December 10, 2012

E-Commerce Simulations: Here's My Framework

Use of simulations are commonplace.  We wouldn't be having vibrant discussions about global warming unless somebody ran a simulation illustrating a forthcoming train wreck.  

And nearly every day, your television station provides you with a four or five or seven day weather forecast ... and it's reasonably accurate.  Weather forecasters are running a myriad of simulations, then they forecast the future.

Remember the movie Apollo 13?  The simulator was pretty important in getting folks back to Earth.  And that was more than 40 years ago.

Formula 1 race car drivers hopped into the simulator to test the new race track in Austin, Texas.

Why, in e-commerce, do we not use simulations?

We don't have to over-complicate the creation of simulations.

Let's use a very simple example.  I take two snapshots of the twelve-month file, one from October 10 - November 9, then another from November 10 - December 9.  I record key attributes about each customer ... recency ... frequency ... monetary value ... channel preference ... merchandise preference ... Christmas shopping preference ... price point preference ... free shipping preference ... discounts/promo preference.

Allow me to over-simplify for a moment, to demonstrate how I run my simulations.  Let's pretend that in the 10/10 - 11/9 timeframe, I segment customers as good, average, and poor.  Then let's pretend that in the 11/10 - 12/9 timeframe, I segment customers using the exact same criteria ... good, average, and poor.

Toss in new customers, assigning them to good/average/poor, and we have a 4x4 matrix.

With this simple matrix, I can calculate how a customer will migrate in the next year.  Look at average customers.  If I start with 1,000 average customers ...

  • 1000 * 0.40 = 400 will become good customers next month.
  • 1000 * 0.40 = 400 will still be average customers next month.
  • 1000 * 0.15 = 150 will become poor customers next month.
  • 1000 * 0.05 = 50 will not be active, and will leave the simulation next month.
This process is repeated for each row, giving us the count of customers starting next month in the simulation.

This process is repeated, month after month after month, yielding simulated results for your business over time.

This is a 4x4 example.  My methodology utilizes a 1,000,000,000 x 1,000,000,000 matrix ... the matrix takes no extra work to program in a computer than a 4x4 matrix used in our example (though many of my models work on a 500 x 500 matrix, FYI).

I append 12-month spend values across channels and merchandise categories, yielding simulated sales totals for the business.

See, this doesn't have to be terribly difficult.  And yet, simulated results yield discoveries that are not easily obtained via normal queries.

Yes, we need to execute e-commerce simulations.  We're decades behind other industries.  The time is now.  Get busy!

1 comment:

  1. The simulation is remarkable, I don't see any flaw on the data that you showed. Who would thought that the computation for this simulation would be not so difficult. If I venture in e-commerce I would definitely use this simulation to predict the outcome and to improve them if deficiencies would appear.


Note: Only a member of this blog may post a comment.

Lone Wolf?

I just went to the Safeway website to look for Tempura batter. Odds are that most people don't go buy Tempura batter ... and that's ...