Monday, June 9, 2014

Predicting Salesforce’s Subscriber Count

So this is my third consecutive post on Salesforce. I promise the next will be Dynamics CRM or something non-Salesforce at the very least.

For avid readers of the blog, you will know that every quarter I review the quarterly submissions of Salesforce to the Securities and Exchange Commission (SEC). Up until January 2011 this included the subscription numbers but a sad thing happened after this date; Salesforce no longer released their subscriber numbers. After July 2011 they also stopped releasing their customer numbers. There was no explanation, the numbers just stopped coming. Here are the data we have from Salesforce’s public statements up to these dates.

Financial Quarter Month Subscribers Customers Average Company Size
2003 Q4 Jan-03 76,000 5,700 13
2004 Q1 Apr-03 85,000 6,300 13
2004 Q2 Jul-03 96,000 7,000 14
2004 Q3 Oct-03 107,000 7,700 14
2004 Q4 Jan-04 127,000 8,700 15
2005 Q1 Apr-04 147,000 9,800 15
2005 Q2 Jul-04 168,000 11,100 15
2005 Q3 Oct-04 195,000 12,500 16
2005 Q4 Jan-05 227,000 13,900 16
2006 Q1 Apr-05 267,000 15,500 17
2006 Q2 Jul-05 307,000 16,900 18
2006 Q3 Oct-05 347,000 18,700 19
2006 Q4 Jan-06 393,000 20,500 19
2007 Q1 Apr-06 438,000 22,700 19
2007 Q2 Jul-06 495,000 24,800 20
2007 Q3 Oct-06 556,000 27,100 21
2007 Q4 Jan-07 646,000 29,800 22
2008 Q1 Apr-07 742,900 32,300 23
2008 Q2 Jul-07 800,000 35,300 23
2008 Q3 Oct-07 952,500 38,100 25
2008 Q4 Jan-08 1,100,000 41,000 27
2009 Q1 Apr-08 1,177,200 43,600 27
2009 Q2 Jul-08 1,287,900 47,700 27
2009 Q3 Oct-08 1,398,600 51,800 27
2009 Q4 Jan-09 1,500,000 55,400 27
2010 Q1 Apr-09 1,660,400 59,300 28
2010 Q2 Jul-09 1,769,600 63,200 28
2010 Q3 Oct-09 2,000,000 67,900 29
2010 Q4 Jan-10 2,102,500 72,500 29
2011 Q1 Apr-10 2,319,000 77,300 30
2011 Q2 Jul-10 2,554,400 82,400 31
2011 Q3 Oct-10 2,790,400 87,200 32
2011 Q4 Jan-11 3,000,000 92,300 33
2012 Q1 Apr-11 3,321,800 97,700 34
2012 Q2 Jul-11 3,640,000 104,000 35

The numbers in red are my best guess, based on the known numbers.

So is there anything else we can use to indirectly infer the subscription numbers?

My Old Idea

Up until this blog, my proxy for the subscription numbers had been Revenue which is still reported every quarter. Essentially I assume the average revenue per subscriber per month and then use this to derive a subscriber count. Generally I pick the revenue PUPM to be around $50 which, for the latest quarterly report, gives us a subscriber count of a little over 8 million subscribers and, assuming the average size of a Salesforce customer is 35 users, this gives us a total of 230,000 customers.

The above estimates rely heavily on my guess for the revenue PUPM and the average Salesforce customer size. While consistent with known historical values, they are still a speculation. There turns out to be a better way.

Daily Transaction Numbers

A couple of years ago, Ross Dembecki (Silverpop/Core Motives/CWR guru and seriously nice bloke) sent me a message suggesting that, if I was interested in Salesforce’s growth, to consider monitoring their daily transaction levels, as published on their ‘trust’ page. It turns out Ross was on the money and it has only taken me two years to catch up. By scouring the internet and using this article, we get quite a few data points for Salesforce’s historic transaction levels.

Date Transactions
28/09/2006 53,000,000
09/10/2007 117,000,000
2009 150,000,000
20/05/2009 195,000,000
10/11/2010 394,000,000
11/2011 500,000,000
17/11/2011 623,000,000
11/2012 1,000,000,000
08/2013 1,300,000,000

Plotting the transactions against the known subscriber levels gives us this.


The x-axis is the number of subscribers and the y-axis is the number of transactions.The “R-squared value” tells us how related the two sets of data are. In this case, there is a 97% match, suggesting the two are very correlated.

What is more, the equation of the closest fitting line (y=150.27x-40,000,000) tells us that for each subscriber that comes on board, the daily transaction count goes up by about 150.

Doing the same for the customer numbers gives us this:


Predictions From the Graph

We know that in August 2013, there were 1.3 billion transactions. Using our linear equation, this suggests that in August 2013, Salesforce had just shy of 9 million subscribers. Looking at the trust site, Salesforce is on the cusp of a 2 billion transaction day, which puts subscribers at around 13.5 million. In terms of customers, we get about 383,000 customers.

This is a lot more that the 8 million subscribers I predicted via the revenue which means one of my assumptions is wrong e.g. my predicted revenue PUPM is too high. In fact, at 13.5 million subscribers, the revenue per user per month is closer to $30. The numbers also predict the average customer size is 35, about one third of the average customer size of Dynamics CRM (4,000,000/40,000 = 100). This average company size is consistent with the known numbers where the last known average company size of 33 (and increasing). Also, the $30 revenue prediction is consistent with the known numbers where the last known value was $51 and falling.


Unless Salesforce release actual numbers, all of this is speculation but, using the transactions seems to be the best approach with minimal assumptions. The numbers predicted are consistent with the known data and their trends from three years ago. While the transaction numbers predict a very large subscription base (three to four times that of Dynamics CRM), if they are correct, Salesforce make a surprisingly small revenue from their subscribers and the average size of the companies using Salesforce is also surprisingly small.


Unknown said...

I'm making an assumption that a transaction equals a CRUD action (Create, Read, Update, Delete). Based on that assumption, I'm wondering if other factors play into the transaction numbers like power users, integrations......

Leon Tribe said...

It is difficult to know the composition of the transaction count; I am unaware of a formal definition. There does appear to be a strong correlation between the transaction numbers and the subscribers/customers for the first five years though.

The assumption in my analysis is that this correlation remains today.