Case Study

A bubblegum manufacturer currently offers two flavors, classic bubblegum and cola, and decides to expand its portfolio and offer additional flavors. Which flavors should they add?

Lots of companies confront this same problem. For example:

- What range of products should a retailer sell?
- What media should they use for advertising?
- What product features should a manufacturer offer?

Raw Data

I’ll start by adding a data set with 712 respondents.

Add data > Displayr > Resources - Documents > Data > TURF > Bubble Gum Flavors.sav

The survey asked respondents which of 11 flavors they like.

In this raw data table, each row represents a respondent, and for each respondent, 1s indicate flavors they like while 0s indicate flavors they don’t like. So, we see the first respondent likes the classic, grape, sour, orange, and cola flavors.

Reach by Flavor

We see here that 66% of people like the classic bubblegum flavor. This is what’s called the “Reach” of the classic bubblegum flavor. That is, a portfolio that has only one flavor, classic bubblegum, reaches 66% of the market.

The cola flavor’s reach is 38%.

If you add these percentages together, you get 104%, which is impossible. You can’t just sum the percentages and ignore the duplication or overlap that exists due to people who like both flavors. We need to calculate the *Total Unduplicated Reach* of the two flavors.

We see the Total Unduplicated Reach of a two-flavor portfolio w/ the classic bubblegum and cola flavors is 75%, or 74.7% to be more precise.

The goal of this study is to find the best combinations of flavors to expand market **REACH**.

Duplication Matrix

The most basic way of analyzing data like this is to create a duplication matrix that has brands, or in this case flavors, in both the rows and columns.

We see column percentages by default, but I prefer total percentages for this type of analysis.

Reading the first row, we see that 66% like just classic bubble gum, 35% like classic bubblegum and super strong, 45% like classic bubblegum and grape, and so on.

Remember, the reach for classic bubblegum and cola was 75%. We can compute that using this duplication matrix.

Classic bubblegum by itself has a reach of 66%.

Cola by itself has a reach of 38%.

And the duplication or overlap between the two flavors is 29%. That is, 29% like both.

So, the Total Unduplicated Reach is 75%.

There’s a lot going on in this table, making it tough to find patterns. Fortunately, we can use Correspondence Analysis to quickly find patterns.

Correspondence Analysis of a Duplication Matrix

I’ll search for “Correspondence Analysis” to quickly find it in the menus.

We have three options, but which should we choose? Let’s try searching for “Duplication Matrix” instead.

Displayr’s built-in expertise recommends a “Correspondence Analysis of a Square Table” where square means the labels in the table’s rows and columns are the same. We’ll trust Displayr and select that.

The closer two flavors are, the greater the duplication between them.

We see the two current flavors, classic and cola, have a lot of duplication between them, so there’s probably no point in adding super strong or grape, as they also appeal to the same group of people.

To expand reach, it probably makes sense to add a non-traditional flavor or flavors since they’re farthest away and therefore have the least duplication.

~~All else equal, we should add watermelon since it’s the farthest from the current flavors and therefore will have the least duplication.~~

~~But, all else isn’t equal. It only makes sense to add watermelon if it’s also popular. So, we can add each flavor’s popularity by adding bubbles to the visualization that represent their reach.~~

~~Unfortunately, watermelon has a small circle, so it’s not popular. Note that the biggest circles are all for traditional flavors, so it might make sense to have multiple of those flavors.~~

While Correspondence Analysis is great for distilling patterns in two-dimensional tables, Factor Analysis can sometimes gives us a more nuanced view w/ more dimensions.

Factor Analysis

+ > Advanced Analysis > Dimension Reduction > Principle Components Analysis

What most of you know as “Factor Analysis” is technically called “Principal Components Analysis.”

This analysis reveals three flavor components or dimensions:

1. The first is sweeter fruits
2. The second is traditional flavors
3. And the third is citrus fruits

This suggests a good portfolio should have representation from all three components.

Displayr, by default, works out the number of components based on something called “The Kaiser Rule.”

It’s usually worth looking to see if this is clearly the best solution or if there’s a bit of noise. We check by changing the output to a “Scree Plot.”

The X-axis shows the number of components while the Y-axis shows how much information they explain. If we see a sharp corner, that tells us that the three-component solution was clearly the best, but we don’t. This curve is pretty smooth, so we should explore other options.

With two components, we basically get the same result as the Correspondence Analysis. We have a traditional component, a fruit component, and sour and chocolate are somewhere in-between. Let’s explore a four-component solution.

The four-component solution is very similar to the three-component solution. We have a sweet fruit component, a traditional component, a citrus component, and a chocolate-only component.

Total Unduplicated Reach and Frequency

While Correspondence Analysis and Factor Analysis are general tools for finding data patterns, TURF analysis is specifically designed for portfolio optimization.

+ > Advanced Analysis > TURF > TURF Analysis

TURF searches through the data and finds the combinations of alternatives, or in this case flavors, that maximize reach.

This output shows us the top 10 portfolios that consist of two flavors. The best two-flavor portfolio consists of classic bubblegum and strawberry, which reaches 82.4% of the market and has a total of 747 choices, where this frequency counts people twice if they like both flavors.

With three flavors, the best portfolio consists of a traditional flavor, a citrus flavor, and a sweet fruit flavor, which is basically what we saw before.

But look at the second best portfolio, which is almost as good. It has classic bubblegum and grape, which we know have high duplication. These portfolios don’t differ a lot, with the best portfolio being only 0.6% better than the second best. This is good news, as it means that we have a fair bit of flexibility in choosing our best portfolio and can trade off tactical and operational factors.

Now, if we’re going to have more than two flavors, we likely want to include the classic bubblegum and cola flavors. We can force the TURF to include those two flavors using constraints.

TURF > Constraints > Must include: Classic and Must include: Cola

Let's switch back to a two-flavor portfolio.

We see again that reach is 75%.

If we add a third flavor, we should add strawberry.

With four flavors, we should add sour.

Grape is added to the mix with five flavors.

With six flavors, chocolate enters the picture, but orange replaces sour. You can see that the third best six-flavor portfolio includes sour and has a reach that’s only 0.3% lower. This is well within noise, so we can force the results to be consistent by including sour as a constraint.

TURF > Constraints > Must include: Sour

Now, let’s put all these TURFs together and visualize the results.

Incremental Gains of New Flavors

An incrementality or waterfall plot is great for visualizing the gains in reach achieved by adding flavors.

We see adding strawberry increases reach by 14% and then adding sour increases reach 6% more.

Adding grape increases reach by only 2.5%, so the trick might be just adding strawberry and sour to classic bubblegum and cola and sticking with a four-flavor portfolio. This is where your judgement and other factors come into play and you need to decide if the incremental gains from adding a flavor are worth it.