Kevin Hillstrom: MineThatData

I ran a logistic regression equation to understand how likely multi-category buyers are to purchase again in the next year ... after controlling for purchase frequency and AOV ... analyzing twelve-month buyers. Here's the results of the Logistic Regression equation ... nerdy stuff you can ignore (though I do want to show my work).

The variable "multicat" has an Exp(B) of 1.361 ... this means that any customer that historically purchased from more than one category was 36% more likely to buy again in the future AFTER controlling for purchase frequency and AOV.

Separately, I was able to demonstrate that "multiple" is important ... whether the number is two categories or seven categories is much less important.

In a prior analysis, we demonstrated that customers are responsive for a few months following a first purchase, then are dormant, then come back at months 11/12/13 and again at 23/24/25 following a first purchase.

A marketing plan becomes obvious.

1. Separate email streams for first-time buyers within three months of a first purchase, attempting to generate a quick second order (preferably) within a second category.
2. Personalize the merchandise in email campaigns for first-time buyers to increase the probability of a second purchase (preferably) within a second category.
3. When a prospect visits via search, the goal is to convert the prospect to a first order and hopefully cross-shop the customer into a second item and/or second category.
4. When a customer reaches recency = 11/12/13/23/24/25, we ramp-up messaging and frequency.
5. Personalize the home page and/or landing pages to show customers what they "need" to see to maximize their future value.

        

Yesterday I talked about Entertainment customers shifting dollars over time to Apparel Tops.

By the time the customer orders for the third time, 34% is in Apparel Tops, just 9% in Entertainment.

Does this happen to all categories? In this case, yeah, it happens to all categories.

Here are first order categories and migration through a third order.

Apparel Bottoms: 47% / 21% (Bottoms vs Tops) in first order, 16% / 35% in third order.

Fashion: 34% / 21% (Fashion vs Tops) in first order, 9% / 33% in third order.

Home: 47% / 19% to 18% / 32%.

Jewelry: 40% / 17% to 6% / 35%.

Workplace: 27% / 21% to 3% / 36%.

Outside: 51% / 12% to 17% / 27%.

Having Fun: 37% / 20% to 15% / 32%.

Seasonal: 34% / 19% to 7% / 31%.

Decorations: 31% / 20% to 9% / 32%.

Two interesting things happening, of course. First, customers on a first order do buy from multiple categories, with Apparel Tops being right there at the top. Second, as the customer evolves, the customer spends more and more money with Apparel Tops, less and less money with the category of a first purchase.

As a marketer, tell me how you plan on using this information to make different decisions? If the answer is "I'm not changing anything", that means something.

Tomorrow, we'll explore an email exchange discussing "doing something different" given we have knowledge of customer behavior.

        

We've learned that "Beans: The Internet's Only Variety Store" is heavily skewed to the November/December timeframe.

I've learned across nearly twenty years of consulting project that it's not healthy for a business to skew so heavily to Christmas. You want to offer products that customers purchase all-year. If there's a reason that subscription-centric brands are coveted by investors, there's the opposite reason that Nov/Dec businesses aren't coveted.

When I see a skewed business, I run a regression model to test the dollar contribution of Oct/Nov/Dec orders vs. orders generated during the rest of the year. For twelve-month buyers, spanning four years of purchase history and one year of "prediction", here's the simple regression equation.

• 1.257 + 0.079*($ Spent in Oct/Nov/Dec) + 0.095*($ Spent Jan-Sep).