I'm deeply skeptical about a lot of the rhetoric about "mobile data explosions" and "tsunamis". In particular, I believe that a lot of the forecasts are unrealistic and often self-serving. Cisco's VNI is the best-known, but many other vendors (eg Ericsson, Huawei) and analysts put out their own data as well.
The predictions of several more years of 100% traffic growth seem a particularly poor fit, given that numerous operators (eg Vodafone) have reported notable falls in growth rate, often to below 50%pa annualised in developed markets with tiered/capped plans.
Clearly, suggesting that networks might get overwhelmed is a great way to suggest that operators should "buy more kit". There's also usually a particular focus on video and the percentage of traffic it makes up.
I also think that overstated & misrepresented data traffic forecasts are mis-used by the operators and industry bodies, especially when it comes to talking about "spectrum shortages", or perhaps the onerous effects of Internet content that should justify non-neutrality of service provision. There is a bigger battle being fought here, with the telecoms industry trying to claim spectrum previously used for TV or government functions. This has two functions - it makes network expansion simpler than using alternative approaches, and it also reduces the strategic and competitive power of the broadcast industry, which is a gating factor on telcos' IPTV opportunity.
SIGN UP TO RECEIVE THIS BLOG BY EMAIL
So it is easy to understand why there is a good reason to high-ball estimates of future mobile data growth. There is also a desire to try and create - or at least influence - self-fulfilling prophesies about the role of mobile broadband.
In any case, market forecasting is imprecise, because often the market itself is dependent on decisions made in the light of people reading them. To be accurate, you'd really need to forecast what new actions will occur as a result of your forecasts being believed, which is clearly a circular argument.
But in any instance, these discussions generally overlook numerous inconvenient issues that ought to be front-and-centre for mobile data before we run to the hills (or the regulators) from the "tsunami":
- Tiered/capped pricing seems to "work" very well in limiting data consumption and congestion, especially if users have a "fuel gauge" and some idea of what activities burn the most of their quota
- Gross measures of traffic "tonnage" don't translate either to costs or congestion. It's traffic in busy hours and busy cells that matters. An extra 10GB of video downloaded at 3am in a rural cell is essentially free. An "offpeak" dataplan might increase reported traffic volumes but have no impact on costs or congestion. In fact, it might generate incremental and very profitable revenue by increasing capacity utilisation during quiet periods.
- For many networks, signalling load (against both the RAN and the core network) is the problem, not data tonnage. Multiple short bursts of data or "pings" clog up the network in different ways to a single, consistent stream. But it's harder to measure and bill for signalling, so it tends to get ignored
- Smaller cells give greater capacity density, at a lower price. They allow better spectrum re-use, reducing the need for new bands. In the long run, we get much more extra capacity by reducing cell size rather than adding extra radio channels - but this conflicts with the desire to grab more spectrum from alternative/competing users.
- Other new technologies and processes will improve network efficiency too: beam-forming, better sectorisation, MIMO and so on. But these are less well-proven than simply adding extra bands.
- "Video" isn't an application, it's hundreds. Amalgamating 500+ different applications and services under a single banner is completely arbitrary and meaningless. It's like saying the web is disproportionately dominated by the colour blue, and it should therefore be treated differently. (It might be green, I don't know).
- The dynamics of demand for mobile broadband are over-simplified. Much of the historic growth has been from "more users" rather than "more use per user". There are additional issues (discussed below) about coverage, device capability and so on.
- Demand growth and capacity growth are not directly linked - especially because "capacity" is impacted by numerous factors such as backhaul as well as radio-network scale. We also find that some base stations are "congested" because they haven't yet been upgraded to the maximum number of radio carriers already available.
- The dynamics of the mobile data market differ for post-paid and pre-paid users. As PAYG (which makes up the bulk of the planet's mobile customer base) becomes predominant, the idea of a monthly allowance will shift to a more usage-linked model. Early evidence suggests that PAYG data users - with smartphones - consume much less data than those on fixed plans. This is not factored into most forecasts.
- Some forecasts start from 2009 or 2010, therefore building in a huge initial leap from a low base. Ignore anything that doesn't use 2011 or 2012 as a start year, as otherwise the statistics will get swamped by vague and patchy measurements of historic data.
NEW REPORT - 10 REASONS WHY 1-800 TOLLFREE APPS MODEL WON'T WORK
We're already satisfying much of the latent demand
I'd argue that in developed markets such as the US and UK, we are already at 50%+ of potential mobile data usage saturation given TODAY's devices, data-plans, apps and user behaviour. Almost anyone who really wants a smartphone already has one.
Many people who want cellular-connected tablets or laptops already have them too. Sure, there are still a few demographics that want them but can't afford them, but that group is diminishing rapidly. The remainder are mostly "laggards" - folk who might like a new device, but who are likely to be "unenthusiastic" in usage behaviour, at least in comparison to those with a 4S or S3 or L900 in their hands 24x7. There are surprising numbers of mobile data "refuseniks" whose attitude is unlikely to change.
If you magically increased the smartphone penetration of all developed countries to 100% right NOW, I'd be surprised if it added more than 50% to tomorrow's data consumption stats.
NEW REPORT - 10 REASONS WHY 1-800 TOLLFREE APPS MODEL WON'T WORK
A thought experiment
Consider a fictional place where 50% of people have mobile broadband (smartphones with data plans, plus some fraction have 2nd/3rd devices), and 50% don't have MBB, either because it's too expensive, or because they live outside coverage, or they're just apathetic.
Let's say that the 50% of current data users are using 1GB/month as an average (mean). There's a mix of capped and uncapped plans, some people are heavy users (with one or two devices), others are more parsimonious, or perhaps just use WiFi a lot.
What is the "unconstrained demand" from these people at around current data-plan prices? I think we can now assume that the mobile broadband marketplace is now pretty efficient at giving people roughly what they want, at roughly the price they're prepared to spend. I'd be surprised if the true unconstrained demand from existing mobile data users would be much more than 50-100% higher than today. (Yes, if we dropped the average price massively there'd be an elasticity effect and demand would rise, but let's leave that option for a moment).
What are those constrains?
Thinking about Mobile Data Demand Constraints
A good way to think about this is to consider "what would happen to overall data traffic if we removed certain constraints?". This is often a counter-intuitive thought process because it involves going beyond the raw statistics and thinking about the real world and user behaviour.
So it's tempting to say that going from 50% penetration to 100% overnight would result in an instant doubling of traffic. But actually, it wouldn't because the remaining 50% would be much less enthusiastic users than the early adopters, especially on Day 1. Similarly, if we had perfect 3G/4G coverage tomorrow, we'd see traffic growth but not that much overall because all the busiest areas are already covered. What's left are big zones of occasional use (rural), and quiet corners of some indoor spaces.
This type of analysis is inherently much more complex, and goes beyond most statisticians' comfort zones. But it's a critical application of common sense. It's a sanity-checking phase too, that often seems absent in a lot of the mobile data forecasts I've seen.
For the thought experiment, let's consider I've got a magic wand to remove constraints. Each time, I'm going to leave the other variables untouched, especially price. What might happen?
1) Device penetration - if you gave everyone a smartphone tomorrow with a dataplan they could afford, plus cellular tablets/dongles at a pro-rata penetration to the early-adopter base, I expect we'd get around another 40% of traffic. (Heavily dependent on existing penetration of smartphones, eg 30% vs 50% - by the time some people read this post, we'll be nearing saturation anyway as it's moving so fast). In fact, the figure might be much lower - maybe just +20-30%, because most of those users will be prepay subscribers who tend to have lower data consumption anyway.
2) If we had perfect cellular coverage everywhere, I reckon we'd get a 20% uplift in aggregate traffic. Network planners aren't stupid - they know where the demand is, and the economics of satisfying it. Providing coverage to every mile of road and rail, or to every small village would definitely be nice - but it doesn't compare to a a big metro area when added together.
3) Now for a biggie. If we improved network speeds to 4G-type rates, with better latency, what would that do to user behaviour? More video streaming? Probably. More web use? Sure. More cloud-based apps? Perhaps. This is a tough one to predict, but we see some indications from people who move from 3G to LTE (although some of that is about getting upgrading to a new & better device rather than having a better network). Some stats say that LTE users typically use 50% more data than 3G users. BUT, that is skewed by early adopters switching first. But given today's apps and data plans and user expectations - which are often met pretty well on 3G after all - I'd say 40% is reasonable if the speed constraints were removed.
4) Device performance is another variable here. A lot of people have quite old devices that are slow, clunky, have low-res screens or are otherwise constrained, irrespective of network capabilities or coverage. But how much of a big deal is this really? Again, most of the real heavy users and enthusiasts do have the latest devices. If you waved the proverbial magic wand and upgraded all the old 3GS's and Galaxy S2's and assorted BlackBerries to today's state of the art, what would happen? Not much I reckon, again if all other variables were kept constant. (There's a bit of a co-dependency with LTE availability as noted above, though). I reckon that across the user base as a whole , we'd see perhaps a 30% uplift in data usage.
Now let's bring all these together to see what might happened if we removed the constrains (except price, and again bear in mind this is with today's typical apps and behaviours):
Device penetration = 1.4x (maybe lower)
Coverage = 1.2x
Network speed = 1.4x
Device performance = 1.3x
Multiplying through, we get an estimate of unconstrained demand = 3x today's constrained demand
But some of this is - in all reality - never going to happen. We're always going to have a spread of device ages and capabilities. We're never going to get 100% coverage. Some people will never use Facebook or Netflix or Dropbox, no matter how fast the network. PAYG prepay users will use less data for various reasons. And some people will hold onto their Nokia 6310 from 2003, even if you try and bribe them with the latest smartphone for free.
So in the confines of this thought experiment, we're probably addressing 50% of *current* unconstrained demand for mobile data, at current prices.
That's probably true of a bunch of other industries as well. I'd guess that we're probably at around 50% of unconstrained price-constant demand for anything from flights (people don't have enough holiday time, or are scared of flying etc), or even unconstrained demand for beer (can't drink at lunchtime before a meeting, too young, health/religious issues etc).
It also passes the "taste test". Most people don't spend all day moaning about how they're only getting a fraction of the mobile data they want. Generally, apart from a few minor gripes (coverage mostly, plus congestion in some hot-spots), people seem pretty happy that their mobile data demands are being met.
Obviously, this is macro-level stuff. Specific places (eg Olympic Park) clearly see much faster growth in demand as there will be localised drivers. Also, the calculation in the thought experiment above will vary by country a lot too - there are different levels of network rollout, smartphone/dataplan adoption and so on. India, for example, is starting from a much lower base for traffic, and so many of the variables will be considerably higher to work out latent demand.
Mapping future demand
So. Let's say that with current devices, networks, apps, behaviour and pricing, we are dampening consumption of mobile data by a factor of maybe two from the theoretical realistic demand in today's developed markets. That's a useful number to bear in mind, as it means that:
Any future mobile data traffic growth is primarily going to come from new demand not satisfying current latent demand
That's important in technology. It's often said that "usage always expands to fill the capability available", or "build it and they will come", but that's not actually true. The reason that computer processor speed has always been exploited (and so quickly) has been that companies such as Intel have spent lots of time and resources on "demand creation", seeding developers with new technology, running marketing programmes and so on.
So where are all these forecasts for 10x, 20x - even 1000x - growth for mobile data coming from?
Is it just the mobile industry's normal ridiculous arrogance (almost a sense of entitlement about growth) and its propensity to ignore lessons painfully learnt elsewhere in the technology industry?
Well, firstly there's still a lot of untapped growth in emerging markets, although it again needs to be born in mind that at current network costs and dataplan prices it is likely that data use will be lower per-capita. A $2 data prepay ARPU is not going to give 100% coverage, $300 subsidy of an iPhone 4S, and a 1GB per-month plan. In general, that part of the market will be using cheaper/less-capable devices, on thinner/slower networks, with more restrictive tariffs than elsewhere. They will also likely adopt different behaviour to "squeeze more from less".
And, as mentioned above, we'll also see growth in smartphone use among late-adopters in developed markets.
So while this means we should see mobile data user numbers continue to grow rapidly, this will also paradoxically drive down the average data consumption as heavy users (who may well be using more data year-on-year) will get diluted by newer and ever more numerous lighter users.
I have not seen this mathematical inevitability called out on any forecasts, yet it happens in virtually all markets as we see a progression of maturity. A grandparent in rural Bolivia getting their first smartphone is unlikely to be using Facebook and streaming video 24x7 on Day 1.
So, where else is the extra usage going to come from? Various sources are possible:
- Even faster / better networks making apps more usable
- Device improvements like bigger screens
- New apps and services (eg mobile cloud-based offers)
- More cellular devices per person (either "attach rates" of personal devices like 3G tablets, or others like M2M that bump up the total)
- Behaviour changes meaning greater time spent on mobile apps
- Lower prices driving elasticity (eg through changing behaviour faster/further)
- Better structured data-plans driving off-peak usage volumes
- Switching from using fixed broadband or WiFi to LTE
NEW REPORT - 10 REASONS WHY 1-800 TOLLFREE APPS MODEL WON'T WORK
Many of these are interlinked, obviously - if the network is faster, it enables new apps, and alters peoples' behaviour.
It is also important to note some downward pressures on users' average consumption of mobile data:
- Better devices and OS's that compress data (eg similar to BlackBerry, Opera Mini & Nokia Asha which route "optimised" data via a server), or third-party software like Onavo's
- Various types of network-based optimisation and compression, especially for streamed video
- More use of adaptive applications (eg HLS-encoded video) which self-optimise to network conditions, or more efficent codecs
- Substitution of 3G/4G data with WiFi, either as true "offload", or (much more importantly) user-driven preference for accessing private WiFi, usually for free
- Older devices being retired (for example, I'm about to cancel my 3G dongle contract as I never use it - I get WiFi almost everywhere I want it)
- App developers becoming wiser and more parsimonious about how their software consumes data, especially if they get better development/testing tools, or are "shamed" into it in appstore ratings
I don't have quantitative forecasts for all of these drivers and constraints. But some appear to me to be especially important:
- More usage per person driven by better apps and devices, and behavioural change. Let's tackle the latter first. I honestly don't think that the average mobile data user is going to increase their time spend on mobile devices by another 2x, 3x, whatever. We're close to saturation on that one already. Better apps and devices? Definitely. I agree that there will be more video and cloud-app usage. I can certainly see scope for an extra 2x or 3x over a 5-8 year period. However, the swing factor here is likely to be tablets and the evidence suggests more extra usage will be WiFi-based.
- WiFi is to my mind the biggest "decelerant" here. Operators are (often) trying to do their own controlled offload of traffic, although I still believe that most examples are going down the wrong path of "seamlessness" with things like ANDSF and Hotspot 2.0. However, that is becoming less relevant anyway, given the huge global explosion of non-operator WiFi and increasing sophistication of users in exploiting it. Partly because of data pricing and caps, users are actively seeking "free WiFi" wherever they go, and becoming adept at using it - especially the high-end power users that normally generate the most traffic. Unless we see lots of WiFi congestion (possible), that move now seems irreversible. Recent Ofcom data shows that more people are becoming "WiFi-primary", just using 3G/4G where they have to.
- More devices per person. Yes, this is going to rise, even though most tablets will likely remain WiFi-only. We'll see various new mobile-enabled gadgets in all walks of life. M2M can, however, be dismissed as a major traffic driver as the vast bulk of products are low-bandwidth. Despite a few high-consuming categories like digital signage or in-car telemetry, there are none that obviously have the scope to scale to billions of units. Against phones and to a less degree PCs, tablets & MiFis, they are lost in the noise.
- Despite better uplink speeds and slow rollout of fibre, I don't see signs of much shift from ADSL/cable to mobile-only. Coverage limitations and need for IPTV and WiFi are likely to keep fixed broadband largely protected except for a few niches.
- I do see quite a lot of traffic growth being driven by "off-peak" data plans. Marketeers and their billing systems are becoming smarter. However, this is essentially irrelevant from the point of view of network capex, spectrum needs and so on.
The price elasticity issue is a difficult one to address. If 4G was completely free & ubiquitous, you'd find people getting LTE-enabled 42" TVs in their home, running HDTV over cellular even when there is nobody in the room. In that case, even the most bullish forecasts would still be too low. Clearly, that's an extreme example, but various other milder scenarios are possible.
This is the paradox in all the forecasts: there seem to be no clear assumptions on pricing. Cisco's VNI projections probably are achievable, but only if mobile data is priced at levels that nobody would make any profit from it. As we found with flatrate, it's easy to drive usage if you're throwing away money.That said, we may see users being encouraged to migrate to LTE with offers of larger data bundles, which effectively reduces prices.
Taken together, my belief is that the bulk of forecasts are over-hyped. I think that too many of the projections are being made without considering how devices, apps or users are actually changing. There also seems little recognition of the "dilution of the average" as late-adopters, prepay users and developing-market subscribers bring down the headline "per-user" numbers.
Quantitative estimates and forecasts
I haven't done a full spreadsheet model analysis & forecast, but since I'm sure everyone's going to ask me anyway, I'll "take a punt" on overall global data traffic volume growth rates, based on existing published stats & bearing in mind the qualitative factors described in this post. It's worth noting that most spreadsheets I've seen are long on detail but a bit short on commonsense, especially with assumptions on per-user data growth continuing inexorably, despite the "average dilution" effects described here. There is also a general assumption that "subscription" remains the main business model, rather than more ad-hoc usage through PAYG. (Although, obviously, the overhyped 1-800 tollfree models won't be happening)
My prediction for overall global data traffic growth (Indexed to full-year 2011) is as follows:
2012: 70%
2013: 55%
2014: 45%
2015: 40%
2016: 35%
2017-2020: CAGR 30%
In other words, 2011-2016 growth is 7.2x - and in developed markets, probably more like 4x, and by 2015 we should see growth rates broadly comparable to that in fixed broadband in developed markets.
Given advances in technology, this should be "sustainable" with ongoing evolution of networks to LTE, especially small cells & better radio technology like beamforming. Some of the traffic growth is likely to be off-peak, as pricing & policy becomes smarter.
This means that growth will be more of a regular tide than a "tsunami", and definitely not something for operators to panic about - and regulators need to learn to be skeptical of shrill demands for more spectrum.
SIGN UP HERE TO RECEIVE THIS BLOG BY EMAIL
NEW REPORT - 10 REASONS WHY 1-800 TOLLFREE APPS MODEL WON'T WORK
Some comparisons with other forecasts:
Cisco VNI expects 18x growth from 2011-2016 (vs. 7.2x from my best estimates)
Ericsson predicts 15x growth 2011-2017 (me: 9.4x for that period)
Reading from a chart in this ALU presentation suggests 30x from 2010-2015, or about 10x from 2011-2015 (me: 5.5x)
My peers over at ABI reckon 8x from 2012-2017 (me: 5.5x for that period)
Informa predicts 10x for 2011 to 2016 (me: 7.2x)
Morgan Stanley gives scenarios for 5x, 9x & 23x for 2011-2015 (me: 5.4x)
Edit, 14th September - AnalysysMason makes very similar arguments to me (and has been similarly pessimistic to me for a while). They reckon 5.5x global growth for 2012-17 (me: 5.6x) and "dangerously low" for Europe. I agree - overcapacity is the risk, not a mythical "spectrum crunch"
The predictions of several more years of 100% traffic growth seem a particularly poor fit, given that numerous operators (eg Vodafone) have reported notable falls in growth rate, often to below 50%pa annualised in developed markets with tiered/capped plans.
Clearly, suggesting that networks might get overwhelmed is a great way to suggest that operators should "buy more kit". There's also usually a particular focus on video and the percentage of traffic it makes up.
I also think that overstated & misrepresented data traffic forecasts are mis-used by the operators and industry bodies, especially when it comes to talking about "spectrum shortages", or perhaps the onerous effects of Internet content that should justify non-neutrality of service provision. There is a bigger battle being fought here, with the telecoms industry trying to claim spectrum previously used for TV or government functions. This has two functions - it makes network expansion simpler than using alternative approaches, and it also reduces the strategic and competitive power of the broadcast industry, which is a gating factor on telcos' IPTV opportunity.
SIGN UP TO RECEIVE THIS BLOG BY EMAIL
So it is easy to understand why there is a good reason to high-ball estimates of future mobile data growth. There is also a desire to try and create - or at least influence - self-fulfilling prophesies about the role of mobile broadband.
In any case, market forecasting is imprecise, because often the market itself is dependent on decisions made in the light of people reading them. To be accurate, you'd really need to forecast what new actions will occur as a result of your forecasts being believed, which is clearly a circular argument.
But in any instance, these discussions generally overlook numerous inconvenient issues that ought to be front-and-centre for mobile data before we run to the hills (or the regulators) from the "tsunami":
- Tiered/capped pricing seems to "work" very well in limiting data consumption and congestion, especially if users have a "fuel gauge" and some idea of what activities burn the most of their quota
- Gross measures of traffic "tonnage" don't translate either to costs or congestion. It's traffic in busy hours and busy cells that matters. An extra 10GB of video downloaded at 3am in a rural cell is essentially free. An "offpeak" dataplan might increase reported traffic volumes but have no impact on costs or congestion. In fact, it might generate incremental and very profitable revenue by increasing capacity utilisation during quiet periods.
- For many networks, signalling load (against both the RAN and the core network) is the problem, not data tonnage. Multiple short bursts of data or "pings" clog up the network in different ways to a single, consistent stream. But it's harder to measure and bill for signalling, so it tends to get ignored
- Smaller cells give greater capacity density, at a lower price. They allow better spectrum re-use, reducing the need for new bands. In the long run, we get much more extra capacity by reducing cell size rather than adding extra radio channels - but this conflicts with the desire to grab more spectrum from alternative/competing users.
- Other new technologies and processes will improve network efficiency too: beam-forming, better sectorisation, MIMO and so on. But these are less well-proven than simply adding extra bands.
- "Video" isn't an application, it's hundreds. Amalgamating 500+ different applications and services under a single banner is completely arbitrary and meaningless. It's like saying the web is disproportionately dominated by the colour blue, and it should therefore be treated differently. (It might be green, I don't know).
- The dynamics of demand for mobile broadband are over-simplified. Much of the historic growth has been from "more users" rather than "more use per user". There are additional issues (discussed below) about coverage, device capability and so on.
- Demand growth and capacity growth are not directly linked - especially because "capacity" is impacted by numerous factors such as backhaul as well as radio-network scale. We also find that some base stations are "congested" because they haven't yet been upgraded to the maximum number of radio carriers already available.
- The dynamics of the mobile data market differ for post-paid and pre-paid users. As PAYG (which makes up the bulk of the planet's mobile customer base) becomes predominant, the idea of a monthly allowance will shift to a more usage-linked model. Early evidence suggests that PAYG data users - with smartphones - consume much less data than those on fixed plans. This is not factored into most forecasts.
- Some forecasts start from 2009 or 2010, therefore building in a huge initial leap from a low base. Ignore anything that doesn't use 2011 or 2012 as a start year, as otherwise the statistics will get swamped by vague and patchy measurements of historic data.
NEW REPORT - 10 REASONS WHY 1-800 TOLLFREE APPS MODEL WON'T WORK
We're already satisfying much of the latent demand
I'd argue that in developed markets such as the US and UK, we are already at 50%+ of potential mobile data usage saturation given TODAY's devices, data-plans, apps and user behaviour. Almost anyone who really wants a smartphone already has one.
Many people who want cellular-connected tablets or laptops already have them too. Sure, there are still a few demographics that want them but can't afford them, but that group is diminishing rapidly. The remainder are mostly "laggards" - folk who might like a new device, but who are likely to be "unenthusiastic" in usage behaviour, at least in comparison to those with a 4S or S3 or L900 in their hands 24x7. There are surprising numbers of mobile data "refuseniks" whose attitude is unlikely to change.
If you magically increased the smartphone penetration of all developed countries to 100% right NOW, I'd be surprised if it added more than 50% to tomorrow's data consumption stats.
NEW REPORT - 10 REASONS WHY 1-800 TOLLFREE APPS MODEL WON'T WORK
A thought experiment
Consider a fictional place where 50% of people have mobile broadband (smartphones with data plans, plus some fraction have 2nd/3rd devices), and 50% don't have MBB, either because it's too expensive, or because they live outside coverage, or they're just apathetic.
Let's say that the 50% of current data users are using 1GB/month as an average (mean). There's a mix of capped and uncapped plans, some people are heavy users (with one or two devices), others are more parsimonious, or perhaps just use WiFi a lot.
What is the "unconstrained demand" from these people at around current data-plan prices? I think we can now assume that the mobile broadband marketplace is now pretty efficient at giving people roughly what they want, at roughly the price they're prepared to spend. I'd be surprised if the true unconstrained demand from existing mobile data users would be much more than 50-100% higher than today. (Yes, if we dropped the average price massively there'd be an elasticity effect and demand would rise, but let's leave that option for a moment).
What are those constrains?
Thinking about Mobile Data Demand Constraints
A good way to think about this is to consider "what would happen to overall data traffic if we removed certain constraints?". This is often a counter-intuitive thought process because it involves going beyond the raw statistics and thinking about the real world and user behaviour.
So it's tempting to say that going from 50% penetration to 100% overnight would result in an instant doubling of traffic. But actually, it wouldn't because the remaining 50% would be much less enthusiastic users than the early adopters, especially on Day 1. Similarly, if we had perfect 3G/4G coverage tomorrow, we'd see traffic growth but not that much overall because all the busiest areas are already covered. What's left are big zones of occasional use (rural), and quiet corners of some indoor spaces.
This type of analysis is inherently much more complex, and goes beyond most statisticians' comfort zones. But it's a critical application of common sense. It's a sanity-checking phase too, that often seems absent in a lot of the mobile data forecasts I've seen.
For the thought experiment, let's consider I've got a magic wand to remove constraints. Each time, I'm going to leave the other variables untouched, especially price. What might happen?
1) Device penetration - if you gave everyone a smartphone tomorrow with a dataplan they could afford, plus cellular tablets/dongles at a pro-rata penetration to the early-adopter base, I expect we'd get around another 40% of traffic. (Heavily dependent on existing penetration of smartphones, eg 30% vs 50% - by the time some people read this post, we'll be nearing saturation anyway as it's moving so fast). In fact, the figure might be much lower - maybe just +20-30%, because most of those users will be prepay subscribers who tend to have lower data consumption anyway.
2) If we had perfect cellular coverage everywhere, I reckon we'd get a 20% uplift in aggregate traffic. Network planners aren't stupid - they know where the demand is, and the economics of satisfying it. Providing coverage to every mile of road and rail, or to every small village would definitely be nice - but it doesn't compare to a a big metro area when added together.
3) Now for a biggie. If we improved network speeds to 4G-type rates, with better latency, what would that do to user behaviour? More video streaming? Probably. More web use? Sure. More cloud-based apps? Perhaps. This is a tough one to predict, but we see some indications from people who move from 3G to LTE (although some of that is about getting upgrading to a new & better device rather than having a better network). Some stats say that LTE users typically use 50% more data than 3G users. BUT, that is skewed by early adopters switching first. But given today's apps and data plans and user expectations - which are often met pretty well on 3G after all - I'd say 40% is reasonable if the speed constraints were removed.
4) Device performance is another variable here. A lot of people have quite old devices that are slow, clunky, have low-res screens or are otherwise constrained, irrespective of network capabilities or coverage. But how much of a big deal is this really? Again, most of the real heavy users and enthusiasts do have the latest devices. If you waved the proverbial magic wand and upgraded all the old 3GS's and Galaxy S2's and assorted BlackBerries to today's state of the art, what would happen? Not much I reckon, again if all other variables were kept constant. (There's a bit of a co-dependency with LTE availability as noted above, though). I reckon that across the user base as a whole , we'd see perhaps a 30% uplift in data usage.
Now let's bring all these together to see what might happened if we removed the constrains (except price, and again bear in mind this is with today's typical apps and behaviours):
Device penetration = 1.4x (maybe lower)
Coverage = 1.2x
Network speed = 1.4x
Device performance = 1.3x
Multiplying through, we get an estimate of unconstrained demand = 3x today's constrained demand
But some of this is - in all reality - never going to happen. We're always going to have a spread of device ages and capabilities. We're never going to get 100% coverage. Some people will never use Facebook or Netflix or Dropbox, no matter how fast the network. PAYG prepay users will use less data for various reasons. And some people will hold onto their Nokia 6310 from 2003, even if you try and bribe them with the latest smartphone for free.
So in the confines of this thought experiment, we're probably addressing 50% of *current* unconstrained demand for mobile data, at current prices.
That's probably true of a bunch of other industries as well. I'd guess that we're probably at around 50% of unconstrained price-constant demand for anything from flights (people don't have enough holiday time, or are scared of flying etc), or even unconstrained demand for beer (can't drink at lunchtime before a meeting, too young, health/religious issues etc).
It also passes the "taste test". Most people don't spend all day moaning about how they're only getting a fraction of the mobile data they want. Generally, apart from a few minor gripes (coverage mostly, plus congestion in some hot-spots), people seem pretty happy that their mobile data demands are being met.
Obviously, this is macro-level stuff. Specific places (eg Olympic Park) clearly see much faster growth in demand as there will be localised drivers. Also, the calculation in the thought experiment above will vary by country a lot too - there are different levels of network rollout, smartphone/dataplan adoption and so on. India, for example, is starting from a much lower base for traffic, and so many of the variables will be considerably higher to work out latent demand.
Mapping future demand
So. Let's say that with current devices, networks, apps, behaviour and pricing, we are dampening consumption of mobile data by a factor of maybe two from the theoretical realistic demand in today's developed markets. That's a useful number to bear in mind, as it means that:
Any future mobile data traffic growth is primarily going to come from new demand not satisfying current latent demand
That's important in technology. It's often said that "usage always expands to fill the capability available", or "build it and they will come", but that's not actually true. The reason that computer processor speed has always been exploited (and so quickly) has been that companies such as Intel have spent lots of time and resources on "demand creation", seeding developers with new technology, running marketing programmes and so on.
So where are all these forecasts for 10x, 20x - even 1000x - growth for mobile data coming from?
Is it just the mobile industry's normal ridiculous arrogance (almost a sense of entitlement about growth) and its propensity to ignore lessons painfully learnt elsewhere in the technology industry?
Well, firstly there's still a lot of untapped growth in emerging markets, although it again needs to be born in mind that at current network costs and dataplan prices it is likely that data use will be lower per-capita. A $2 data prepay ARPU is not going to give 100% coverage, $300 subsidy of an iPhone 4S, and a 1GB per-month plan. In general, that part of the market will be using cheaper/less-capable devices, on thinner/slower networks, with more restrictive tariffs than elsewhere. They will also likely adopt different behaviour to "squeeze more from less".
And, as mentioned above, we'll also see growth in smartphone use among late-adopters in developed markets.
So while this means we should see mobile data user numbers continue to grow rapidly, this will also paradoxically drive down the average data consumption as heavy users (who may well be using more data year-on-year) will get diluted by newer and ever more numerous lighter users.
I have not seen this mathematical inevitability called out on any forecasts, yet it happens in virtually all markets as we see a progression of maturity. A grandparent in rural Bolivia getting their first smartphone is unlikely to be using Facebook and streaming video 24x7 on Day 1.
So, where else is the extra usage going to come from? Various sources are possible:
- Even faster / better networks making apps more usable
- Device improvements like bigger screens
- New apps and services (eg mobile cloud-based offers)
- More cellular devices per person (either "attach rates" of personal devices like 3G tablets, or others like M2M that bump up the total)
- Behaviour changes meaning greater time spent on mobile apps
- Lower prices driving elasticity (eg through changing behaviour faster/further)
- Better structured data-plans driving off-peak usage volumes
- Switching from using fixed broadband or WiFi to LTE
NEW REPORT - 10 REASONS WHY 1-800 TOLLFREE APPS MODEL WON'T WORK
Many of these are interlinked, obviously - if the network is faster, it enables new apps, and alters peoples' behaviour.
It is also important to note some downward pressures on users' average consumption of mobile data:
- Better devices and OS's that compress data (eg similar to BlackBerry, Opera Mini & Nokia Asha which route "optimised" data via a server), or third-party software like Onavo's
- Various types of network-based optimisation and compression, especially for streamed video
- More use of adaptive applications (eg HLS-encoded video) which self-optimise to network conditions, or more efficent codecs
- Substitution of 3G/4G data with WiFi, either as true "offload", or (much more importantly) user-driven preference for accessing private WiFi, usually for free
- Older devices being retired (for example, I'm about to cancel my 3G dongle contract as I never use it - I get WiFi almost everywhere I want it)
- App developers becoming wiser and more parsimonious about how their software consumes data, especially if they get better development/testing tools, or are "shamed" into it in appstore ratings
I don't have quantitative forecasts for all of these drivers and constraints. But some appear to me to be especially important:
- More usage per person driven by better apps and devices, and behavioural change. Let's tackle the latter first. I honestly don't think that the average mobile data user is going to increase their time spend on mobile devices by another 2x, 3x, whatever. We're close to saturation on that one already. Better apps and devices? Definitely. I agree that there will be more video and cloud-app usage. I can certainly see scope for an extra 2x or 3x over a 5-8 year period. However, the swing factor here is likely to be tablets and the evidence suggests more extra usage will be WiFi-based.
- WiFi is to my mind the biggest "decelerant" here. Operators are (often) trying to do their own controlled offload of traffic, although I still believe that most examples are going down the wrong path of "seamlessness" with things like ANDSF and Hotspot 2.0. However, that is becoming less relevant anyway, given the huge global explosion of non-operator WiFi and increasing sophistication of users in exploiting it. Partly because of data pricing and caps, users are actively seeking "free WiFi" wherever they go, and becoming adept at using it - especially the high-end power users that normally generate the most traffic. Unless we see lots of WiFi congestion (possible), that move now seems irreversible. Recent Ofcom data shows that more people are becoming "WiFi-primary", just using 3G/4G where they have to.
- More devices per person. Yes, this is going to rise, even though most tablets will likely remain WiFi-only. We'll see various new mobile-enabled gadgets in all walks of life. M2M can, however, be dismissed as a major traffic driver as the vast bulk of products are low-bandwidth. Despite a few high-consuming categories like digital signage or in-car telemetry, there are none that obviously have the scope to scale to billions of units. Against phones and to a less degree PCs, tablets & MiFis, they are lost in the noise.
- Despite better uplink speeds and slow rollout of fibre, I don't see signs of much shift from ADSL/cable to mobile-only. Coverage limitations and need for IPTV and WiFi are likely to keep fixed broadband largely protected except for a few niches.
- I do see quite a lot of traffic growth being driven by "off-peak" data plans. Marketeers and their billing systems are becoming smarter. However, this is essentially irrelevant from the point of view of network capex, spectrum needs and so on.
The price elasticity issue is a difficult one to address. If 4G was completely free & ubiquitous, you'd find people getting LTE-enabled 42" TVs in their home, running HDTV over cellular even when there is nobody in the room. In that case, even the most bullish forecasts would still be too low. Clearly, that's an extreme example, but various other milder scenarios are possible.
This is the paradox in all the forecasts: there seem to be no clear assumptions on pricing. Cisco's VNI projections probably are achievable, but only if mobile data is priced at levels that nobody would make any profit from it. As we found with flatrate, it's easy to drive usage if you're throwing away money.That said, we may see users being encouraged to migrate to LTE with offers of larger data bundles, which effectively reduces prices.
Taken together, my belief is that the bulk of forecasts are over-hyped. I think that too many of the projections are being made without considering how devices, apps or users are actually changing. There also seems little recognition of the "dilution of the average" as late-adopters, prepay users and developing-market subscribers bring down the headline "per-user" numbers.
Quantitative estimates and forecasts
I haven't done a full spreadsheet model analysis & forecast, but since I'm sure everyone's going to ask me anyway, I'll "take a punt" on overall global data traffic volume growth rates, based on existing published stats & bearing in mind the qualitative factors described in this post. It's worth noting that most spreadsheets I've seen are long on detail but a bit short on commonsense, especially with assumptions on per-user data growth continuing inexorably, despite the "average dilution" effects described here. There is also a general assumption that "subscription" remains the main business model, rather than more ad-hoc usage through PAYG. (Although, obviously, the overhyped 1-800 tollfree models won't be happening)
My prediction for overall global data traffic growth (Indexed to full-year 2011) is as follows:
2012: 70%
2013: 55%
2014: 45%
2015: 40%
2016: 35%
2017-2020: CAGR 30%
In other words, 2011-2016 growth is 7.2x - and in developed markets, probably more like 4x, and by 2015 we should see growth rates broadly comparable to that in fixed broadband in developed markets.
Given advances in technology, this should be "sustainable" with ongoing evolution of networks to LTE, especially small cells & better radio technology like beamforming. Some of the traffic growth is likely to be off-peak, as pricing & policy becomes smarter.
This means that growth will be more of a regular tide than a "tsunami", and definitely not something for operators to panic about - and regulators need to learn to be skeptical of shrill demands for more spectrum.
SIGN UP HERE TO RECEIVE THIS BLOG BY EMAIL
NEW REPORT - 10 REASONS WHY 1-800 TOLLFREE APPS MODEL WON'T WORK
Some comparisons with other forecasts:
Cisco VNI expects 18x growth from 2011-2016 (vs. 7.2x from my best estimates)
Ericsson predicts 15x growth 2011-2017 (me: 9.4x for that period)
Reading from a chart in this ALU presentation suggests 30x from 2010-2015, or about 10x from 2011-2015 (me: 5.5x)
My peers over at ABI reckon 8x from 2012-2017 (me: 5.5x for that period)
Informa predicts 10x for 2011 to 2016 (me: 7.2x)
Morgan Stanley gives scenarios for 5x, 9x & 23x for 2011-2015 (me: 5.4x)
Edit, 14th September - AnalysysMason makes very similar arguments to me (and has been similarly pessimistic to me for a while). They reckon 5.5x global growth for 2012-17 (me: 5.6x) and "dangerously low" for Europe. I agree - overcapacity is the risk, not a mythical "spectrum crunch"
No comments:
Post a Comment