## September 08, 2019

### Training The Customer To Spend Less

In retail, my client base spent 20 years telling the customer to never visit a store. And guess what? Customers listened!! Malls have been emptied, stores have been closed, and the experts are left staring at the rubble wondering why the failed omnichannel thesis didn't grow sales?!

The same thing happens when you teach a customer to pay less for an item than the item used to sell for. Be it discounts/promotions or actual price cuts, it turns out that customers notice when you do this stuff, and they don't forget ... they like paying less.

Of course, if the customer pays less, you make less, so that's not a good equation if your job is to increase profitability, right?

For every item in your assortment, I calculate whether the customer paid an average or above-average price for the specific item or a below-average price for the item.

Then, I create four variables.
• September 9, 2017 - September 8, 2018 total demand spent on items at or above the average item selling price.
• September 9, 2017 - September 8, 2018 total demand spent on items below the average item selling price.
• September 9, 2018 - September 8, 2019 total demand spent on items at or above the average item selling price.
• September 9, 2018 - September 8, 2019 total demand spent on items below the average item selling price.
With the four variables, I run two regression models.
• Next Year's Above-Average Volume based on Last Year's Above Average and Last Year's Below Average spend.
• Next Year's Below-Average Volume based on Last Year's Above Average and Last Year's Below Average spend.
The two regression equations allow us to determine how much a customer will spend next year based on actions of the past year. Fair enough? Let's look at the results. Here is the model for next year's items sold below their average historical selling price.

For every \$100 spent on items below average last year, the customer can be expected to spend \$26.10 on items below average next year.

For every \$100 spent on items above average last year, the customer can be expected to spend \$13.40 on items below average next year.

So clearly, there is some affinity for customers who bought below-average priced items last year to continue to do it again next year.

The big question is this ... what will customers who spent money on below-average items prices last year do when presented with above-average item prices next year? Here's the equation (yes, this is actual customer data being presented to you).

For every \$100 spent on items below average last year, the customer can be expected to spend \$24.10 on items above average next year.

For every \$100 spent on items above average last year, the customer can be expected to spend \$31.40 on items above average next year.

So this is where things get interesting.

Let's pretend that a customer spent \$100 on above-average priced items last year. The image below depicts what we can expect from this customer in the next year:

We expect the customer to spend \$12.18 on below-average-priced items and \$29.88 on above-average-priced items, for a total of \$42.06.

Say instead you offered the customer 20% off ... and let's assume that the customer spent MORE because of the generous discount, spending \$100 regardless. What do the equations tell us about this customer?

The customer will actually spend more ... \$47.46 instead of \$42.06. But look at the reduction in spend on above-average-priced items ... we go from \$29.88 down to \$22.58 ... a 24% reduction.

Let's pretend that gross margins are 50% for above-average-priced items and 40% for below-average-priced items. Last year's above-average-priced customer would generate \$19.81 while last year's below-average-priced customer would generate \$21.24. You'd actually enjoy an increase in gross margin dollars. How about that??

But what if you offered 40% off ... so that gross margins are 50% for above-average-priced items and are just 20% for below-average-priced items? Now you are comparing \$19.81 to \$16.26 ... you've harmed the business.

The key, then, is to find the right balance ... let the equations above drive what the optimal level of discounting and price adjustments should be.

Make sense?

P.S.:  These are the type of posts that garner criticism from the stat/math/data-science folks. They'll criticize that the methodology is too simple, or doesn't take into account the "right" factors. Well, the ones criticizing are right. But how am I supposed to teach the concepts from a machine learning algorithm factoring in 29 different variables? Here, it's easy to see the outcome, right? Sometimes you go simple so that you can teach the concepts. I hope that's an acceptable outcome. If not, ask your favorite vendor to do the work for you the way you want it done, and pay them. You're getting this information for free.