March 23, 2013 § Leave a comment
We got this client feedback through our partner about Adaptive Questions where respondents can see responses from others.
“I spoke to the client today and they are excited about the prospect of this project. They have only one concern which I could not address… they are worried about the likelihood of negative comments surfacing and being shared within the survey group, and possibly shared outside the group. “
One of our best long-term customers from TurboTax put this in a most elegant way, “You have to expect negative feedback to make any progress. If you filter it out, you are doing a disservice to your company.”
I usually like to put both positive and negative people in the same Adaptive Question because we can see how they agree and disagree. Think about how valuable it is to see agreement and disagreement between Promoters and Detractors, happy and unhappy customers. Personally, I prefer to allow respondents to say anything they want because…
- Bad ideas, negative ideas, complaints don’t usually get seen by very many people because they don’t usually get a lot of agreement because we word questions to get constructive feedback, not complaints
- Only a small portion of respondents see the negative ideas
- Only a small portion of customers participate
- Your mother told you, “If you don’t have anything nice to say, don’t say anything at all.”
Given the sensitivity of some clients such as this one, there are a couple of fairly easy solutions.
- Moderate the comments – only allow constructive comments to be seen. CloudMR has an option to require moderation. Respondents only see the comments you approve. That way we can limit the pool of ideas to positive and constructive improvements. The client will need to give us some guidance which is a part of our process. Downside: If you are not aggressive about approving legitimate ideas during the early part of the fielding process, you could go through all of your sample with just the initial 10 seed ideas.
- Separate positive and negative respondents into two Adaptive Questions(TM) – It is pretty natural to get a rating such as satisfaction or likely to recommend before the Adaptive Question. Simply use the built-in logic to get improvement suggestions from unhappy respondents and positive sound bites from happy customers. This isolates the negatives to an extent.
- Combine 1 and 2. Adapative Questions are ideal for getting WOM. The analysis will include both general buckets of ideas and individual comments. If you want to WOM for your promotions or advertising, we will ask respondents to identify themselves for the quotes.
March 15, 2013 § Leave a comment
When we started thinking about text analysis, one of the issues we discussed was sentiment analysis. Lots of people have tried to figure this out and some companies claim a pretty high success rate using algorithms or other techniques.
Our feeling is that language is so complex and that the subtleties of things such as sarcasm make it really difficult to be very accurate. So we took a different approach to the problem. What if we leave the determination of sentiment up to real humans? If we do that, it simplifies the task to finding which comments to read.
CloudMR’s patent-pending algorithm sorts a list of comments in priority order. The comments most likely to resonate with other respondents are sorted at the top. If you want to know sentiment, just read the top comment for yourself. If you are feeling really energetic, read the first ten. Sentiment problem solved…