Where are my sales? — Part 1

Corner office. 20th floor of the company headquarters. You’ve a thirty minute connect with the VP of revenue management. The sales report for the previous quarter are in; and they are poor to say the least. You were hired six months ago as the sales director of the FMCG leader to improve sales. However, you seem to have failed.

And you want to know why.

Sales driver analysis draws from the paradigm of key driver analysis, wherein you find the importance of some predictor variables(e.g. age or socio-economic background)towards predicting an outcome variable(e.g buying preferences). We build a driver model, which is basically an equation describing the relationship between the outcome(dependent variable) and predictor variables(drivers).

In sales driver analysis, we try to understand the factors which impact sales.

• A sales driver model tells you the relative importance of different factors affecting your sales. For e.g Price, execution(how well you sell a particular product) or some macro economic factor.
• Multiple linear regression is the most common and robust technique used for sales driver analysis. It gives you an R-Squared value(more on this here) which gives you an idea of the model fit. As an e.g., an R squared of 80% would mean that 80% of the variation in your sales can be explained by the model, while the rest 20% variance is unexplained.
• A sales driver model is almost always built at a Store-Product level. What this means is that each store-product combination will have different factors explaining sales. As an example, the sales of sodas(cold drinks) will be impacted by weather while the sales of rice may not be affected much by weather. On the same lines, a store in an urban market(Bangalore) will respond differently to price fluctuations than a rural market(Patna). Hence, it becomes necessary to have separate models for each Store-Product combination.

This article is the first part of the series “Where are my sales?”. The 4-part series is structured as follows:

1. Introduction to sales driver analysis
2. Variable design and data treatment
3. Modeling in R
4. Interpreting the results and drawing actionable insights

Sales measures

Let’s quickly go over a few FMCG/CPG specific metrics that we will need to be familiar with.

• Unit Sales: The number of packages (e.g. 5 packets)
• Volume Sales: The weight of the product x Unit sales(e.g. 500g x 5 units = 2500g)
• $ Sales: No. of Units x Price per unit( e.g. 5 packets x $10 per packet= $50)

Slightly more tricky to understand is the concept of equivalized(EQ) volume and prices. We will go over them in part three of this series.

Key Drivers of Sales

Price

Price of a product, in general, explains ~50-70% of the variation in sales. When running a store-product level model, Price/unit is considered. However, when we want to run a store level model(i.e. across all products), we need to find the EQ prices. However, we would be discussing this in Part 4 of this series.

Distribution

People can’t buy a product if it’s not available in the stores. On the same lines, a product that is placed in a supermarket(bigger store by revenue in general) would have more chance of selling than if it was placed in a local grocery store. This is where ACV and %ACV distribution comes in.

ACV(All commodity volume) is the total $ sales of ALL stores in a region where the product sells. Consequently, %ACV distribution is the $ sales of the store where the product sold divided by the $ value of all the stores in the region.

Shown below is an ACV calculation example. As we can see, Reliance has the highest ACV(and consequently highest %ACV). Therefore, if a product sold more in Reliance than Loyal City, it could be due to the larger size(in terms of revenue) of the store. Hence we have to account for the same in the form of %ACV distribution.

Example ACV Calculation

Execution

Many a time you been stuck in a retail store, hunting for a product that you just can’t seem to find. That’s what execution is. It’s about having the right product at the right time at the right location in the store. Now execution as a metric can be in different forms. Generally however, it is in terms of where a product was placed(start of store,end of store or on a special shelf and so on).

Here is an example of what execution data may look like. We have the weekly sales of a product and we can clearly see that whenever the product had an advertisement banner, it sold well and vice versa.

Example of execution data

Additionally, ACV Display % and ACV Feature % can also be used as execution metrics. The higher the ACV% for display/feature of a given product, the more number of stores had displays/features for that product.

Note: In the US, displays are when a product is placed in a special temporary display without any discount/promotion. Feature occurs when a product is ‘featured’ in a retailer flyer(advertisement pamphlet), typically coupled with some price promotion. Different markets have different definitions of displays/features. Read more on Promo ACVs here and make sure you understand the nuances of the market you’re working with.

Seasonality(Structural Demand)

Seasonality captures the regular and predictable changes that recur every calendar year in sales. Seasonality is calculated by taking an index of the sales volume over at least 2 years.

An example is shown below where the seasonality index is calculated on a Monthly level.

Calculation of seasonality index

However, there are a couple of points to note here:

• The volume on which the index is calculated should be price controlled i.e. it should be the volume keeping the product at it’s base/EDV(everyday) price. If price is not controlled, the variation in volume(and hence index) will be because of the price(reductions or promotions) and not seasonality. We will go over this in detail in part 2(Data Preparation) of this article.
• Seasonality index is typically calculated at a category level than product level. This is usually done because the demand varies across seasons by category and not individual products. For e.g. Soft drinks and rice will have different indexes. However the index will not vary much between two soft drinks or two types of rice.

Pantry Loading Effect

Pantry loading is an effect that usually occurs when an item is on promotion. It essentially means that the consumer bought more of a product than they usually buy or need to buy because the offer was so good. Consequently they won’t buy that item again soon because they have stocked up. For example, if a consumer usually buys an 8 pack of toilet paper every week, but because of a great sale offer buys a 24 pack, which will last them 3 weeks, that consumer will not buy toilet paper again for 3 weeks.

As an example, we can see that whenever the Discount Depth transitions from Deep discounts(>40%) to Shallow/No Discounts(in Week 2,3,4 or 7,8,9,10), the sales volumes are low. This can be explained by the pantry loading effect.

Example of pantry loading effect

In the next part, we will be covering a few more factors and the isolation of drivers in a typical sales driver analysis pipeline.

Please do let me know your thoughts, questions or suggestions in the comment section below.