An Update!

In the next few weeks I will unveil a new product offering ... "Hillstrom's Marketing Budget Experiments"! You can click here for project pricing (as well as seeing project pricing for other work I perform).

As always with new product offerings, the kinks aren't always worked out ... that's why you, the intrepid blog follower, get first crack at the offering at a reduced price. You help me work out the kinks, I help you achieve an outcome at a great value! For this project, you will pay 60.3% of the project fee. Stupid inexpensive. Let me know before the end of the week if you wish to be part of the pilot and save 40% in the process.

More data and examples begin tomorrow. Elements of Category Dynamics will be folded into our experimentation framework. We are headed toward the following product.

• Enter your monthly marketing budget by major marketing channel.
• Supply 5 years of purchase history. This is used to forecast cohort performance, ultimately calculating CLV at a segment level.
• My product allows us to vary marketing spend by month, seeing what the long-term impacts are of the decisions we're making today.
• My product produces a five-year sales forecast and calculates down to variable profit / contribution.
• We "experiment" ... seeing if there are combinations of channels, ad spend, and merchandise productivity improvements that yield the result you expect of your business over time.

Fun is coming!

The Best Categories & New Customers

For this brand, Categories 12 and 19 are the "suns" of their category ecosystem. Both categories yield high-value customers, as illustrated below. Customers buying from either category tend to reside in the top 40% of the value segments I created.

Smart companies do smart things. They focus on the categories that drive high-value customers, and within those categories they make sure that winners get proper attention and they make sure that new items are properly developed via low-cost / no-cost marketing scenarios.

Are Those December Newies All That Bad?

Well, maybe.

Remember when I mentioned that the brand we are studying acquired good customers in August and lousy customers in December? I plotted fraction of new customers from December by new customer quality. What does the image suggest to you?

Among the best newbies, 15% to 20% of 'em were acquired in December. In other words, December newies aren't uniformly "bad" ... there are plenty of fabulous new buyers acquired in December.

Meanwhile, the worst two segments (24 and 25) do not have a huge share of December newbies. In other words, December newbies aren't the "worst" ... though the percentages are very high for not-so-good segments from 19-23.

Remember when I mentioned that August newbies were "good"? Well, what does the graph below tell us?

Oh, isn't that interesting?

Turns out that August newbies are better ... until you get to percentile groups 22-25, where a ton of August newbies exist. What the heck is going on there?

Well, it turns out that there is one merchandise category that is largely offered in late summer, and that category delivers new customers who perform poorly. Take a look at Merchandise Category 15.

The customers acquired from this category perform poorly, don't they? And in a separate analysis I demonstrated that these customers were generally acquired between late June and early September ... with August being the primary month for sales for this category.

There is this bubbly interaction between what you sell, when you sell it, and how you market to customers. It's not something you can easily separate, and it's not something you come up with simple rules to deal with. You have a lot of interactions, and those interactions dictate future success.

You Want To Acquire The Good Customers, Right?

The brand we've studied this week acquires a lot of customers ... and those customers have differing levels of future twelve-month demand. Take a look ... the graph below shows 4%-tiles of customer quality on the x-axis vs. future twelve-month demand on the y-axis.

There are all sorts of good customers acquired by this brand ... the top 20% of the file delivers customers worth between $69 and $141.

The bottom 20% of the file delivers customers worth between $17 and $28.

You'd want to know which customers end up in the top twenty percent, and which new customers end up in the bottom twenty percent, correct?

All this stuff is straightforward, it gives you plenty of knowledge, and you end up protecting the future health of your business in the process. 

Get to know your new customers, ok?

Many Items or Fewer Items: The $100 Question

Yesterday I illustrated an example of a newly acquired customer in August - the customer spent $100 on three items.

The customer is valuable - worth $81.18 in the next year.

Let's experiment a bit ... instead of the customer buying three items at $33.33 each, how about we see what happens if the customer bought one item at $100.00.

This customer is worth less ... $76.96 in the next year. The difference happens within rebuy rates ... 43.9% for the customer buying three items in a first order, 41.6% for the customer buying just one item at $100. Remember - the AOV is the same, but the presence of multiple items in the order yield a customer with better future value.

In most of my projects, it is better to encourage a customer to buy more items in a first order, given a comparable AOV.

Marketing Budget Experiments: Future Value of New Customers

Category Development and Future Value meet up in a typical Marketing Budget Experiments project.

I commonly develop twelve-month future demand projections for customers with varying attributes in a first purchase. I build a Logistic Regression model to predict how likely a first-time buyer is to purchase again next year ... and I build an Ordinary Least Squares Regression model to predict how much a first-time buyer will spend in the next year if the customer repurchases. Coefficients from both models are input into a spreadsheet so I can experiment with various customer attributes.

You'll want to click on the image to see the numbers.

Essentially, I have a series of predictive inputs ... AOV on a first order, items purchased on a first order, share of demand from new items, share of demand from items selling below their historical average price point, month of first purchase are all included on the left side of the image above. The middle set of coefficients represent different merchandise categories that the customer could purchase from. The right-side set of coefficients represent different marketing channels that the customer purchased from in a first order.

So, this customer spent $100 on three items, buying only existing items at/above the historical price point of the items purchased. The order was in August, the customer bought from Category 19, and the customer bought from Marketing Channel 4 (which in this case was email marketing).

• 43.9% chance of buying again next year.
• $185.08 spent if the customer buys again during the next year.
• $81.18 future demand value.

The data is only useful if we compare the customer to another customer, correct?

Let's say that the attributes are identical, but instead of the customer being acquired in August the customer is acquired in December. Does the story change?

The story changes:

• 42.1% chance of buying again next year.
• $153.99 spent if the customer buys again during the next year.
• $64.90 future demand value.

Acquire a customer in August and get $81.18 in future demand value.

Acquire a comparable customer in December and get $64.90 in future demand value.

The foundation of Marketing Budget Experiments is this work ... we first need to understand how valuable different customers are. Once we understand how valuable the customers are, we can simulate five-year customer value and determine the optimal monthly marketing spend to achieve our business goals. And yes, Category Development plays a key role here ... we'll see later this week how important different Categories are to building high-value customers.

A Delightful Outcome of a Life Table Analysis

Life Tables are an integral part of any Customer Development work performed by e-commerce brands. Life Tables also reveal delightful outcomes that confound the mind.

Here's a typical Life Table for a brand with existing customers that yield an approximate 31% rebuy rate on an annual basis.

You've all looked at a situation like this.

Let's run a small experiment. In the first month, let's just pretend that zero (0) of the 1,000 initial buyers repurchase.  How does the rest of the year progress, even if repurchase rates after the first month are identical to normal?

Did you see what happened?

Read across the Month = 2 row in each table ... you have 50 purchasers in the top table .. in the bottom table (where nobody purchases in Month = 1) you have 54 purchasers. You make up four buyers. In Month = 3, you make up 3 purchasers. And after a full year, the 63 initial customers who were lost are reduced to 308-261 = 47 buyers.

The same thing happens in reverse ... if business is great, you get the benefit of great business early on and then, even with constant rebuy rates, the actual number of customers buying is smaller.

I receive feedback from readers - readers typically want to know how to "improve" rebuy rates. There are two ways to do this, of course.

• Sell better merchandise, and/or develop winning new items that become winning existing items.
• Spend more money marketing to customers.

Both methods are valid ... but interestingly, both methods are limited by the math in the life table above.

Your rebuy rates are highly dependent upon what you sell and how often customers want/need what you sell. Under normal conditions, math conspires to hold your rebuy rates within a common band.