Monday, March 15, 2010

Sexy model curves on iPad pre-orders

No this is not some spam post offering a pseudo-porn iPad app. Just go read my previous post to see what this is about. Also, don't miss PED's quoting my random musings on this stuff (and outing me! LOL):

PED: Apple iPad orders drop sharply

On to the "sexy" curves, click to enlarge (hey, some extra hits due to my title shenanigans can't hurt, and maybe someone somehow learns something new thanks to my mischief).

This one's pretty self explanatory. A nice, round rump evolves after 12 hours into a relatively straight line slightly sloping up. I'll just add that this slope will increase in the next few hours and days, that is, the line will curve in the opposite direction right about now (72 hours) and get steeper, and then it'll become straight once again for some time. Once we get past 100-150 hours, I expect a new, more subtle rump to form in which the slope will very slowly get shallower as the days and weeks pass.

This one's also easy to read. It's an attempt to measure the value of this slope over time (an approximation since I can't actually do infinitesimally small intervals), also called the first derivative, or rate of change. It simply represents the order volume rate per hour. Notice the sharp drop off early on and then it manages to maintain that small value for a longer period of time. The flat part here corresponds with the straight slope upwards of the previous curve.

Please don't pay attention to that harsh bump at 10 hours as it's due to a noisy sample most likely off by a few minutes from its true time value as recorded in Apple's servers.

Again, this curve will start to creep back up just a bit over the next few hours, and maintain a higher plateau for a longer period of time, and then after a few days keeping its ground it'll start showing a very slight decline only noticeable over much longer periods.

I'd also like to mention here that I left out a few points that would go in the tiny sliver of space between the vertical axis and that first point up there at 70 thousand. These points would plot much higher, higher than 100, and one of them was over 500 (that is, half a million orders per hour, but of course there never was half a million orders placed, just that the time span during which this happened was a minute or two, so 500,000/60 minutes lasting for a minute or so would generate something like 8-10 thousand orders).

The problem with including these points on the chart (despite them being really fun to look at) is that in order to show them so high up there, the zooming out of the vertical scale would have made the later part of the curve where it's more stable as if it was laying flat against the horizontal axis, thus potentially misleading one into thinking that the order rate just died. It did not die, it's oscillating a couple of times at around one thousand orders per hour without falling further, which is quite respectable for a weekend.

But that part is still really hard to see, like, those oscillations are important as they show the day/night volume patterns, and it'd be nice to get a closer look at those values. That's the motivation for the next chart.

I think this is the coolest one. Please ignore those two outliers at a value of 1, which correspond with that noisy sample I was talking about before. As to what the values on the axes mean, well it's just the logarithm of the same values plotted previously. Something similar would happen if you used log scales on both axes, except the normal values for hours and thousands of orders would show at different spacings.

To help you find your place there, remember that log(1)=0, log(10)=1, log(100)=2, etc. So, on either axis where the zero is, it's the 1 hour point in time or 1k orders/hour level. Where the 1.0 is shown, it's the 10 hour point, or the 10 thousand level, the 2.0 is where we'll get to the 100th hour or when the rate was 100k orders per hour (only during the first 19 minutes of sales, which is 0.32h and log(0.32) is right around -0.5 which is the left side of the graph so those very early orders are not shown). As to what this curve is saying, what it means, I'll leave it up to you to interpret. Why do the dots fall much more in line? What would it mean if future dots suddenly stop falling or start turning up? All I'm going to say is, don't think those last few dots where it shows a temporary recovery for a cluster of them near the right side of the curve are noisy data points. These points are several hours and thousands of web order numbers apart, so a couple of minutes off creates no noise here. And there are several dots plotted at the temporary recovery (during daytime on Saturday and Sunday).

Finally, this last graph (also shown in PED's article) is similar to the second one, except here I've only used a few points spaced out about 12 hours apart. A clarification is needed about the yellow label there which says "iPad sales" which should have said "iPad unit sales" so it's not showing $ sales. But the most important addition here is I show my estimated volume rates for orders that don't include an iPad pre-order.

Wow this turned out to be quite a long post. Sorry if you were expecting some hot chicks in bathing suits, but if you read all the way through here, I appreciate your effort and I'm glad I could keep your attention this far.

Stay tuned over the next week or two.


Unknown said...

Nice analysis.

A small comment:
Cumulative has only one "m".

Daniel Tello said...

Thanks Judy!

smirky said...

You know, you could have easily found a suitable photo and placed it in the background of the graphs that would have "accentuated" the sexy curves a bit. We're living in the 21st century! The graphical display of data pseudo-porn no longer have to be mutually exclusive.

Anonymous said...

Great analysis! It will be interesting to see what happens this week and to add in estimated reservation units. I'd expect 75-80% of those reserved will actually be picked up on Day 1. But I'm fairly sure that reserved units not picked up on Day 1 will be bought by those in line who didn't reserve one.

I did a similar analysis back on Friday with just a few order numbers and had my first results by around 10am. I then recalculated after 10 hours with more data points. I didn't really believe my results until I saw yours. Thanks a lot! Love the graphs.

I'm going to try a few more things.

First day US sales:

Apple has 222 stores in the US. They will each be open 12 hours on April 3rd. That's 2664 store hours on the first day of US iPad sales. Picture long lines of customers and lots of sales and checkout staff. If an Apple store can sell just 80 iPads an hour, thats another 200,000 or so iPads sold on April 3.

If that seems like a high number, it's not really. It's probably low. The iPhone launch in June 2007 (Apple had fewer stores then but also sold iPhones from a small number of AT&T stores) sold 500,000 iPhones in a weekend from Friday at 6pm til 9pm Sunday. I'll have to somehow figure out how to parse sales of reserved units and unreserved units.

At any rate, the iPad first weekend U.S. store sales should beat the iPhones for a lot of reasons but even if it just matches it, that's another half a million on that weekend alone.

Non-US sales:

Apple said the iPad would go on sale in Australia, Canada, France, Germany, Italy, Japan, Spain, Switzerland and the U.K. in late April. I'm going to apply my previous analyses to these markets based on their population and Apple presence/market share their.

Future rate slope increases:

These are pretty obvious I guess. Expect a slope increase when 3G iPad US sales start til the end of April. Then add in international starting late Spring and Summer. Then back-to-school buying in August/Sept. Then Christmas 2010.

Bottom line is I think Wall St. projections for iPad sales for 2010 (4-5 million) are wrong and extremely low.


Anonymous said...

was expecting hot chicks, not in their bathing suits though.

jmmx said...

Another set of orders not covered by your analysis is the business orders. We have no way of knowing what these are, but reports they they are also being limited to 2 per person suggest that they are also very robust.

jmmx said...

I think you are being a little conservative in your estimates. Given this response, I think we will see 1 million by end of first week, and an average of close to 1 million / month thereafter. I will throw my hat into the 7M by year's end ring.

Last thought - nobody expected that early rate of sales to hold up. No one expected it to be that high to begin with.

JavaJack said...

You got a complimentary, well deserved mention this morning.

Anonymous said...

how good is your data for regular orders on I guess at one point it will be difficult to separate iPad pre-orders from regular orders, don't you think. Do you have historic data for regular purchases (including seasonality or at least day of week distribution)?

Anonymous said...

nice analysis, this proved that Apple still one of the most valuable brand

Anonymous said...

Are you going to take into count Amazon pre-orders?

Sacto Joe said...

One fascinating bit of news: According to Apple Insider, the iPad keyboard dock, the iPad soft case and the iPad 10W USB power adapter are now in back-order mode. So those of us who ordered early are now officially the birds that got the worms….

Anonymous said...

At best this just shows the ultra early adopters that will buy this sight-unseen. At worst, this is pure speculation.

JavaJack said...

I'm getting anxious to see some 2nd quarter numbers. When will we start to see some data?

Anonymous said...

The analysis was insightfull knowing that Credir Suisse reported their estimate of 90 days to sell 1M Ipads. I would like to see some Q2 and Q3 earnings earnings estimates.