SEO Werkt!

Are you one of millions of marketers/analysts trying to prove that engagement exists, and more important, leads to increased return on investment?


The pick up a broom and start sweeping … I’ll show you a method that loyalty marketers have been using for twenty years to prove that loyalty marketing delivers a return on investment.  I use more technical versions of this methodology to calculate the value of loyalty programs for current clients, running a full profit and loss statement on the outcome of this analysis.  Hint:  It works!


Step 1:  Create your engagement measure.  This will be a different metric for everybody, so there’s no sense spending time discussing it, you are the expert at knowing your business.  Customers who you consider to be “engaged” receive a value of “1″, while customers who you do not consider to be “engaged” receive a value of “0″.  Only use the timeframe up to 10/31/2011 for your engagement period.


Step 2:  Create RFM-based variables.  For each customer, through 10/31/2011, calculate months since last purchase, number of 12-month purchases, number of 13+ month purchases, and historical average order value.


Your audience is comprised of all customers with 1+ purchase (via Step 2).


Step 3:  I will assume that you don’t have profitability data, so let’s make this really easy.  Create a variable called “Future” … it has a value of “0″ for all customers who did not purchase from 11/1/2011 to 11/30/2011 … it has a value of “1″ for all customers who did purchase between 11/1/2011 to 11/30/2011.


Step 4:  Match the query in Step 3 to the query in Step 2.  Then, match these queries to Step 1, all at a customer level.


Step 5:  Run a Logistic Regression (you can take this much further if you have profitability data … Logistic Regression for response, OLS for spend/profitability).  Regress Future against Recency (usually Square Root of Recency), 0-12 Month Orders, 13+ Month Orders, Average Order Value, and Engagement.


If “Engagement” is a significant predictor with a positive coefficient, then you just proved that, for the month of November, engagement during October led to an increased probability of a customer purchasing in November.


Ok, here’s the SPSS code required to run the Logistic Regression procedure I described above:



LOGISTIC REGRESSION VARIABLES future 
  /METHOD=FSTEP(WALD) root_recency freq12 freq99 average_order_value engagement 
  /CRITERIA=PIN(.05) POUT(.10) ITERATE(20) CUT(.5).
execute.

Here’s the outcome of a trial I ran earlier today:

In this example, the square root of recency is easily most important.  Recent orders carry about 5 times the weight of older orders.  Large dollar orders result in customers less likely to buy again in the future.  And most important (while not terribly significant), customers who were “engaged” in October were more likely to buy in November, after accounting for all other RFM-based variables.


In fact, “engaged” customers were 22.7% more likely to buy again, all things being equal.


Armed with this outcome, we can put a profitability number on “engagement”.  We’ll make the financial analysis terribly simple, for demonstration purposes.


Let’s assume that we have 250,000 customers in the database.  Of the 250,000 customers, 5,000 are considered “engaged”.  Let’s assume that our 250,000 customer database has a 3% chance of buying in November.  And let’s pretend that, if a customer purchases, the customer will spend $100.  Finally, let’s pretend that 35% of demand flows-through to profit, and let’s pretend that one employee is responsible for improving engagement, at a cost of $8,000 per month (salary + benefits).


Total Expected Housefile Demand = 250,000 * 0.03 * $100 = $750,000.


Now, we know that 5,000 engaged customers are 22.7% more likely to purchase because they are “engaged”, right?  So, our calculation changes a bit.


Total Expected Housefile Demand = (245,000 * 0.03 * $100) + (5,000 * 0.03 * 1.227 * $100) = $753,405.


The impact of engaged customers is … $753,405 – $750,000 = $3,405.


At 35% profit, this translates to $3,405 * 0.35 = $1,192 profit.


But, we hired an individual to generate engagement, and we paid the employee $8,000 to get the job done in November.


How many “engaged” customers do we need to make the effort worthwhile?  Well, we have 5,000 engaged customers who generated $1,192 profit, prior to employee costs, or $0.2384 profit per customer.  To offset employee costs, we need 8,000 / 0.2384 = 33,557 engaged customers.


So, at this point, here’s what we know:

1. We demonstrated that engaged customers are 22.7% more likely to purchase, all things being equal.  In this example, Engagement does lead to improvements in customer loyalty. And isn’t that what you really wanted to demonstrate?
2. We only have 5,000 customers meeting “Engagement” criteria.  As a result, we only generated $3,405 of incremental demand.
3. After accounting for employee costs, our engagement efforts are not generating a positive ROI.
4. Every engaged customer is worth an additional $0.24 profit, per month, to the company.
5. If subsequent engagement activities result in having at least 33,557 engaged customers, we can demonstrate that increased engagement can be accomplished at break-even levels.

We are entering into a new era, ladies and gentlemen. Well, “era” may not be the right word considering how quickly things change in these here mobile parts, but the fact remains the same: Quad-core mobile processors are here. And the ones that aren’t quite here yet are coming.

While many of our brilliantly geeky readers need no tutorial on the advantages of four processing cores, some of you may be thinking “Uh… OK, why do I care?” So I took it upon myself to place a few calls and get some of the big guns — Qualcomm, Nvidia, and TI — to explain why exactly you should care (or shouldn’t), and what kind of differences technology like this can make in the average user’s daily phone usage.

Right off the bat, there are a few myths we need to squash, the most prominent being the misguided belief that doubling cores automatically doubles processing performance. That’s not so true. Upgrading from a single-core CPU to a dual-core processor yields 50 percent better performance, while upgrading from dual-core to quad-core increases performance by just 25 percent. The second commonly held but utterly untrue belief is that all mobile processors are created equally. These companies actually work extra hard to differentiate themselves, which is difficult when the end-user has little say over which processors get stuck in which devices.

Generally speaking (as in, with no particular brand or model in mind), a quad-core CPU should most noticeably do two things. The first is to improve performance during multi-tasking or use of multi-threaded applications. Web browsing, for example, is a multi-threaded process, as are many advanced gaming applications. Android is also natively multi-threaded. The second noticeable improvement quad-core should yield is an increase in battery life. Now, your average CPU usually only consumes about 15 percent of your battery life during regularly daily usage, so the improvements won’t usually be that staggering. Still, battery life is a big problem right now in mobile and any improvement is a worthwhile one.

Nvidia has been the first to bring quad-core processing to mobile, in the form of its Tegra 3 Kal-El SoC. Aside from the general benefits afforded by four cores, Nvidia specifically differentiates itself with what it calls a Companion core. The Companion core is a patented fifth core that maxes out at speeds of 500MHz. It uses patented technology known as variable symmetry multiprocessing (vSMP), which allows the processor to power cores on and off based on the device’s workload.

The Companion core handles just about everything during low performance tasks and in stand-by mode, like email and monitoring the network for incoming calls. When you start on something more performance-intensive, like web browsing, facial recognition or photo stitching, other cores are powered on to handle the task. This is Nvidia’s way of improving performance while saving battery life, while others have found different ways to make quad-core stand out.

Qualcomm, for example, is about to release its APQ8064 SoC, which has a special trick. Most multicore processors clock up and down at the same time. Qualcomm’s processor, on the other hand, is able to clock one core at the max while clocking the second needed core only to the speed it needs to complete the task.

In other words, since Qualcomm’s processor cores can be clocked individually, a task that overflows on the first core may only need the second core spinning at 60 percent of its max speed. So just like Nvidia’s Companion core hooks you up on the battery life front, so will Qualcomm’s individual clocking technology.

Texas Instruments, however, has yet to outline plans for their quad-core offerings and seems to be sticking with dual-core OMAP SoCs for the time being. That said, TI maintains that its OMAP 5 SoC equipped with a dual-core Cortex A15 processor (and two Cortex M4 cores) is a mature system that is more efficient at handling instructions. Some even refer to it as a quad-core system, though TI itself still calls this a dual-core SoC. And they believe it’ll compete. The company went so far as to say that its smart multi-core architecture actually takes 30 percent more instructions than the Cortex-A9 MPCore’s four processing cores as seen in Nvidia’s Tegra 3.

The truth is this is just the beginning when it comes to the migration toward four cores, and there’ll be plenty more to learn in the coming months.