The limitations of ARPU as a metric for mobile phones are fairly well-understood - people have multiple accounts, it's skewed by certain things like BlackBerries and business users, and has a hard time dealing with pre-bundled services. Yet observers and investors continue to use it, despite these caveats, because they now have a "feel" for how to interpret it against this complex background - and also because there isn't really something else convenient to use. Particularly when you separate out between prepaid and postpaid, ARPU can still convey something of use, if you're sufficiently cynical about it.
Unfortunately, I think this complexity is going to get worse with mobile broadband, rather than better. At least with phones, people tend to use them regularly, whether with prepay or postpay subscriptions. And phones tend to be either actively used for making calls/SMS, or not used at all. Nobody uses a mobile handset exclusively "offline".
In the early days of mobile broadband, it was much easier - there was usually a fixed monthly subscription, plus any extra megabytes and roaming charges. No really onerous accounting judgements to make.
Now, things are changing - various models of prepaid mobile broadband are popping up, either with traditional "top-ups", or pre-packaged. I saw an offer yesterday for a HSDPA dongle with 12GB of data pre-bundled, usable at any point up to 12 months from purchase. Also, there are various "free notebook" offers that massively hike the apparent ARPU by bundling in the implicit consumer finance of the computer with the data access charge.
That's a bit like buying a car with a built-in cellular tracking system, and charging the whole monthly finance charge against ARPU.
Going forward, the picture is going to get even more complex. There will be more "connectivity built-in" devices like the Amazon Kindle. If I (say) pay $300 for a mobile-enabled games console, with zero monthly fees, how does that count in ARPU calculations? Something based around the wholesale fee charged by the MNO to the device manufacturer? But what happens if it's offset against (say) using the device for the operator's TV and advertising channel?
What happens with instances of free or sponsored use? Let's say the Tunisian Tourist authority offers to sponsor 10MB of roaming data as long as a visitor checks out their restaurant guide & completes a survey form? What about free broadband given to delegates at a conference? How does that fit into either the "revenue" or "number of users" calculation?
What about bundling? Fixed+mobile broadband, with a DSL modem plus integral 3G dongle so I can take my broadband with me when I go out? That should keep the accountants busy.
If I get an embedded-3G notebook with a 2-month "free trial", am I counted as a user? And what happens if I sign up to iPass, which has various MVNO deals in a number of countries - am I one mobile broadband user, or six? And what happens where I offload traffic to a femtocell, or WiFi - am I double-counted, and how is the revenue split recognised?
And all this is before we get to any of the really clever Telco 2.0 style business models. If an MNO sees clusters of active 3G notebook users around certain locations, and sells the data to an advertising agency planning a campaign for Dell or Toshiba, does that count as mobile broadband "revenue"?
None of this lends itself to a nice, convenient headline metric like ARPU unfortunately. But that's the nature of convergence, so we have to ditch it before it obscures the reality of the market.
I'd suggest that a more useful metric might be the number of active individual users, or active devices, coupled with some quantifier of service revenue split between end-user and wholesale/3rd-party. Hardware sale/subsidy/financing should be stripped out and reported separately.
And there needs to be absolute clarity on definitions and assumptions - I see too many charts which give bland numbers of "mobile broadband subscribers" without defining what a "subscription" actually is.
And if you're an investor - now is the time to press CFOs on this. There isn't much legacy data on this market, so now is the ideal time to pick a useful metric to start with, rather than a superficial but increasingly meaningless one.