[ Home ]    [ FAQ ]    [ Download ]    [ Contact / About / Privacy ]

Get the Drilling Down Book!
http://www.booklocker.com/jimnovo
Now also available online through
Amazon and Barnes & Noble

Check out:

The Marketer's Common Sense Guide to E-Metrics - 22 benchmarks to understand the major trends, key opportunities, and hidden hazards your web logs uncover.  I wrote this manual with Bryan Eisenberg of Future Now, the visitor conversion specialists.  Download a free white paper on the topic:
Marketer's Common Sense Guide to E-Metrics

Prior Newsletters:
http://www.drilling-down.com/newsletters.htm
-------------------------------

In This Issue:

# Topics Overview

# Best of the Best Customer Marketing Links

# Tracking the Customer LifeCycle: Recency

# Questions: Data Mining
--------------------

Topics Overview
=============

Hi again folks, Jim Novo here.

This month we've got the usual "best of" Customer Marketing article links, oriented towards retail in recognition of the season.  We also kick off the first Real World Example using the Recency Metric with The Web Retailing Example, and a fellow Driller wants to know what the heck Data Mining really is and when one should use it.

Let's do some Drillin'!

Best Customer Retention Articles
====================

This section flags "must read" articles moving into the paid archives of trade magazines before the next newsletter is delivered.  If you don't read these articles by the date listed, you will have to pay the magazine to read them from the online archives.

Note to web site visitors: These links may have expired by the time you read this.  You can get these "must read" links e-mailed to you every 2 weeks before they expire by subscribing to the newsletter.

It's Outta Here
Expires November 30, 2002  Catalog Success
Timely to the season, many people just don't understand how to get rid of merchandise that does not sell.  It's not always best to just hack the price and sell it yourself; you can ruin your price credibility and turn your customers into low margin bargain hunters.  10 tips for getting rid of merch that doesn't move.  And try using customer behavior to segment your offers.

A Housefile Divided
Expires November 30, 2002  Catalog Success
Too many people just blast out the same offer to everybody on their e-mail list.  Sure, it may be easier but it is very sub-optimal, you are leaving margin dollars on the table.  If you start segmenting your list and making different offers according to behavior, you will make a lot more money.

Tracking the Customer LifeCycle: Real World Examples
=====================
If you are new to our group and want to review the previous LifeCycle metric - Latency - that discussion is here, along with the Real World examples
Hair Salon and B2B Software.  The previous piece on Recency is here.

Recency: The Web Retailing Example

"Yes, things are sweet", thinks the owner of IMissAsia.com (not the real name of the site), ending a phone conversation with a supplier.  Who would have thought!  In just one short year, with the remains of the dot-com bust scattered all about, IMissAsia.com was running at sales of $25,000 a month.  With an operating margin of 40%, "The Miss", as the owner liked to call the site, was going to generate about $120,000 this year pre-tax - certainly enough to keep the spouse warm, fed, and reasonably happy.

Who knows what next year would bring?  Could the business double in size?  The owner figured The Miss could do about $2 million in annual sales with the current infrastructure set up - a web site / shopping cart that cost about $40 a month and assorted software / hardware purchased for a total of about $1,000.  The owner put up the very simple web site and ships every box - the order management system is highly integrated with both the shopping cart and UPS WorldShip, so order processing and customer service are a breeze.  All other costs of the operation were basically variable to sales.  Sweet, indeed.

Of course, the owner / only employee has done a lot of things right in the first place, including the pursuit of a naturally web-friendly business idea.

IMissAsia.com is a site for people who used to live in Asia and now don't, and are scattered all over the world.  The core business idea takes natural advantage of what the web is very good at - aggregating niche demand. 

A person who used to live in Asia but now doesn't is cut off from a lot of things they liked and now "miss" - food, clothing, beauty items, culture.  IMissAsia.com aggregates all those things into one web site, and then offers it as a "one stop shop" to a very geographically dispersed group of people all over the world. IMissAsia.com is "patient zero" in all this, the intersection of diverse needs with a fragmented customer base - a perfect application for the web.

The site is set up smartly, using mostly free resources to attract and hold on to traffic - news feeds, free newsletters, and discussion boards.  The store is tightly integrated into all the content, so there are many opportunities to get visitors to take a peek at the merchandise.  The Miss gets pretty high natural search rankings for important search terms because it's a plain HTML site without a lot of script and database-driven components, and has been written carefully with search engine optimization in mind.

In other words, the site is a little cash machine that requires almost no maintenance.  Shipping packages, customer service related to those shipments, and the newsletter are about all the day-to-day work done on the business. However, storm clouds are on the horizon.

Response to the weekly newsletter is falling, and the owner is thinking of going bi-weekly or even monthly.  In addition, to try and keep response up, the owner has been discounting more aggressively in the newsletter, and this practice is starting to depress margin.  This situation is of deep concern to the owner, because the newsletter generates a big chunk of sales volume.

Niche markets are a double-edged sword. While they fit perfectly into the natural search-driven model of the web, by definition, niches are small.  This kind of business has a tendency to ramp up very quickly, but then plateau as the entire niche is discovered and filled out.  You can quickly capture 80% of the market, but then there is nowhere to go. 

 And as soon as you are successful, you will attract copycats, who chip away at your share, often undercutting your prices in start-up mode.  The copycats have now started to appear.  How will the owner grow the business when it already dominates the niche, and defend against the copycats?  Not to mention address the worrisome situation with the newsletter.

So, what the heck does any of this have to do with Recency?  Everything, fellow Drillers, absolutely everything, in more ways than you could possibly guess.  Next month, we'll find out just how important - no, critical - Recency is to the future success of IMissAsia.com.

If you would like to read the next installment of 
Recency: The Web Retailing Example, click here.

-----------------------------------------
If you are a consultant, agency, or software developer with clients needing action-oriented customer modeling or High ROI Customer Marketing program designs, click here.  If you are in SEO and the client isn't converting the additional visitors you generate, click here.
-----------------------------------------


Questions from Fellow Drillers
=====================
Jim's note: If you still don't know what RFM is and how it can be used to drive increased profitability in almost any business, read this.

Q: We're trying to develop an action-oriented customer retention program and are starting to build our trip wire metrics.  I've been "hit up" by some of these firms that specialize in data mining - all of the neural networks stuff - can you help me understand why I should NOT go with one of them, versus develop this internally using your approach?

A:  I don't know that my approach would be superior without understanding your objectives and a bit more about your data.  But I can give you some "need to knows" about data mining in general that should help you decide...

Note to the data mining community: please don't hate me for stating the case in plain English.  People really need to understand this stuff.  If you are different, then simply prove it to your prospects.  Love ya, I really do.

Understand that mining was originally developed for use by highly advanced modelers, not beginners.  These are people who have used every human-based tool to do modeling and have squeezed every bit of info out, but want more.  The only way to get more is to use a machine to look for patterns.  The fact these services are sold to people just beginning the data modeling process is not particularly honest, in my opinion.

Here is what may happen: you will pay the money and spend the effort and they will come back and tell you that the Recency and Frequency of the trades are key to defection - and say their miner figured it out.  And it did.  The fact you could easily have already known that with a few simple tests is beside the point.

It is the ability of mining to improve on human modeling by 1% that is valuable, the ability to "go granular".  If you have never done any human-driven modeling, you will see a tremendous benefit in using your first models generated by the miner.  But the real question is this: could you have generated something nearly as good yourself, at a fraction of the cost, using "universal" behavior models?

Mining works best (meaning, delivers highest ROI) under these conditions:

1.  Your data is super clean.
2.  You have an enormous amount of data, no logical way to aggregate it.
3.  You have done everything you can to understand the data on your own.
4.  You have tested theories and generated results that can be analyzed.

Reasons the above are true:

1.  It's a machine, it can't distinguish between  bad and good data.  Dirty data = dirty results, garbage in = garbage out.

2.  This one is tricky.  You probably think you have more data than you really do, and that it is more complex than it really is.  Example: Do you really think what stock was traded and how many shares were traded is meaningful to your end objective of predicting customer defection?  Doesn't it make sense that a slowing in aggregate volume of trades or the number of months since last trade would be more relevant to spotting customer defection?  I think so, and this approach means you would have much less data to analyze and the data is a lot less complex that you thought.  If you are convinced the detail is important, use data mining.  This is really a question of scale, and what data you believe is relevant, which leads to #3:

3.  You have to "train the miner".  This can take months.  Basically, it keeps running scenarios and you tell it when it is wrong.  This continues until it gets something right.  In order to provide these instructions, you have to know what is wrong and right, you should  have already run some analysis.  When it comes back and says "customer defected because they traded 100 shares of SUN in Sep 1999" you have to say "Um, I seriously doubt it".  If the mining company says  they "already have the training done", fine, but it's not on your data, so the granularity that is the real benefit of mining is lost.

4.  If you are looking for direction on "what to do", there has to be "outcomes" for the miner to analyze.  If you have not done any customer retention programs and generated results from the programs, there is no "result" the miner can tie to "why".

So my advice is this: I think it is always best to make sure you have generated enough meaningful data for the miner to analyze.  To do this, you have to do some basic modeling yourself and try some things, see how it goes.  Doesn't have to be very complex, just generate some activity along the lines of the objective you have.  Then if you are not satisfied, call in the miners.  At least you will then have the kind of data that the miner can use, and will know enough to be able give the miner instructions. 

Another alternative: hire a real human modeler for a time, and see what they come up with (not me, I use universal models or hire a modeler myself).  I guarantee a hand-built regression model will be much better than using RFM, and much better than what the miner will come up with, because human intuition plays a huge role at the beginning of the modeling process.  At the end of the process is where mining comes in, when it becomes an issue of "brute force".  

And if you do hire a modeler, be prepared for the first question they are likely to ask: do you have any results I can look at from universal models like Latency, Recency, or RFM? 

Jim

-----------------------------------------
I can teach you and your staff the basics of high ROI customer marketing using your business model and customer data, and without using a lot of fancy software.  Not ready for the expense and resource drain of CRM?  Get CRM benefits using existing resources by scheduling a workshop. 
-----------------------------------------

That's it for this month's edition of the Drilling Down Newsletter. If you like the newsletter, please forward it to a friend - why don't you do this now while you are thinking of it? Subscription instructions are at the top and bottom of the newsletter for their convenience when subscribing.

Any comments on the newsletter (it's too long, too short, topic suggestions, etc.) please send them right along to me, along with any other questions on customer Valuation, Retention, Loyalty, and Defection right here.

'Til next time, keep Drilling Down!

- Jim Novo

What would you like to do now?

Get the book with Free customer scoring software at: