Challenge

Product portfolio management is a very sensible topic in product-based organizations where there’s a significant pace of innovation, though most organizations can benefit from some level of portfolio optimization or re-arrangement through exhaustive analytics.

Volume of sales, a traditional metric, doesn’t tell the full story and advanced analytics can help to better gauge your portfolio of products as well as its performance over time.
For this organization that decided to take a deep look onto the overall portfolio, the structure has a relative simple hierarchy, as per the following data:

Data summary
Name data_sku
Number of rows 143
Number of columns 8
_______________________
Column type frequency:
character 6
numeric 2
________________________
Group variables None

Variable type: character

skim_variable n_missing complete_rate min max empty n_unique whitespace
sku 0 1 8 9 0 143 0
product 0 1 4 12 0 43 0
description 0 1 11 40 0 141 0
product_range 0 1 2 2 0 4 0
id 0 1 1 3 0 143 0
product_ID 0 1 6 6 0 45 0

Variable type: numeric

skim_variable n_missing complete_rate mean sd p0 p25 p50 p75 p100 hist
ssp 0 1 433 112.8 199 349 429 499 999 ▆▇▃▁▁
cash_margin 0 1 175 55.6 44 135 171 213 323 ▂▆▇▅▁


So there are 4 different product categories (or ranges), 43 products, and more than 143 SKUs for one single market. However, the complexity of the portfolio can be observed through the following chart, where seasonal product sales can be notices, products in maturity stage and other products being introduced in the market too.

Random plots for different products showing rather different patterns

Random plots for different products showing rather different patterns

Now, looking into the cumulative sales for different SKUs, we can note different patterns by SKU:

Different patterns for cummulative sales, depending on product category (left) or SKU with low-volume sales at granular level (right)

Different patterns for cummulative sales, depending on product category (left) or SKU with low-volume sales at granular level (right)

On these figures, in SKUs 65,97,98, sales have been growing for the last months indicated a growth phase,in SKUs 107 (top left) that might indicate a seasonal product,or SKUs 103,120 that might have already reached its maturity phase or decline phase, in SKUs 130 and 131 that has just been launched into the market, entering into introduction phase.
These different phases or stages imply additional actions downwards,like renew the product or retire it in maturing phases , or exploring and exploiting in initial product phases.
It’s important to realize that within each market, the product has a different segment and audience, therefore the sales metrics can be different by category. Let’s explore the amount of sales within each category by making a histogram of monthly sales across the different product categories over the last two years.

Monthly sales distribution over time with latest data-points, that confirm the overall pattern.

Monthly sales distribution over time with latest data-points, that confirm the overall pattern.

A long tail of sales can be seen now and most of the time the monthly sales fall under less than 125 monthly-sales units. The number of weekly sales depends as well of the product life-cycle status: products new to the market, product in a mature state and therefore declining,…as we saw above.

Let’s wrap-up what we have learnt so far

  1. Product portfolio can be structured through this simple hierarchy
    - 140 SKUs
    - 43 products
    - 4 products categories
  2. Monthly sales pattern may reveal which product-lifecyle stage the product is in
    - launch
    - growth
    - maturity
    - decline
  3. Majority of monthly-sales were less than 250 unit per month, and the same pattern is confirmed in the last 3 months.

Solution

But with more than 140 SKUs, how can we categorize all SKUs without going 1-by-1 in market A, then the same for market B and so forth?

Based upon historical sales, can we make a first attempt to categorize the different SKUs and therefore cluster them in different buckets like the traditional lifecyle products? Let’s see what the outcome is by clustering the data in 4 clusters. In the following chart we see 4 clusters, and for each cluster we have captured pretty well the sales trend over time, so what else can we see?

Graphical representation of the 4 clusters containing 140 SKUs

Graphical representation of the 4 clusters containing 140 SKUs

Let’s focus on the yellow squares, indicating those SKUs with higher sales variation in the last months (cluster 4), or with sales in the first months(cluster #1),those having flat sales (cluster #3)… Ok,so we have the following distribution of SKUs into the different buckets:

## 
##   1   2   3   4 
##   4  10   8 145

Let’s dig deeper into the different clusters by plotting different SKUs:

Cluster - The dogs to be retired

Figure showing the products that need to be commissioned from a marketing perspective

Figure showing the products that need to be commissioned from a marketing perspective

We have here a small number of different SKUs (under 8 different products) that had a tiny growth in the first months, but now they are not contributing to the overall portfolio. We can highlight that EC07AM might be a seasonal product with sales in some specific months, so might need a deep review.

Cluster - The Cash-cows

The great products that every company aspire to launch

The great products that every company aspire to launch

First and foremost, remember that the bulk of monthly sales were about ~250, and here we see a 10x-20x increase in sales for some months over the average, so they are products in the best moment, the rock stars.
In this cluster have only 4 different SKUs (under 3 different products) that had an exponential growth in the last months, and they are having a great contribution to the overall portfolio. Having two different SKUs under CF11SV is something that deserves a look, perhaps some sales cannibalization that could be avoided with some offset in product launches. Marketing or sales led organisation? It’s clear from this, isn’t it?

Cluster 2 - The promising stars

Before having great performing products you need to have promising stars, like the ones in these figures

Before having great performing products you need to have promising stars, like the ones in these figures

We have here 8 different SKUs (under 5 different products) that had a significant growth a few months ago, that might need a boost to keep the sales growth. Looking carefully at PC32X3, you’ll realize that it’s in the same situation with sales in the last months, though might might need a deep review too (negative sales means a recall from one SKU). CF11SV might be a seasonal product.

Cluster 3 - The bulk

The majority of products, the reamainder, not great ones but helping in many ways: brand, awareness, adoption,...

The majority of products, the reamainder, not great ones but helping in many ways: brand, awareness, adoption,…

We have then represented 145 SKUs, under 44 products and most of them represent low-volume sales, that shouldn’t deserve very much time, though a few caveats must be raised:

  • There are a few ones which might be a niche market,with sales up to 600 units. Here is where domain knowledge comes to play again, and it’s the opportunity to engage the commercial teams.
  • There are as well several SKUs with peaks over different but at periodic times that might be seasonal products.
  • FS12SV,FS28CY and EC04TP very likely should be move to cluster 2. Probably a few more, but again. that’s the opportunity to humanize the data and bring the heroes (commercial teams) that need to move the needle on these numbers

Action

With this simple exercise we have inspired action through data: transformed a set of independet SKUs that were managed in an isolated way, to a coherent and consistent portfolio of different clusters with SKUs in very similar stage and where each bucket needs to undergo some specific marketing investments as well as contribute with the right returns.

Though the brand new portfolio needs still some tweaks to make it 100% consistent, the clustering algorithm has been helpful during the process, identifying where the different SKUs should be allocated to with a simple performance criteria. This exercise has been solely focused on product sales metrics though, so external metrics like market share and positioning would help to allocate the different products in the right place before getting the final decision-making.

However it’s pretty easy to scale to other markets, in fact, the algorithm can build in a few minutes a quick assessment of any other portfolio with the same structure and propose a first set-up do discuss upon.

Do you want to run a similar assessment for your portfolio? Drop me an email should you wanted to know more!