Retail Stores

Here's a quote I now use on Twitter.

• "A store is a purpose-built structure designed to extract money from customers and prospects who want to be there."

The quote leads us to this article:  Sprouts - a grocery store, is closing large stores and opening small stores.

And then another article worth your time - this is so similar to what happened to in-store shopping, via the movie industry (click here).

What's coming for retail is, on one hand, going to be messy and unpleasant. But on the other hand, what retail professionals currently under the age of 45 are going to do to reinvent retail ... that's going to be amazing! Smaller stores aren't the answer, they're a step in getting to the answer. The movie theater article above? That's a parallel to turning a store into a distribution center ... it's not the answer, it's a step in getting to the actual future.

The actual future? I'm excited about it!

Odd Quirks in Category Targeting Models

Here's one of the odd quirks you find when creating Category Targeting Models. This brand has twelve (12) main categories. When I decile my findings, I see oddities. Look at this customer (decile 10 is best, decile 1 is worst).

• Category 01 = 10.
• Category 02 = 10.
• Category 03 = 10.
• Category 04 = 8.
• Category 05 = 10.
• Category 06 = 10.
• Category 07 = 10.
• Category 08 = 10.
• Category 09 = 10.
• Category 10 = 10.
• Category 11 = 10.
• Category 12 = 10.

You'd want to target this customer with EVERY category, correct? Yeah! What's interesting is that this customer possesses weighted dollars within Categories 1/6/7/8/9/10/11. That's a really good customer. And all of that cross-shopping suggests that the customer would also be a prime candidate for Categories 2/3/4/5/12, even with no prior purchase history in those categories.

This is not an uncommon outcome.

Your best customers should be exposed to EVERYTHING you sell, period.

It's the marginal customers where targeting makes a huge difference. Look at this customer.

• Category 01 = 2.
• Category 02 = 2.
• Category 03 = 3.
• Category 04 = 4.
• Category 05 = 2.
• Category 06 = 4.
• Category 07 = 4.
• Category 08 = 2.
• Category 09 = 2.
• Category 10 = 7.
• Category 11 = 2.
• Category 12 = 2.

There you go ... target that customer with Category 10, right?

Each brand possesses some odd quirks in their Category Targeting Models. Figure out what those quirks are, and when you are targeting your customers take advantage of the odd behaviors you observe.

Category Targeting Model Framework

I've explained this previously, and the statisticians in the studio audience are going to belly-ache about what I'm about to explain, but here's what I do ... I'm transparent about my methodologies.

Step 1:  Select twelve-month buyers as of 12 months ago.

Step 2:  Calculate historical spend in 0-12, 13-24, 25-36, and 37-48 month buckets.

Step 3:  Calculate future spend in the next twelve-months.

Step 4:  Regress the four variables in Step 2 against the dependent variable in Step 3.

Step 5:  Evaluate the coefficients for the 0-12 month spend variable, 13-24 month spend variable, 25-36 month spend variable, and the 37-48 month spend variable. Let's pretend that they are as follows ... 0.604 ... 0.369 ... 0.235 ... 0.134.  Divide each value by the value of the 0-12 month coefficient (0.604), giving us 1.000, 0.611, 0.389, 0.222.

We now have our weights ... we weight 0-12 month dollars by 100%, we weight 13-24 month dollars by 61%, we weight 25-36 month dollars by 39%, and we weight 37-48 month dollars by 22%.

Step 6:  For each category, we calculate weighted dollars spent during the past four years, as of one year ago.

Step 7:  For each category, we calculate total dollars spent in the past year.

Step 8:  We run a regression for each category ... the independent variables are weighted dollars spent in each category, the dependent variable is amount spent in the past year for a specific category. This way, we learn which categories contribute to future spend within a specific category.

Step 9:  We repeat Step 2, shifting our date ranges one year forward.

Step 10:  We score each customer for each category for the next year.

Step 11:  If desired, results are placed into deciles for each category.

At this point, you have completed Category Target Modeling ... you now have targeting variables for every category. If you wish to send an email campaign to Mens Outerwear customers, you use the Mens Outerwear Category Targeting Model results.

Again - statisticians are gonna grumble about the methodology. Let them come up with their own ... this methodology is simple and effective and proven to work in the real world.

P.S.: Yes, you can use logistic regression here as well with a 1/0 dependent variable. Works well.

Category Targeting Models

If you want to develop your categories, you'll need two things.

1. A style/sku plan to properly manage winning items, new items, and existing items.
2. A marketing plan to target merchandise to customers most likely to purchase from various categories.

Next week we'll talk about "Class Of" reporting, reporting that helps us see what problems we have within various categories.

This week, we'll talk about Category Targeting Models. All of you should have Category Targeting Models, and when you feature anything you should feature it to the customers most likely to purchase it (though exposure is important as well).

Tomorrow, I'll share with you my Category Targeting Model Framework.

Loyalty Program Expectations

There is a reason so many retail brands tie loyalty programs to credit programs. The goal isn't necessarily to increase customer spend - the goal is to increase the amount of interest charged to a customer.

In my work, it is generally true (your mileage will vary) that loyalty programs increase customer spend by around 10%, plus/minus.

For a loyalty program to work, you need high value customers. Pretend you have a typical e-commerce customer with a 30% chance of buying again, purchasing 1.6 times at $100 each if the customer repurchases.

• Normal Conditions = 0.30 * 1.6 * 100 = $48.00 of expected spend next year.
• Loyalty Program = 10% Bump = $48.00 * 0.10 = $4.80.

It doesn't take a rocket scientist to notice that increasing spend by $4.80 a year per customer in a loyalty program won't do anything, after you subtract all normal business expenses and then subtract marketing expenses associated with a loyalty program.

Take a customer with a 70% annual rebuy rate, 5 purchases per year at $100.

• Normal Conditions = 0.70 * 5.00 * 100 = $350.00 of expected spend next year.
• Loyalty Program = 10% Bump = $350.00 * 0.10 = $35.00.

You can make an argument that $35.00 is a credible amount of increased spend, so yeah, have at it, create a loyalty program for this customer.

The problem, of course, is that you don't have many of these customers.

On average, a customer needs to achieve a fifth purchase before the customer has a 60% or better chance of repurchasing next year. Once a customer has a 60% chance of buying next year, the customer generates sufficient net sales to make loyalty programs both noticeable and meaningful (see my first example above for a situation that is not noticeable and is not meaningful).

Here's an exercise you can perform with your own data.  Segment all twelve-month buyers by number of life-to-date purchases as of one year ago today. Select all customers with 5+ life-to-date purchases from that audience. Then calculate total spend in the next year from that audience. Finally, multiply that number by 10%. That's the expected amount of sales increase a credible loyalty program "might" deliver. Again, your mileage will vary.

For a typical e-commerce business, the audience of 0-12 Month 5x+ buyers generated maybe 20% of annual sales, plus/minus. Multiply that amount by 10% and it means a credible loyalty program "could" add 2% to annual net sales levels.

This is why large retail brands tie loyalty to credit ... it allows them to make profit off of the interest.

This is why I advocate for e-commerce brands to simply go find another new customer ... the math works out so much better over time.

Statistical Significance of A/B Tests

There is a ton of misinformation out there about A/B testing. Those lacking rigorous statistical training tell you that you need "x" responses for a valid A/B test. 

That's not how this stuff works.

More than thirty years ago (Lands' End), we developed an equation to estimate the variance associated with our A/B mailing tests. It turned out that the variance of our estimates was non-constant. In other words, the variance might be "x" when the dollar-per-book was $3.00 ... it might be "1.5x" at a dollar per book of $5.00.

We developed an equation ... as long as dollars-per-book was >= $2.00 variance could be estimated as -188 + 192*x, where "x" was the dollars-per-book in a test group (aside ... there are going to be statistical experts who balk at creating this equation ... one that accounts for non-constant variance ... and will say that everything that follows is garbage ... just want you to know that view is out there, I need to be forthright here).

Let's pretend that our control dollar-per-book was $3.00, and we expected the test dollar-per-book to be $3.25. Let's pretend that we wanted 10,000 customers in the test group and 10,000 customers in the control group. Would our results be statistically significant?

The t-test equation looked like this:

• Test Group Dollar-Per-Book = $3.25.
• Control Group Dollar-Per-Book = $2.75.
• Test Group Sample Size = 10,000.
• Control Group Sample Size = 10,000.
• Variance of Test Group = -188 + 192*3.25 = 436.
• Variance of Control Group = -188 + 192*3.00 = 388.
• T-Score = (3.25 - 3.00) / SQRT(436/10000 + 388/10000) = (0.25) / (0.29) = 0.86.

The T-Score is nowhere close to 2.00 ... so the results are not statistically significant.

Now, does that mean that the results aren't meaningful? Maybe. What would happen if we had 100,000 customers in each of the test/control group?

• T-Score = (3.25 - 3.00) / SQRT(436/100000 + 388/100000) = (0.25) / (0.09) = 2.78.

Now the results are statistically significant.

What was the difference?

Well, the original sample size was too small.

We used the equation above to determine the appropriate sample size for all tests based on the amount of variance associated with our expectations for test group performance and holdout group performance.

When I worked at Nordstrom, we used a comparable equation - one specific to Nordstrom. We learned that we needed 100,000 or 200,000 customers to measure what we wanted to measure ... not 10,000 or 20,000.

Comparable issues impact website conversion. You don't want to measure conversions ... you want to measure sales per visitor. Your test group might spend more/less than the control group once the decision to purchase is made ... so you have to measure sales per visitor. 

Follow the math.

There Goes Google ... A Channel All By Itself

In a recent project, here's what the data told us about Google.

• 7.8% of sales were derived from one specific Category.
• 20.4% of all sales attributed to Google were from this category.
• 75% of sales attributed to Google were from first-time buyers.

Again, you can preach a seamless/frictionless customer experience, and that's fine. Preaching sameness?

Never.

Google does what it wants, and it attracts customers with specific interests. Your bidding strategy interacts with customers possessing specific interests. The result is a customer that is not representative of your overall category mix among customers.

Get to know how customers, channels, and categories interact. Then do something about what you learn via all of the marketing tactics you possess. Or hire me (click here please) and we'll get busy discovering all of these interesting relationships.