So in this part, we'll start talking about descriptive data collection. So descriptive data collection, as you recall, basically talks about trying to understand for instance, who are our customers? What does the? And all those kinds of question where you have to have hard numbers. So how can you do this? This can possibly be done in two ways. One is active data collection, and one is unobtrusive data collection. In the Active Data Collection, you an start thinking about again, two ways broadly of thinking about data collection. One is surveys, which is the mainstay of market research for many, many companies. The other one are self-reports coming from your customers. We'll talk about both of the them. So let's start with surveys. Surveys, pretty much used by every Fortune 500 company. Regularly used for gathering customer attitudes. You can think about sentiments. You can think about purchase habits. Many different things actively gathered by surveys. And data of course, can be helped to segment customers. Start thinking about who our customers are. Start thinking about, who do they buy from? All of those questions that you need to understand to set your marketing strategy. Now, there are many, many companies out there that can help you do these surveys. I'll give you some examples. Qualtrics is a very famous company that helps you conduct free surveys sometimes. Another example is Survey Monkey. Now, both of these companies not only help you do sometimes free surveys, but they can also be full service companies. So for instance, if you look at the pricing plan of one of these companies. They allow you and they price you differently based upon whether you would like them to find your customers, you would like them to setup the survey, analyze the data and give you that data. So in that sense, when you start thinking about doing surveys, there are already companies out there that can help you reach out to customers, collect that data, and analyze the data. But of course, surveys are not the only way in which you can get information from customers. You can actually ask customers, directly, who self-report some of those surveys. So if you look at, for example, mobile surveys. That's the next frontier. And these are basically companies that are giving these suveys on mobile devices. Again, some common examples, Qualtrics is one company that does both. It does surveys on the desktop. It also does surveys on mobile devices. Another company, for example, is Mixpanel. Again, what's the idea here? The idea here is, you want to basically send surveys to customers in the moment of purchase sometimes. So what do mobile surveys allow you to do? They allow you to capture customers' reactions in-situ rather than being retrospective. For example, you can actually send a survey to a mobile device of a customer at the time that they're making a purchase decision rather than one month later. So clearly the kind of sentiment, the feelings that customers might have at the time of purchase, would be better captured, rather than making them think about that purchase one month later. The questionnaire can be tailored based on location and context. So again, looking at where the mobile device is. Now, the words it tells you where the customer is. If the customer is in a mall, you can ask questions about what they’re doing in that mall. If the customer is at a restaurant, you can ask questions tailored based on that. So, very tailored surveys can be done. But what’s the caveat? You don’t want to overdo it. Marketers should be very careful, rather than kind of, again, using this leverage of making a tailored survey. You don't want to keep sending surveys again and again to the same customer. You quite often see a huge amount of survey fatigue setting in. So it's important again, use this power of mobile surveys but up to a limit. So now that we've talked about different ways of conducting surveys, either mobile surveys and so on. Let's go into more depth. What are the kinds of questions you can ask using surveys? And what are some dos and don'ts? Now, before implementing the survey, there are two big issues that come forth. What are the different kinds of questions, and how do you validate a survey? In other words, is it worth collecting what you're collecting? Let's go over the first issue. What are the different kinds of questions? Now, this is in some sense a list of different kinds of questions. Of course, there are many kinds of questions out there, but these are the important ones. So what I'm going to do in the next few slides is to go over each type of questions, look at what are the positives and negatives, in some sense the pros and cons. And then, kind of talk about best practices. So let's start with the first one, itemized category. Here's one example of this. How satisfied are you with your health insurance plan? It could be different buckets. In this case, you have five buckets from very satisfied to very dissatisfied. Notice, the extent of category descriptions are quite clear, and there is a balance of favorable and unfavorable categories. What does that mean? There's a middle point. Neither satisfied or dissatisfied. On the other hand, you have two categories which are above that. Quite satisfied and very satisfied. And two categories below that. Quite dissatisfied and very dissatisfied. So on the surface of it, I think this is very good way of asking a question. But what are some cons with this? Well, one big con is, compare to what? Of course, if the person who's answering this question does not have health insurance, clearly this question is not relevant. It could be possible that the question is being answered by a person who is thinking about insurance that he had before. If that's the reference point, the answer that he might be getting might be quite different across people. In other words, the problem with this is you don't know what people are comparing it against. So that's one issue with the itemized category. Let's look at another one which tries to address this issue. Here, which we would call as a comparative question, you directly ask compared to private clinics in the area, the doctors in private practice provide a quality of medical care which is very inferior to very superior. So what have we done here? You've tried to address a problem with the previous set of questions, which is, you're explicitly telling people what to compare against. But what's the problem here? What are the big loss of information? The big loss of information is, both alternatives might not be that great. So you're comparing two alternatives which might be both below the bar. But one is better than the other. So what do we see here? Depending upon the type of question you can ask, there might be some loss of information that always happens. So what I want to do again in the next few slides is to show you different kinds of ways you can ask questions which tries to get at the heart of the problems. Here's another one. It's called ranking questions. An example of this would be the following. Please rank the following characteristic of, let's say, cellphone service in terms of their importance. 1 to 8, there are 8 categories given. 1 here is the most important, 8 is the least important, and typically when you're asking these questions, no ties allowed. What that means is, only one of these things could be the most important and only one of them is the least important and so on. So what do we see here? First of all, the type of categories are quite clear. But what it also involves is a lot of comparisons. In other words, people would have to do a lot of comparisons when they're going across all of this. So if you look at it, for the first one, for the first rank, you're comparing across eight different categories, and giving one of those categories as the top rank. Let's say reception clarity for you is the most important one, so you give it number 1. Then, when you're going through giving rank number 2, you again have to do seven comparisons. It's a lot of different comparisons. What that means is typically such type of data might not be very beneficial to collect. If you have a lot of categories that you want people to compare, what might end up happening is that people give the rank 1, 2 and 3, perhaps thinking a lot, and after that, it might be too many comparisons for people to make. So the typical rule of thumb, or the best practice here, is that you don't give too many categories, maybe 8 might actually be already quite a lot. So maybe 6 to 8 might be a good number of categories to give. If you give more than that, you might get good quality data only for the top one or two, and after that it might not be a lot of distinguishing data. Another example of this is something called a paired comparison. In fact, if you think about it, which we'll cover later on in these sessions, something called conjoined. If you think about it, you will see that this type of comparison data is coming actually from a conjoined type of survey. Which of the following two products do you prefer? On the left-hand side you have Honda Accord, price of $18,000, automatic transmission and a luxury package. On the right-hand side you have Toyota Tercel, $16,000, manual transmission, standard package. What are we trying to do here? What we're trying to do here is forcing people to compare among the two objects. And this way, by looking at what they choose and what they don't choose, you try and understand what is it that people care about when they're choosing among these two products. This looks like a great way of understanding what people like, why? Because this actually mimics what people probably do in the real world. Imagine you want a laptop. What do you typically do? You go down, let's say, to Best Buy or any other store, or maybe on Amazon, whatever is your preferred provider. You go and start comparing different laptops. You choose the different things that you want to compare things on. So for example, for a laptop it might be the screen size, it might be how heavy it is, what's the CPU, and so on. So this really mimics what people really do in perhaps real life. But what are some issues with this? Issues are the following. Again, if you have, let's say, two comparisons here, two products, Honda Accord, Toyota Tercel, people might prefer Honda Accord to Tercel, but might actually hate both. Let's say you're given another option. It might be the case that amongst these two, Honda Accord is preferable but it's still below the bar in terms of what they like. Another problem of course is, large number of brands cannot be compared. Imagine yourself competing among six or seven different brands with lots of different kinds of features. Very, very difficult for you to make that decision, why? Because, again, there'll be lot of comparisons. What's the best practice here? Typically, you have about two to three brands. So that way you can get good data in terms of how people are comparing across different brands. And also you don't want too many features or too many attributes of these brands. Typically about six is a good number to have. In this example, we have four which is the brand name, the price, what kind of transmission is it and what kind of package is it. So about four to six features per brand and about two to three brands in terms of comparison. Anything more than that, I think it'll become very difficult for respondents to clearly understand what the differences are and give you good reliable data. The next one is the one which is the most common form. What's the story here? The idea here is you have many statements, typically on the horizontal. Which is here you see for the first one might be I buy many things with a credit card, and so on. And in each row you answer whether you agree or disagree with these kinds of statements. So this gives you an ability to collect a lot of data in terms of what people like and what people don't like. This is called the Likert Scale, it's the most common form of questioning and is used very frequently when you want people to think about lots of different statements, in this case about credit cards, and related ideas. Here's another example, something called the Continuous Scale. The idea here is if you have, for example, some things that you want to show people and you want what is called in-situ preferences. What that means is you want preferences as they are thinking about or looking at a particular, let's say, video or movie clip and so on. What do people do? Typically it's a bar, in some sense it can done on the internet very easily, very popular with computer mediated surveys. You can have a bar, a mouse click and so on and people, as they are watching a video or as they are watching an advertisement and so on, they can move this bar between do they like this or do they not like this. So this is very popular, especially in computer mediated surveys. When you want information, or how people are looking at your product, and how that preference changes as they're going through the different products. So if they're looking at a particular video. Let's say you're an ad provider and you want to see how people's preference for that ad changes as they're viewing the ad. So while they're viewing this, you can keep changing that counter. Many of us might see this more recently in election polls. So what happens during elections is when they're having debates and so on, they typically have an audience which has this meter which can go back and forth. So as they're going through an argument, you can kind of see how people are preferring towards one or the other candidate. So this is what is called as a continuous scheme. Now, what I wanted to do here was basically give you a broad overview of the different kinds of questions. Notice these are not an exhaustive set. There are many other kinds of questions that are out there. But what I wanted to take away from all of this is that each type of question that you ask, whether it's a rating scale, is it a comparative scale, Likert Scale and so on, each one of those questions has some pros and cons. So thinking carefully about what kind of question to ask depends upon what the end goal is. That brings me to issue number two. What is the end goal here? The end goal can be of two forms, one is called validity and one is called reliability. In other words, is what you're collecting going to be worth anything at all? So lets take that issue. Validity, basically the idea is of predictive validity. So for instance, let's say you're asking a net promoter score which is something that I'll talk about in the next slide itself. Net promoter score typically is one where you measure customer satisfaction, okay? And you're trying to see whether people are going to refer your product to other people. Now, what you'd like to hope for is that net promoter score typically predicts for example, customer profits, firm profits or other kinds of different variables that you might be interested in. If it does, then what you would say is that that particular survey, net promoter score, has good predictive validity. What that predictive validity means is that it's worthwhile collecting that survey data because it predicts a particular type of deeper variable that you as a firm are interested in. It could be profit, stock prices, other kinds of behavior. Another way to look at how good surveys are is using the reliability. One specific form of reliability is called test, retest reliability. What that means is very simple. It basically says how stable is it, that what you're collecting. If you were to remeasure for example, customer satisfaction and so on, does it vary a lot? If it does, from one time we measure it to the next time, what it tells you is perhaps it's not a very stable measure. It has a lot of volatility. Ideally what you would like to do is to have a measure or a scale or a way of measuring things, which are reasonably stable. So that you take comfort in the fact that once you measure it, it's not going to change very volatile. So those are two ways in which you can measure how good a survey is, validity and reliability. So in the next few slides, we're going to take a concrete example of a survey, the net promoter score, see what are the dos and don'ts of survey design as well. But before I do that, I just want to summarize what are the pros and cons of surveys. Let's first look at the pros. Low cost, relatively easy to implement, good way to learn about potential customers. On the cons, not very easy to write a survey that's non-biased. And we'll bring that up. We will talk about the best practices. The second big issue is how do you get the right respondents? Who are the people that should answer the survey? Again, something I'll bring up when we talk about the dos and don'ts. And what about products that requires some use? So there, surveys might not be the best way to do, why? Because these are products that actually requires customers to use the product. So in that case, what will the people do? Typically you might do, for example, a focus group, where you ask people to first look at the prototype, touch the product, feel the product, use the product, and then perhaps do a survey after that. So these are the big pros and cons of surveys. When you think about implementing some surveys, keep these in mind.
