When I used the term "scientific" I don't necessarily mean it has to follow the highest level of scientific scrutiny. I have some colleagues that will disagree with me because they are scientific purists. Something simple like a survey needs 3 years of preparation and contextual literature review. Don't forget validation! Silly!
If you are trying to change public policy with a survey you should go through all of the steps...but if you are simply trying to gain some understanding of what your customers want then you might be ok asking straight forward questions. Conducted in the right way gracious customers will give you more information than you want or need.
A problem arises when we want to justify the answer through a line of research. In our head we believe we know what the study will tell us and in turn subconsciously design the questions that give us the answers you seek. I see people do this in their everyday conversations when they have a point to make and skip over all the logic. The conclusion is not justified in the least! Leaving out important information, framing things incorrectly, designing leading questions, etc...all lead to biased results.
What you measure will also determine what results you get so make sure you are get into the details. Read this article on digital economy and its influence, or lack thereof, on GDP calculations (Click Here). If your really really really motivated you can read the full study and understand the studies design and what it truly indicates (Click Here) You will see that the way we calculate things can have some serious economic consequences.
How that data is collected and processed determines the overall value of the outputs. This is one reason why studies have limitations. It can't be everything to everyone and the inherent design of the study creates limitations! You should be well aware of what they are before you collect data!
Designing strong studies that have internal validity are important in the overall process of understanding the market and creating some predictive modeling. Be sure you know what you want to measure, do your background research and design the study in a way that leads to answering that specific question. If your lucky you can create a model and apply that to other places.
Sometimes you must complete preliminary research to ask the right questions. You think it would be easy but it really isn't. Having a general sense of the problem but might need to explore the idea a little to define it specifically for a research question. This research question becomes the basis for everything else within the study. Your single focus is/are to answer this question (s).
You will find as you begin to collect the data that it can start getting jumbled up and you will lose track of what you are trying to accomplish. Data is the source of all experiments and you might want to find some way of taking in that data, analyzing it, manipulating the variables, etc... Consider that collection may require specialized equipment or round about ways of getting the information while the encoding and analyzing of that data is more statistically oriented.
Once you have this mountain of data and it has been scrubbed and categorized there will be a need to analyze that data. The type of analysis you do will depend entirely on what you want to accomplish and what type of data you have. Some scientists want you to conduct huge analysis for a survey. In the business world you used what is most prudent and provides the most useful data. Statistics, regression analysis, etc.... is a little too complex for this discussion.
Three points of key advice:
1. Define your problem as everything revolves around this.
2. Design your study to ensure it is measuring what you say it is measuring.
3. Ensure you collect and encode data properly so as to create a stronger analysis at the end.