012.
66% of the United States’s mobile data traffic will be offloaded in 2017.
The amount of traffic offloaded from smartphones will be 64% in 2017, compared to 59% at the end of 2012.
The amount of traffic offloaded from tablets will be 78% in 2017, compared to 77% at the end of 2012.
So yet again we’ve had a dramatic change in assumptions about offloading, without much explanation or any retrospective view of whether the prior estimates were remotely accurate. Indeed, if the true amount of offload traffic (from smartphones and tablets combined) has increased from 21% at the end of 2010 to 49% at the end of 2011 to something over 60% at the end of 2012, it is far from clear that the share of offloaded traffic on these devices will hardly grow at all (1% p.a.) in the next five years. (Note that the large projected growth in overall offloaded traffic is an artifact of the change in mix, with total traffic becoming dominated by smartphones and tablets, so laptops play a much less important role).
In reality, users of Cisco’s own Data Meter Application already offload more than 80% of their traffic and may be rather more representative of the longer term smartphone market, at least in North America. One consequence of Cisco’s assumption that offload growth will slow, is that data traffic growth is projected to accelerate once again in 2013 compared to 2012 (e.g. North American traffic growth is projected to be 70% in 2013 compared to 64% estimated in 2012, and on a global basis growth of 78% is projected in 2013 compared to 70% estimated in 2012), which seems rather unlikely.
Even more perplexing are some of the individual changes in estimates between 2011 and 2012, which appear to relate to totally undocumented retrospective revisions to the 2011 data (which can be extracted by use of the VNI tool).
In North America, Cisco estimated last year that traffic would grow from 118,972 TB/mo in Dec 2011 to 259,253 TB/mo in Dec 2012 (a growth rate of 118%). Now Cisco estimates that traffic in North America was only 222,378 TB/mo in Dec 2012, with growth of 64% (per the VNI tool), so in other words, Cisco’s estimated 2011 traffic was ~136,000 TB/mo (14% more than the original estimated).
Incidentally, traffic in the US was estimated to have grown by only 62% (almost exactly as I predicted from the CTIA data last October) to 206,854 TB/mo in Dec 2012. However, in absolute terms Cisco’s number for the US appears far too high: extrapolating CTIA’s 6 month statistics (based on real data from almost all US mobile operators) indicates there should be no more than 150,000 TB/mo of mobile data traffic in the US by Dec 2012.
Similarly, the Cisco numbers for Western Europe indicate growth of 44% between Dec 2011 and Dec 2012 (to 181,397 TB/mo), compared to an original projection of 103% growth, but the 2011 estimate has been retrospectively revised to about 126,000 TB/mo, or 31% lower than the 180,370 TB/mo originally given in the 2011 report.
In addition, the previous assessment that significant traffic would be generated by “home gateways” has been completely erased and significant changes have been made to all of the regional traffic totals for 2011, as given in the table below.
UPDATE: Apparently a later version of the February 2012 Mobile VNI forecast (not that linked below) corrected a typo in the Middle East & Africa regional total, so that the discrepancy between the sum of the regional traffic and the global total traffic (597,264 TB/mo) was eliminated. The chart above has been updated to incorporate this modification.
Thus, though Cisco tries to direct our attention to future growth potential (not least by deleting access to previous reports – for your benefit, here are copies of the reports published in February 2010, 2011, 2012 and 2013 respectively), I’m hugely concerned about whether the traffic data that is supposed to be covering historic periods is at all reliable, given the enormous scale of unstated retrospective revisions to these numbers, and the lack of correlation with much more detailed studies such as that by CTIA. If the current data isn’t accurate, then its hard to see how much reliance can be placed on Cisco’s forecasts for future periods.
Nevertheless, Cisco appears to have bowed at least partially to reality, acknowledging that 2012 growth was “slower than expected in some regions” and side-stepping its prior claims of a “data deluge“. I wonder if those, including the FCC Chairman, who have made so much fuss about the “spectrum crisis”, and have repeatedly cited Cisco’s over-optimistic projections to justify their argument, will now do likewise. Perhaps they might even give less credence to these numbers, especially given the apparent inaccuracies in Cisco’s estimates.