Good Article, Bad Timing

Here's an article from the Omniture Blog (a decent vendor-based web analytics blog), with some interesting tidbits about Borders and online analytics. Too bad the interview features improvements at Borders, just one day after Borders announced sweeping management changes amid significant sales declines.

One of the big challenges for our industry is to prove that software or analytics improved business at a time when the rest of the business cratered. It is 100% possible that software/analytics made a difference. It has been my experience that almost nobody wants to hear that message.

Update: 11:36am 1/6/2009: This article suggests that customers at Borders maintained spend in Nov/Dec 2008, compared with 2007. If you are to believe this, then in order to achieve a -13% comp, new customer traffic had to completely fall off the table.

Vendors: Above I mention two articles that demonstrate flat or improved performance, at a time when the company is struggling mightily. If our industry is going to improve credibility, if anybody is going to believe in us, we have to find a different way to communicate.

Twitter Update: Two Weeks In

Thought you might like to know a little bit about the audience following my posts on Twitter.

• About 10% are big company CEOs, EVPs, or VPs, most at multichannel (catalog + online & maybe retail) companies.
  
• About 25% are Entrepreneurs, running their own startup businesses, and they represent a very diverse set of business models!
  
• About 30% are Consultants or members of the Vendor community.
  
• About 25% are either Web Analytics experts or Business Intelligence experts.
• About 10% are difficult to classify, or are in the Social Media field.

• About 20% of the audience is from Europe.
• The audience is concentrated in NY/NJ/PA, the Midwest, and Seattle/Portland.

• The audience has few members of the multichannel/catalog industry, or the e-mail marketing industry. Wow. Wow.
  

Remember, I am doing this different than the social media elite suggest Twitter needs to be done. My goal is to link to thought provoking articles, those I agree with, and those I disagree with. The goal is to get you, the loyal reader, to consider different concepts.

Twitter Page: @minethatdata, or the old-school http://twitter.com/minethatdata.

Follow along, or subscribe to the Twitter MineThatData RSS feed.

Modern Segmentation, Modeling, And Planning

Much of the segmentation/modeling/planning process involves predicting a future purchase, followed by the determination of an appropriate targeting strategy.

For instance, in this catalog example, we predict two things.

1. We predict the Response Rate to a future catalog.
2. We predict the Average Order Size for a segment being mailed a future catalog.
   

Based on these two predictions, and a forecast for the cost of mailing a catalog, we arrive at the following segment-level mailing prediction and profit/loss statement (after online/retail matchback):

	Prediction
Response Rate	1.8%
Avg. Order	$125.00
$ Per Book	$2.25
Flow-Through %	35.0%
Flow-Through $	$0.79
Book Cost	$0.70
Profit	$0.09

The marketing world of 2009 requires a different level of sophistication.

In the future, we will change the planning and prediction process. This segment will be split into two sub-segments.

1. Subsegment #1 = Customers with the same RFM-style classification, but never historically purchased using Paid Search, Affiliates, or Shopping Comparison Sites.
2. Subsegment #2 = Customers with the same RFM-style classification, but historically purchased using Paid Search, Affiliates, or Shopping Comparison Sites.

In each case, we'll measure future response, but we'll also predict the expected marketing cost associated with self-service customers using Paid Search, Affiliates, or Shopping Comparison Sites. If the catalog or e-mail drives customers to these micro-channels, we incur additional marketing expense. Here's the sub-segment prediction:

	Subseg #1	Subseg #2
Response Rate	1.8%	1.8%
Avg. Order	$125.00	$125.00
$ Per Book	$2.25	$2.25
Flow-Through %	35.0%	35.0%
Flow-Through $	$0.79	$0.79
Book Cost	$0.70	$0.70
Pred. Search/Aff/SC Cost	$0.02	$0.18
Profit	$0.07	($0.09)

In this example, Subsegment #2 generates additional expense, because they like to use Paid Search, Affiliates, and Shopping Comparison sites after receiving a catalog. Therefore, we have to predict what the amount of incremental expense is likely to be. The same level of prediction is required to properly manage future e-mail campaigns.

For Statistical Modelers, this opens up a whole new area of exploration --- it's like drilling for oil in areas where exploration was prohibited.

For the Catalog Circulation Director, this gives you the opportunity to fundamentally change the contact strategy for self-service online shoppers, while generating a boatload of profit for your brand.

For the E-Mail Marketer, you have a once-in-a-lifetime chance to motivate your Executive team to deliver e-mail campaigns to unprofitable customers less often --- and you'll have the proof!

For the vendor community, especially for matchback vendors, you have a whole new product you can develop --- one that integrates purchases and expenses in a holistic and actionable manner. Or maybe the folks at Coremetrics or Omniture can get a jump on the catalog vendor community, and take ownership of this new opportunity.

Best of all, all of you e-mail vendor employees who regularly read this blog have a chance to build an application that improves the profitability of e-mail marketing efforts for your clients --- a good thing!!!

Catalog ROI Is Overstated Because Of Search

Last week, we chatted about how E-Mail ROI is mis-calculated. My stats tell me that you found the article interesting.

Catalog advertising causes the same issues that e-mail marketing causes, often on a larger scale.

The typical catalog marketer matches paid search orders that occur within 30/60/90 days of a catalog mailing back to the catalog that the circulation team believes is responsible for creating the order.

However, the typical catalog marketer does not match back unconverted paid search expenses to the catalog responsible for causing unconverted paid search to happen.

Take a look at this profit and loss statement.


This is a fairly typical catalog profit and loss statement.

Notice converted paid search orders. These orders are matched-back to the catalog. Some catalogers match the paid search expense of those orders back to the catalog.

Almost nobody matches the unconverted paid search clicks back to the catalog that caused paid search to happen. In this example --- a reasonably honest assessment of a catalog profit and loss statement, the catalog caused 3,200 paid search orders to happen. However, at a 3% conversion rate, the catalog caused about 100,000 paid search clicks to happen.

The average cataloger does not allocate the cost of the incremental 96,800 unconverted clicks back to the catalog that caused the clicks to happen.

So three things happen.

1. The cataloger significantly over-circulates the catalog, because the additional expense is not allocated to the catalog driving paid search. The catalog marketing effort is less profitable than it appears.
   
2. The cataloger significantly mis-understands the impact of catalog marketing. In this case, circulating 1,000,000 catalogs caused 100,000 paid search clicks. The marketer fails to see that the catalog caused a 10% "engagement rate". This is a big deal --- the catalog is causing far more customer engagement than is typically measured.
   
3. A portion of the 100,000 paid search clicks result in purchases with the competition, reducing your Net Google Score.

Eventually, we'll create a database infrastructure that allows us to capture appropriate customer interactions. This will fundamentally change how we market to customers.

• We will attribute unconverted paid search clicks back to the customer/catalog combination, in our promotional history files. Instead of recording an $0.80 cost for the catalog, we'll record a $0.80 + $0.50 = $1.30 cost to the customer, incorporating the cost of the search. Ask your database, co-op, or web analytics vendor if they are able to do this for you.
  

• When we make mailing decisions (e-mail or catalog), we will make the decision based on the historical paid search expenditure of the segment we're considering. We won't send as many catalogs or e-mails to customers who augment their experience with unconverted paid search. This is a big deal, folks ... we'll be much more profitable when we make this transition.

• Example: Say your break-even on an $0.80 catalog is $2.50. Now you have a customer who loves to click on paid search ads when she receives a catalog. Your "real" cost of mailing the catalog is $1.30, driving your break-even over $4.00.

• Example: E-Mail marketing is essentially free, until it isn't free! The new e-mail marketing discipline will require us to make e-mail marketing decisions, at a segment level, based on anticipated paid search expense. All of a sudden, e-mail marketing is fundamentally changed --- the discipline becomes nearly identical to catalog marketing.
  

• Another Issue: We have the same problems with Affiliate Marketing and Shopping Comparison Sites. If catalog marketing drives a customer to an affiliate, and that affiliate skims 7% off the top of an order, the catalog needs to receive an expense penalty for driving demand to the affiliate.
  

We've spent a decade doing matchback analytics. Now, we need to provide the vendor community some leadership, so that matchback analytics account for the expense side of the ledger. We are continually making bad decisions because our database infrastructure fails to capture important information.

Who do you see doing this type of work out there, and what was the impact of this style of analysis?

Customer Resume

Take a look at this customer:

• One order, on October 29.
• Spent $200 on one item.

We like to classify customers in some sort of RFM segment. In other words, this customer is a 0-3 month, 1x, $150+ AOV customer. Based on this classification, we develop targeting strategies and we analyze business performance against this segment.

The customer resume demands that we compile a robust profile about the customer.

• Customer received a catalog on October 1.
  
• Customer did not return any of the merchandise ordered on October 29.
• Customer paid the standard fee for shipping.
• Customer did not use a discount code when placing the order.
  
• Customer is an e-mail subscriber.
• Customer only purchased one item.
• Customer ordered on a Wednesday, one day after receiving an e-mail campaign.
• Customer had a referring URL from a social media site.
• Customer ordered 28 days after receiving a catalog.
• Customer did not use a catalog key-code when ordering.
• Customer lives in a Zip Code Forensics zip code classified as "Online Bliss".
• Customer ordered from merchandise division "X", and did not order from merchandise division "Y" or "Z".
  
• Customer has not visited the website since.
• This was the first order placed by this customer.
• Customer lives more than 100 miles from a store.

The Web Analytics professional seldom deals with the customer, she is always evaluating anonymous outcomes, always looking into the past. The Business Intelligence professional is always summarizing customer data down to a small number of manageable dimensions. Mind you, this isn't the fault of the Web Analytics or BI guru --- software makes it difficult to create a "customer resume".

The Customer Resume demands that we evaluate the customer on the basis of everything we know about the customer. Here, we have a rich profile of who this person is, and if we know more about who this customer is, we can imagine what our relationship with the customer could be.

For instance, the fact that this customer ordered 28 days after receiving a catalog tells us that the catalog, at best, mildly influenced the order. The fact that this customer ordered on a Wednesday, one day after receiving an e-mail marketing campaign, tells us that the e-mail may have had a strong influence. This customer paid full price, a positive harbinger for long-term value. This customer only purchased one item, often a negative indicator of long-term value. This customer did not return to the website, which might be a negative harbinger.

The goal of the customer resume is to know as much about a customer as possible, to be able to learn about subtleties in customer behavior that may indicate changes in future behavior. We want to derive information about the customer, we don't want to reduce behavior to a small number of esoteric dimensions --- we can always do that later.

If you want to liven up a boring "Information Technology Steering Committee" meeting, bring the resume for just one customer to the meeting, hand it out to every attendee, and then ask each attendee to craft a communication strategy for this one customer.

Channel Differences

I recently read an article where the catalog marketing expert mentioned that customers who use the telephone to place orders are inherently "better" than online customers that purchase and do not visit again. The premise of the argument was to focus marketing efforts on telephone shoppers.

It is not uncommon for customers in different channels to have different motivations, and different levels of future value. When we attempt to break down silos and focus on "the customer", we harm our ability to think about the different needs of each customer. We need to get past this issue.

Customers using the telephone to place orders are often motivated by different needs. The telephone customer might need additional assistance. The telephone customer might want companionship. The telephone customer might feel unsure about buying a $300 item online, needing to ask numerous questions before pulling the trigger, while the online customer feels perfectly comfortable buying a $30 item on the website.

Analyze your customer data, and you're likely to notice that telephone customers purchase more items per order, and spend more per item than do self-service online buyers. This doesn't mean that phone customers are "better". Rather, they are "different". So treat them in a way that they want to be treated!

Many of you have children. Do you treat each child the same way? Is each child forced to wear the same clothing each day, or do you tailor the clothing to the unique needs of each child? Do you spend more time on homework assignments with one child, then spend more time at soccer practice with another child? You analyze the unique needs of each child, and then you tailor the experience for each child accordingly.

Why not do the same thing for your customers, across the many channels you now manage?

Churn Rates Mask Retention Problems

We frequently read about a metric called "Churn Rate".

For subscription-based services, the churn rate illustrates the percentage of customers who leave the service in any given month.

Typically, churn rates look favorable. A company will report a monthly churn rate of 6%, suggesting that it keeps 94% of subscribers each month.

But churn rate has a damaging cumulative impact that can be illustrated on an annual basis.

For instance, a 6% monthly churn rate yields an annual retention rate of just 48% (1 - 0.06) %^ 12.

In the Multichannel Forensics framework, monthly churn rates translate as follows.

• Retention Mode (60% or Greater Annual Retention) = Less Than 4% Monthly Churn.
  
• Hybrid Mode (40% to 60% Annual Retention) = 4% to 7% Monthly Churn.
• Acquisition Model (0% to 40% Annual Retention) = Greater Than 7% Monthly Churn.

In other words, if your monthly churn is less than 4%, you're in Retention Mode. If your monthly churn is greater than 7%, you are in Acquisition Mode.