February 10, 2016

Winners / Contenders / Others

If you are a marketer, you are probably exasperated with your merchandising team. They beat you over the head for contacting the wrong customers at the wrong time in the wrong way. They look at your email marketing performance and think that one customer purchasing for every seven hundred deliveries is an apocalyptic catastrophe. If you are a cataloger, then it is common for the merchandising team to blame the performance of 160 pages of merchandise on the marketer. I've been in the meetings. I've witnessed the carnage.

As a marketer, you can hold your merchandising team accountable for performance. It's not rocket science. And no, you won't hold them accountable for performance with a square inch analysis. That style of analysis helps you figure out how to merchandise/paginate a catalog, and that is of the utmost importance if your organic percentage is under 30% and you care deeply about catalog response. For everybody else (which is almost all of you now), you care about a merchandise assortment that works the whole year 'round, and works in email and affiliates and paid search and SEM and comparison shopping engines and via direct load traffic. You care about an assortment that attracts new prospects.

I recommend that you start simple.

I will also say, right up front, that there are plenty of consultants out there who will do a great job for you and will disavow my style of analytics. Please, work with those folks. They will help you improve response/conversion.

I use a concept called "Winners / Contenders / Others".

Step 1 = For the desired period of time, count how many customers purchased during that time frame.

Step 2 = Now, sum total demand per item, and total units per item for that same time frame. Do this for each item.

Step 3 = Divide demand by customers. Divide items by customers.

Step 4 = If an item is in the top 5% of all items for demand/customer or it is in the top 5% of all items for items/customer, the item/style is deemed a "Winner".

Step 5 = If an item is not a "Winner", but is in the top 45% of item performance in terms of demand/customer, then the item is deemed to be a "Contender".

Step 6 = All items not classified as a "Winner" or a "Contender" are segmented as "Other".

Using this simple framework, you'll see that +/- 40% of annual demand comes from a handful of "Winners". Another +/- 50% of demand comes from "Contenders". And then you have 55% of the assortment that contribute +/- 10% of annual demand, and are generally not worth measuring.

Perform a simple longitudinal study of Winners / Contenders / Others by Category. This simple analysis is going to tell you if you have a merchandising problem. Count Winners over time. In almost all merchandising challenges, you'll see a reduction in Winners, a reduction that is often sourced from poor new item performance from 1-3 years earlier.


  1. Hi Kevin,
    Quick question on this scoring methodology - is there a way to adjust for variables that may have influenced the demand, such as position in category (items on page 1 of a category's list page are seen more, clicked more and bought more than those buried, and not necessarily buried in results for a strategic merchandising reason)? In other words, the rich get richer. Ditto for products featured in email campaigns, etc. ?

  2. Of course you can do that ... have at it! In fact, you can use your analysis to prove that marketing plays a role in the merchandise that eventually wins.

  3. I mean can you/how to remove that bias to truly uncover some untapped potential. Is there a way to adjust scoring to mitigate the influence of promotions and arbitrary features? (Let's face it, a lot of marketers/merchandisers are guessing) Perhaps there's a way to filter out visitors from a test group that have existing accounts/email subscriptions, or a way to continually randomize category results - just wondering if there is a tool or approach to measure it this way

  4. You can easily use regression models with dummy variables to get at what you are talking about.

    For instance, let's pretend you only sold 100 items. You sum up annual sales for each item. Then, you create dummy variables. Pretend that five of the items were offered at 20% off, while the rest weren't (dumb example, I know, but work with it for a moment). You create a 1/0 variable ... 1 = offered at 20% off, 0 = not. In a regression scenario, the items sold at 20% off would perform "x" percent better. The regression dummy variable would capture this impact.

    Repeat this process for each factor you wish to quantify.

    Run your regression model. Only keep significant variables (those that aren't significant are important for business reasons - it tells you that those marketing tactics are utterly meaningless).

    Now, go back in, and for each item, subtract the value of a coefficient if the dummy variable had a value of "1".

    If the item sold 10,000 items, and a promotion caused a 10% lift, you subtract 1,000 items from the total, leaving 9,000 adjusted items sold without promotions. Repeat this process for each dummy variable, and you've controlled for the impact of marketing promotions and home page placements and email strategy and whatever else you do from a marketing standpoint. Now you can compare each item without bias (theoretically).

    That's one approach. There are many. All are imperfect in some fashion.

    Good job - good question - I very much appreciate smart questions!!!!


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

Speaking of That Community You've Grown ... Known As Your Email List

I'm analyzing a business, and here's what I see. 27% of the sales come from items selling below the historical average price point o...