Have an idea?

Visit Sawtooth Software Feedback to share your ideas on how we can improve our products.

Umbrella brands - CBC Design considerations

Our client has a market where there are 3 major brands A, B, C. One of those three is really an umbrella brand with 3 product lines (3 small brands A1, A2, A3). The whole CBC experiment will consider brand+taste+price+pack-size. I can think of at least 2 strategies: a) Make brand 5 level A1, A2, A3, B, C b) Make a brand 3 level ABC and the whole design attribute specific (with sub-brand as additional attribute for brand A). Unfortunately I can see shortcomings of both:

a) Overrepresents Umbrella in experiment and makes it more psychologically avliable  - I expect it will boost whole brand A simulation result and cause problems with independent valuation of each brand.
b) Underrepresents each sub-brand of A, and gives less precise estimates. Unfortunately brand A happens to be our client brand.

From the shopper perspective I suppose that b situation is more likely to happen on the average shop's shelf. Thou I'm not completely sure.

Is my reasoning right? Is there any "c)" which overcomes problems of solutions a) and b).
Any concepts are welcome...
asked Sep 17, 2013 by lkomenda Bronze (2,810 points)

1 Answer

+1 vote
I think your reasoning is right.  A question:  how much shelf space does each brand have at retail?  If A has 3x the shelf space as either B or C, then your option (a) may well accurately reflect the market.  If A, B and C have equal shelf space, then your option (a) may well bias results in favor of the Brand A portfolio.  In this case your option (b) might work.  In addition, an option (c) might work, too:  allow each brand in your choice sets in proportion to their retail shelf space.  So if A, B and C have share in a ratio 3:2:1 maybe you have six products:  A1, A2, A3, B1, B2, C.
answered Sep 17, 2013 by Keith Chrzan Platinum Sawtooth Software, Inc. (106,150 points)
Don't forget that you can do some pretty easy tricks to rebalance the design.  If you do a 5-level brand attribute with levels:

A1
A2
A3
B
C

...and you worry that brand A is appearing three times as much as brand B or C, then you just modify the design initially to have 9 levels:

A1
A2
A3
B
B
B
C
C
C

...and generate the design.  Then, export the design, open it in Excel, and collapse the 9 codes back to 5 codes (e.g. collapse the three level Bs into one level and the three level Cs into one level).  Then, modify the SSI Web study to have 5 levels again for brand...and import the modified design.  Now, each brand is represented an equal number of times, but there are three variants of brand A
Thank you Keith,
Thank you Bryan,

This "real shelf corrected" design sound good.
And as I understand, using Bryan's trick I could easily rebalance to any shelf proportion I like...
Do you think that recoding (A1,A2,A3)->A will give reasonable brand utilities - assuming a reasonable proportion of A, B and C in my design?
...