Have an idea?

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

Approach to find a niche market

Hi, I am going to use CBC/HB to find a niche market.

1. How many respondent approximately needed overall and  in the niche segment ?
2. What segmentation technique should I use to find a niche market? (Orme recommends 300 resp. for each normal segment)
3. Dose increasing the number of segments in K-mean or Hierarchical clustering helps to find a niche market?
4- Is there any segmentation technique  to focus more on a specific type of variable when searching for clusters?

Many Thanks
asked Aug 9, 2017 by Robin59 Bronze (545 points)

1 Answer

+1 vote
 
Best answer
Robin,

1.  By their nature, niche segments are likely to be small, right?  That means you may well need large total sample sizes to find them at all and once you've found them you want to have enough respondents to profile them and to understand them well (else why find them in the first place).  Looking for niche segments isn't for the faint-hearted (or for those with slim budgets).
2.  I think you're asking about what segmentation method.  I'd avoid k-means, which tends to produce roughly equal-sized hyperspheres.  Hierarchical might be a better bet.  If you're using conjoint results you might also look into latent class MNL as being a way to find segments of respondents.
3.  I suppose it depends on how many segments there "really" are.  If you have 3 big segments and a niche segment  it's not clear to me how much adding a 5th segment will help you, assuming that structure really exists and you're using a method capable of finding it.  
4.  I'm not sure what you mean by "type of variable."  Some methods are more sensitive to some types of variable scaling than to others, but I don't know that that's what you mean by "type of variable."
answered Aug 9, 2017 by Keith Chrzan Platinum Sawtooth Software, Inc. (116,275 points)
selected Aug 10, 2017 by Robin59
Thanks Keith,

Regarding 4th question, assume I have included 3 competition brands(could be also different color, design etc.) plus a new anonymous brand in CBC HB. Majority of respondents prefer those three brands. There is a small sub set of respondent that prefer the new brand in terms of calculated main effect partworth. When I run market segmentation based on all partworth of all attribute levels (putting number of segments on automatic) I get three segments. But none of these segments have the new brand as a salient characteristic. Therefore, I increased number of segments to 5 for example but again the same issue....
Is there any method that find clusters based on focusing on a specific variable or parameter ( in our example new brand partworth)?
Some segmentation methods allow you to do variable weighting.  At a minimum you could just copy the number of times the column containing that brand part worth appears in your data set - this will implicitly weight the influence tat variable has on your solution.
...