It's Just A Matter of WHERE

Last week, we took a look at how to better utilize your customer traffic data to incorporate the footsteps and conversion rates into useable demand signals. This week we'll add on a clever allocation tool to determine WHERE that customer is going to walk into next or click on as they find your offering compelling enough to purchase.

First off, I'm going to list the data points you are likely going to need to make this task less onerous and more insightful.

1) Some sort of INSIGHT or LOYALTY data that allows you to identify the customer attached to the purchase. This can be Air Miles, Aeroplan, or your very own rewards or subscription program; anything that allows the customer to share their demographics in exchange for value.

2) City/District/Region demographics data that shows you the population breakdown by Postal or Zip Code if possible. This allows you to assign a geographic area to the stores you have in that area, essentially a TRADE AREA for each store.

3) SMALL INVESTMENT: Companies like Argus Advisory Group have done an incredible job at capturing Credit and Debit Card spends in Canada in a very easy-to-use format. This service generally costs < $5000 for an annual report and the usefulness of this data will pay back 10 fold. This is $ spend BY RETAILER by Postal/Zip Code for every participating retailer and major card. Combine this with your payment token information to associate payments with customers.

The Data does the Magic

Now that you have these three data sets, you can start to decipher just how your customers are going to disperse their spending with you. Using the first set of data, you can now start to examine just HOW many transactions you do that have some customer data attached to it. It just needs to be a relative small percentage to statistically extrapolate the data across total sales data. You are trying to determine what POSTAL CODE this person resides in and with that, should you have multiple locations in an area, how often they shop away from that code. This is what I call your ROAM RATE by customer:

Eg. Customer X spends $650 annually with you. 63% is in their home store, 24% in your big mall store, and 13% in the store by the office. You now can allocate that spend by store. Total Value = Spend in Store 1 + Store 2 + ONLINE, etc.

You can now take the demographic data you have collected in Step 2 above and start to build out your total market size that you are playing in, and what percentage of that you are already capturing, versus how much more you can potentially gain. If you have 2000 customers shopping in a total population area of 100,000, you have captured 2% of the total market in that trade area. You can use you Loyalty data to get an average spend by LOYAL vs OCCASIONAL and start to prize out how much a % gain can deliver.

This is where you're #3 spend will come into play. You have a total spend by retailer by Postal Code. If you assign these retailers a % of the total category spend in the postal code and how much you've captured, you start to get an idea of just HOW MUCH spend is available.

The Insight

With some simple calculations, you are now able to do some pretty clever manipulation to estimate the following:

1) How much will a loyalty customer spend by channel/location

2) How much an occasional customer will spend with you by location

3) How large is the category potential by location

What you have is a simple building block tool to be able to roll up from the customer/location a estimated sales and consequent inventory plan by category/location. This provides you yet another validation check point to the viability of your plan. You can then capture your forecast accuracy of the method to and use machine learning to test what assumptions should be weighted by what means.

This can eventually replace your triple exponential smoothing algorithms. For now, build and develop this technique to match your current method's results and train the model to replace it in a future upgrade.