How Much Quota Do Internode Customers Actually Use?

All ISPs who keep sensible usage records know exactly how much the users on each of the plans they offer actually use and the time frames in which they use it – a rule of thumb is that the average usage (across any 12 month period) is less than 40%.
John Linton’s Personal Musings

The formula by which Internet Service Providers (ISPs) in Australia price their ADSL2+ plans is pretty standard.  The rough rule of thumb is “a dollar per day and a dollar per GB”.  So, for example, you can expect an 80GB plan to be priced at around \$30+\$80=\$110 per month.  Give or take.  Certainly some ISPs offer additional large “off-peak” allowances (that can only be used in the wee hours of the morning).  Or they bundle broadband plans together with telephone line rental & calls.  But if you look at standalone ADSL2+ plans, and consider only their “on-peak” allowances, the dollar per day and a dollar per GB rule is remarkably consistent from one company to the next.

Internode are little bit different, and perhaps unique, from other ISPs in the way in which they structure their plans.  In June 2007 they completely revamped their plan pricing structure and now cost average within plans, not between plans.  Each pricing tier is self-supporting.  In other words there are no loss making plans being subsidised by profitable plans.

Knowing that each pricing/allowance tier acts independently it may be possible to derive some sort of average customer usage from information that is publicly available.

Below is a summary of Internode’s Home Extreme ADSL2+ plans (for simplicity I’ve rounded up to the nearest dollar):

10GB \$50
25GB \$60
40GB \$75
55GB \$90
80GB \$120
100GB \$150

If you run a regression line through these data you’ll find that they can be very neatly described by the formula:

monthly cost = \$45.28e.0121quota

The model is saying that before a customer even starts downloading data there is a fixed cost of \$45.28 per month (i.e. when quota=0, e0=1).  I’m not sure that an ADSL2+ port connection is really costing Internode anything like \$45 per customer per month.  To be honest I’d be surprised if it was more than half that.  But what do I know?  Nevertheless, that’s our starting point.

Taking out the \$45 fixed monthly “supply charge” from the overall plan cost an indicative price per GB in each tier can be summarised as follows:

10GB \$5 \$0.50
25GB \$15 \$0.60
40GB \$30 \$0.75
55GB \$45 \$0.82
80GB \$75 \$0.94
100GB \$105 \$1.05

It’s very interesting to me, looking at the table above, that the indicative cost per GB (after taking out the \$45 fixed monthly access charge) delivered to the customer varies from tier to tier.  From 50c a GB in the lowest quota plan up to \$1.05 in the highest quota plan.  Common sense would dictate that it costs Internode the same amount to deliver a GB of data to you regardless of what quota plan you’re on.  If cost-to-the-ISP is constant then a possible explanation for the variation could be that customers use different proportions of their quota from tier to tier.  Apparently, and perhaps not surprisingly, customers on low quota plans use proportionally less of their full allowance than people on high quota plans.  If it’s possible to make a guess as to what it costs Internode to deliver a GB of data to its customers, it may be possible to determine what those proportions are.

From what I’ve read, it costs an ISP of Internode’s size something in the order of \$130-\$200 per mbps of bandwidth delivered during the 12-hour peak time noon to midnight period (off peak is significantly less from what I understand).  So let’s split the difference and say \$165 per mbps.  Delivered 12 hours a day over a month this equates to (165x8x1000) / (30x12x60x60) = roughly \$1 a GB per month.  That’s the raw bandwidth cost, ex-GST.  It does not include all the other costs of running an ISP such as tax, rent, equipment, and staff.  Not to mention their profit margin.  And it does not take into account the cost of uploads which Internode absorbs into their download pricing (uploads don’t count toward your quota at present).  I don’t think it’s unreasonable to assume that it costs Internode somewhere in the order of \$1.50 to deliver a GB of data in a month.

Dividing each of the nominal costs per GB across each tier in the previous table by the estimated fixed cost of \$1.50 calculated above results in the following estimates:

10GB 33% 3GB
25GB 40% 10GB
40GB 50% 20GB
55GB 55% 30GB
80GB 63% 50GB
100GB 70% 70GB

So for example looking at the 40GB tier: \$0.75/\$1.50=50% of the monthly allowance=20GB per month.  In other words it appears that, on average, subscribers to Internode’s Home Extreme ADSL2+ 40GB plan use just 50%, or 20GB, of their monthly quota.

Can this analysis be correct?  Maybe.  I get a good vibe from it.  The numbers feel right.  It kind of fits in with John Linton’s quote at the start of this blog entry.  And the average of the estimated averages (I know, I know… average of an average and all that) across the tiers comes to about 30GB… about right I think for ADSL2+.  But obviously the estimated proportions are dependent on my wild guesses at Internode’s fixed costs.

Anyway, the point is it suggests the typical customer is on a plan too large for their average requirements.  They could potentially save money by dropping down a tier without loss of service.  Perhaps people are unaware of their average usage.  Or perhaps customers are willing to pay more for that extra overhead.

Or perhaps I just got it totally wrong?

7 Responses

1. Hi Stanley,

Simon Hackett has responded to this, so I won’t go into a great level of detail, however a couple of easy hit points that would make the equations different:

1. Those on higher quotas are much more likely to use ‘all’ of their quota. My (personal) thoughts would be that as you hit 80GB and 100GB plans you’d be much more like to see 90% usage.

2. Blended costs. ie with any IT infrastructure, the price increases for increased bandwith are not linear, in fact as networks become more complex your multiplier (per head) goes up. As an example of this, when our business was 800 staff 3.5 years ago, average cost per user, per year was around \$2,000. Now that we’re 4,000+ staff our average cost per user sits at around \$4,000 per user, per year.

This takes into account not just your (relatively) fixed costs like bandwith connections, but increased staff (i’m a project director, and as complexity increases, so does the need for highly skilled PM’s to manage non-revenue generating projects, which, in turn, come with expensive price tags!), increased security requirements, add-on services (like video conferencing, a nightmare that i’m now managing) etc etc.

Don’t get me wrong, the services that are provided today are worth far more than double what they were 3.5 years ago, however the increase in blended costs does mean that (to go back to your post) a GB is not really just a GB.

2. Excellent analysis.

I certainly overestimate my bandwidth usage across all devices/locations as often (e.g. for mobile) the penalty for going over is cripplingly high.

3. I love it. Unfortunately it’s not nearly as half-baked as I’ve come to expect and take incredible delight in (I would suggest at least three-quarter baked..). I will however suggest one thing – you assert the port cost is around \$45, and then go on to say of your data cost of ~\$1/GB

“That’s the raw bandwidth cost, ex-GST. It does not include all the other costs of running an ISP such as tax, rent, equipment, and staff. Not to mention their profit margin”

whereas I think you’d find that overheads such as staff, tax, rent etc would be related more to the number of services rather than the amount of data said services use. Therefore, these costs are essentially incorporated into the port cost.

• Yes, I think you’re right. Most of the overheads would be incorporated into the port cost, not the data cost.

• Just been having another little think about this.. I think the real “cost of data” can be determined by modelling the business plans, rather than the residential plans. The Business plans are always touted as being uncontended, therefore the logical extension is that data included in these plans is completely paid for down to the last byte.

So, if we do a simple plot, we discover that the Included Data vs Cost regression comes to being exactly linear (i.e. R²=1) with C = 69 + 1.2Q, where C is mothly cost in dollars and Q is monthly included quota in gigabytes.
(It would be reasonable to assume that the ‘port cost’ is higher for business plans, given a priority assistance line and SLAs, so a \$69 intercept is not unreasonable.)

Once you know the real ‘cost of data’, I think you can start to make more reasonable estimates of the proportion of people using their quota.

– Alex

• Hi, Alex

Thanks for the feedback! So you’re saying I should have used \$1.20 (the coefficient of your linear regression), rather than \$1.50, as the estimated cost per GB in my analysis? I have a “feeling in my waters” that you’re closer to the truth than I was. \$1.50 really was a bit of a Wild Arsed Guess. That would change my proportions quite a bit.

Thanks again.

• I think the closest that we can get, for “Data delivered by Extreme ADSL”, is the Business plan price. Indeed, it is singular that the Business plan prices follow a linear regression so well, almost in fact as if that was how they were set.. (Conspiracy, anyone?)

Assuming the Business data co-efficient to be correct, the prospect of discoveraging average usage for the residential user becomes simpler, viz: this piece of time wasting.
The ‘average’ use for the residential user is therefore reasonably simple to deduce graphically.

Interestingly, we all know the Extreme SOHO plans are priced an arbitrary \$20 more expensive than their home counterparts, yet this merely indicates that their ‘port cost’ is higher, whereas I would probably think that someone who has need of a static IP address, and Annex M, would probably use a higher proportion of their data than residential users. This of course, is merely conjecture, and is probably untrue. I cannot yet properly rationalize a mechanism for the SOHO pricing, and I suspect that the \$20 differential is indeed arbitrary and unrelated to data usage.