Best Customer Fail

How often have you heard quotes like this?

• "Multi-Channel customers are the best customers."
• "Your best customers participate in Social Media, so you have to join the conversation so that you can reap the rewards."
• "The biggest spenders love loyalty programs, so you need to reward best customers to remain competitive in today's highly complex retail landscape."

Putting these comments up on Twitter is like throwing an aerosol can into a bonfire.  BOOM!!  The comments go viral, they're accepted without any analytical rigor.  Folks seldom question the data, because the comments sound so darn logical that they must be true.


If you have access to customer data, you can do some research on your own.  And when you do some research on your own, you learn about a curious pattern that happens when drafting simple queries based on "best" customers vs. all other customers.


Look at the data in the table at the start of this post.  Let's query the table.  We'll calculate the average amount of spend a customer had if they possess the attribute, and the average amount of spend the customer had if the customer does not possess the attribute.  Here we go.


Lives In An Urban Zip Code:

• Yes = $900.  No = $475.
• Strategy:  Find urban customers, because urban customers are your best customers.  Or encourage your customers to move from a rural environment to an urban environment, because they will spend $425 more if they move, based on your findings.

Pays With A Visa Credit Card:

• Yes = $900.  No = $475.
• Strategy:  Partner only with Visa on special promotions, ignoring all other tender types, because Visa purchasers are your best customers.  Or offer a promotion to get AMEX customers to switch to Visa, because you'll benefit from the switch.

Follows You On Twitter:

• Yes = $733.  No = $600.
• Strategy:  Give discounts and promotions to Twitter followers, because those are your best customers and they will evangelize your brand for you.

Has an iPhone:

• Yes = $850.  No = $580.
• Strategy:  Ignore Android and focus only on Apple, because your best customers love Apple.  Apple customers outspend other customers by $270 each.  Or better yet, encourage Android customers to switch to the iPhone, because it will benefit your brand.

Is In Your Loyalty Program:

• Yes = $967.  No = $425.
• Strategy:  Get all of your customers to participate in your loyalty program, because customers in the loyalty program outspend non-members by more than $500.

Buys In Stores And Online

• Yes = $733.  No = $600.
• Strategy:  Fully invest in a bricks 'n clicks environment like Borders or Blockbuster, because your best customers shop multiple channels.

Responds To Offline Marketing

• Yes = $800.  No = $600.
• Strategy:  Develop a catalog marketing program that is targeted to your best customers, because best customers love receiving paper in their mailbox.  Or try to force your customers to shop from a catalog, because they will become better customers.  Explore television commercials, billboards, and radio advertising as well, because your best customers are offline mavens.

Pays For Expedited Shipping

• Yes = $1,033.  No = $375.
• Strategy:  Force all customers to pay for expedited shipping, because when you do that, you end up with customers who spend 2.9 times as much as customers who don't pay for expedited shipping.

Customer Returns Merchandise For A Refund.

• Yes = $975.  No = $233.
• Strategy:  Give a customer $10 off their next order if they return merchandise today, because customers who return merchandise spend $700 more than customers who don't return merchandise.  By providing a financial incentive to return merchandise, you will increase customer value.

Customer Enjoys Playing Yahtzee!

• Yes = $700.  No = $600.
• Strategy:  Send every customer in your database a free Yahtzee! game, because customers who possess a Yahtzee! game outspend all other customers by $100.

Have you ever attempted this analysis in your own customer database?


It's a fascinating analysis, because for 85% of the attributes you analyze, you'll find that customers who possess the attribute outspend customers who do not possess the attribute.  It's a mathematical "dilly of a pickle", if you will ... it turns out that your best customers tend to do everything, and that skews the results, causing nearly every query to exhibit a "best customer bias".


Seriously, go perform this analysis on your own customer database.  If you're a vendor, go run this analysis on your customer base, you'll see the same thing.


"Best Customer" analytics that depend upon simple queries provide highly misleading results ... in fact, the outcomes are utterly impossible to decipher.

And you know this is true.  If it were not true, then you would have capitalized on all of the ideas that you read in trade journals, blogs, Twitter, you name it, and your annual retention rate would have increased from 45% to 60%, right?  But you go back a decade, and you see that you've added channels and loyalty programs and advanced analytics and increased online spend and participated in offline marketing programs and social media and mobile and ... well ... you haven't increased your retention rate ... and spend is up at an inflationary rate.

Oh boy.


A better, but more complicated analysis, is to pick a period in time (say January 1, 2010).  Freeze customers as of 12/31/09, tabulating customer attributes at that time.  Then regress all attributes (like the ones in the table above) against spend during 2010.  If the coefficient for an attribute is significant and positive, then you know that the attribute contributes positively to future spend.

I know, it's humbling to know that everything you've been taught isn't yielding profitable increases in customer retention.  It's frustrating to know that the secret to success is having a profitable customer acquisition program.  Run your own analytics against your own customer database, and learn for yourself why traditional "best customer" analyses are hopelessly flawed.

Multichannel Customer Value: A Case Study

Click on the image to enlarge it.

We make a lot of decisions on the basis of multichannel customers. And why not? Our industry leaders tell us our customers demand a seamless multichannel customer experience.

For me, frustration boiled over this past week, when industry leaders cheered J.C. Penney's decision to consolidate marketing and merchandising functions across channels.

When did it become so fashionable for industry leaders to cheer the loss of jobs, all in the name of operational efficiencies that allow vendors to profit from the sale of multichannel solutions?

The phrase that vendors, industry experts, research organizations, and trade journals most frequently use to promote a multichannel agenda is this one:

"Multichannel customers are the most valuable customers".

We now know that this statement isn't accurate. Business Intelligence teams that use fair queries, controlling for other factors, do not observe this relationship.

The analytics used to defend the statement are highly biased. The metric is backward looking, not forward looking. Analysts query a database, splitting customers into two groups --- those that purchased from multiple channels last year, and those who purchased from a single channel last year. Next, the analyst computes the mean of last year's net sales in each group.

By default, the multichannel group will have spent more. The query is designed to make this happen. A single channel customer is disproportionately skewed toward one purchase. A multichannel customer, by definition, had to purchase at least two times. This greatly biases the results of the query.

The bias benefits our entire vendor industry, from those offering inventory solutions to those promoting the use of paper advertising to those offering e-mail solutions to those promoting pay-per-click marketing to those providing website marketing products and services.

Now honestly, you might find that multichannel customers are your best customers. But it will be because your unbiased queries prove they are your best customers, not because somebody surveyed 849 customers and extrapolated the results to 300,000,000 residents of the United States.

What happens if we try to eliminate this bias?

The image at the start of this post illustrates future twelve month net sales, based on customer activity in the past twelve months for a client. Sure, this metric will be viewed by some as being biased. Maybe I'll spur a discussion that improves how we view our field.


Here's how you run the query.

Step 1: Identify all customers who purchased from your brand in 2006.

Step 2: Within this audience, select customers who were customers prior to 2006 (in other words, eliminate all new customers in 2006).

Step 3: Identify the top 25% cutoff point for spend in this audience in 2006. Say that amount is $350. Keep that amount in mind for subsequent analyses. Select only customers who spent at least $350.

Step 4: Within this universe of great customers, calculate the average number of channels the customer purchased in during 2006.

Step 5: Calculate the mean net sales spent by this audience in 2007. If a customer did not purchase in 2007, the customer spent $0.

Step 6: Repeat steps 1-5 for all prior years, using the $350 cutoff (or whatever your dollar cutoff is for really good customers) in each year you run the analysis.

Step 7: Plot average channels purchased from (x-axis) by mean net sales in the next twelve months (y-axis).

Step 8: Review your graph (the graph attached to this post).


If the experts are right, then this graph should have a linear relation, with future sales increasing as prior channels increase.

In this example, we don't observe a linear trend, do we? In fact, the correlation is negative. Among really good customers, years where there was high multichannel activity were followed by years of lower spend.

What caused increased spend? Merchandise productivity! When the brand offered great products, customers spent more. When the brand didn't have great products, customers spent less. Channels didn't play a significant role in increasing or lowering customer spend.

Our industry demands that we improve the multichannel customer experience, offering little proof that customers will spend more.

Data consistently tell us that when customers love the merchandise we offer, customers spend more.

Focus on merchandise, the main reason customers buy from your brand. Then allocate proportionately fewer resources to the vendor / expert / research / trade-journal agenda of multichannel excellence.

Best Customers

There is an amazing difference between "new" forms of direct marketing (like paid search) and classic direct marketing.

Paid and natural search is all about intercepting a customer when the customer has a need. You don't necessarily know if this customer is your most loyal customer, or one who has never heard of your brand.

Classic direct marketing is all about "pushing" a message at a customer. And in order to get the best return on investment, it is perceived that it is best to "push" your message at "best customers".

Take your average multichannel retailer. If you believe in the old adage that "multichannel customers are your best customers", then your marketing efforts have to be focused on these individuals.

• If you are launching a new catalog title, you'll send it to this individual.
• If you are moving from one e-mail campaign per week to two e-mail campaigns per week, you'll want the new campaigns blasted to the best customers, giving them the best chance to succeed.
  
• If postage costs are chewing up your profits, you are unlikely to pull back spend among your best customers, who spend enough to offset increased costs.
• If you're having a store event, you'll want to notify your best customers about the event.
• If you're opening a new store, you'll want to notify the best online/catalog customers in that trade area about the new store.
  
• If you have a loyalty program, you want to reward your best customers, right?

In each situation, direct marketing activities are going to be focused on "targeting" best customers.

The majority of companies employ this type of strategy --- each marketing campaign is unique --- ignoring the targeting strategies being employed by future/concurrent/prior campaigns.

Is the "best" strategy one that focuses on "best customers"? We'll talk a bit about this topic this week. You are encouraged to offer your thoughts.