February 23, 2009

Web Visits And Lifetime Value And Data Mining

One of the great sins of e-commerce is the website visit that does not result in a purchase.

We've been told for fifteen years that this is a disaster, and can usually be fixed with a few easy steps.

Of course, the problem is never easily fixed --- always improved, but we never see website conversion rates increasing by up to 339% after implementing various vendor strategies, do we?

Have your data mining expert give this exercise a try. Instead of building models using recency and frequency and monetary value and whatever additional variables you want to include, have your modeling expert add dummy variables for the following.
  • Shopping Cart Abandoned 1 Month Ago.
  • Shopping Cart Abandoned 2 Months Ago.
  • Shopping Cart Abandoned 3 Months Ago.
  • Website Visit, No Shopping Cart, No Purchase, 1 Month Ago.
  • Website Visit, No Shopping Cart, No Purchase, 2 Months Ago.
  • Website Visit, No Shopping Cart, No Purchase, 3 Months Ago.
What many folks find is that unconverted website visits and abandoned shopping carts are not the end of the world! In fact, these activities tend to increase future customer value, as illustrated in the graph above (months four and twelve in the graph).

Future value is sort of like a heartbeat ... each additional purchase (months six and eighteen in the graph) results in a shot of adrenaline, causing the heart to beat.

Each unconverted website visit or abandoned shopping cart results in a mini-jolt, as evidenced in many statistical models. Given that we observe this behavior, we don't get in a panic if the customer doesn't purchase. Instead, we view the activity as part of a relationship.

We seldom ask somebody to marry us after a first date, and in e-commerce, we have the opportunity to view each visit as a stepping stone in a long-term relationship. Simulations show that these unconverted visits are not the end of the world ... rather, they are part of an overall relationship with the customer. If the customer comes back multiple times, that can be a better outcome than demanding conversion in one visit, never to see the customer again.

Or at least that's what the data suggests.

2 comments:

  1. Yes, Recency of Visit - any kind of Visit - is predictive of likelihood to visit again.

    This is why people testing "shopping cart abandonment discount" e-mails should be using null control groups - you have to determine what % would have come back and bought anyway without the e-mail if you want to maximize profit.

    On the other hand, if your work performance is not measured at all, or is measured just on Sales rather than Profit, don't bother with the above - it's just more work, right?

    Some day, some boss *will* care. So you might as well explore some of these ideas before that day.

    ReplyDelete
  2. Every movement has a resistance!

    ReplyDelete