Speaker 1: Now let's take a look at the simulation options that we set up here. We have information about our existing competitors. We have information about the existing market share within the market. We've selected the alpha rule for our decision rule. We did incorporate the incremental revenue information, and we set the revenue for our base product at $50. The first type of simulation we're going to run is going to be based on optimal product design. In other words, we're going to look at all 36 potential combinations of cooler design. When we run the simulation, let me first emphasize that the parts worth estimates, the things that we saw earlier, have not changed. We're simply now taking these results and incorporating additional information about the marketplace to derive simulation estimates. Before we consider introducing our new hypothetical products into the market, let's look at what our conjoint experiment predicts will occur amongst the existing competitors. Let's see if what we actually know about the market share information of these five brands maps onto what our model will predict is the market share based on people's parts worth. Here we can see that according to our predicted model, King Cool should have about a 55.3% based on people's expressed parts worth, but in reality, they only have a 50% market share. Cool Pro and Bro Dude Chill, on the other hand, do have small market shares, but according to our predicted model, based on people's preferences, the market share should be even lower. These discrepancies can come from two places. One, our model can be imperfect. Two, perhaps our predicted market shares are entirely accurate and the actual market shares are entirely accurate, but these differences could be due to differences in distribution, perhaps there was some intense promotional activity that we weren't capturing. So in other words, Cool Pro and Bro Dude Chill, perhaps they are way more accessible in the market. So even though people don't prefer them, they purchase them sometimes. In King Cool, perhaps sometimes there's a stock out situation where people do in fact want to purchase King Cool, but it's not available, so they pick the next best option. With the application of this alpha rule to take people's preferences and map it onto their actual marketplace purchase predictions, we can see that not everybody is assigned a 100% chance of buying their most preferred existing competitor. If you look at participant one and two, we can see that they had a really strong preference for King Cool compared to the other options. With the application of the alpha rule, this did in fact map onto a 100% predicted chance they'd purchase King Cool in the real world. But look at participants three and four. Here participant three had a strong preference for Monster, which maps onto an 86% predicted probability that they'll purchase it, but they also had a relatively high preference for King Cool as well, which corresponded into a 13.9% prediction. For our fourth participant, they really did like the Chillax model, but also were pretty favorable towards the Monster. And again, overwhelmingly, the alpha rule says that we predict they will purchase the Chillax model, but there's still a chance that they'll purchase the Monster instead. Now that we've seen what the model predicts people will purchase before we introduce our new product, let's see what happens when we consider introducing these new products into the market. First, amongst these 36 possible combinations, Ingenious identified that there were three optimal products that were non-dominated. We talk about domination here, we're talking about the relationship between trade-offs between incremental revenue and overall market share. Notice for optimal product three, which would be a 48 can model, seven days of cooling, a 90 day limited warranty, and hard top material, Ingenious predicts that this version of the product is going to have the largest market share, even more so than our other two optimal products. However, it does come at the expense of not as much per unit incremental revenue. You might remember that it costs us quite a bit of money to introduce a 48 can option with seven days of cooling, and that's reflected in this prediction here. These other 33 dots represent the other combinations of the cooler. We simply do not want to entertain any of these options, because in all instances, we could pick either option one, two, or three, and have superior market share and incremental revenue. Now that we've identified three optimal products, each one with a slightly different trade-off between incremental revenue and market share, we can see what our estimated market share and predictions will be against the existing competitors. Notice how if we introduce optimal product three, it predicts that we'll have the largest market share. Based on this chart alone, it seems very tempting to say that we definitely should go with optimal product three. Why not have the most market share? However, let's inspect this a little closer. With optimal product one, our market share is predicted to only be about 20%. However, our revenue per unit improves by $31, where our weighted revenue, which is simply multiplying our market share by the weighted revenue, will be $6.1. In other words, we won't have as high of a market share, but we do think we'll make more money than if we go with optimal product three. At this point, this is where management needs to make a decision. What's our initial goals for introducing this new cooler into the market? Do we want to try to maximize our market share and be happy that we're still making some money? Or are we willing to forego some of that market share and make even more money? Let's see this simulation one more time, but this time, rather than considering all 36 combinations simultaneously, let's select that specific new product profile that we had drawn earlier, the light, elite, and unique option. It's important to keep in mind that these three options that we're about to simulate were also in the previous 36 options. The results look just a tiny bit different. In a world where we're only considering these three versions of product combinations, the simulation tells us that the light and the unique are definite losers. With the light model, we get zero market share. In other words, we're just utterly dominated by all other existing market offerings. Nobody prefers us and nobody wants us. With the unique version of the cooler, we would have a market share of about 5%, but our weighted revenue is relatively meager. The elite option, in fact, seems to be pretty worth entertaining. A relatively high market share and our weighted revenue is quite strong. In a world where we only were able to pick from these three options rather than all 36, it's pretty clear we should bring the elite version of the cooler to the market first. Now that we've concluded our introduction to conjoint design, conjoint experiments, and conjoint analysis, and seeing how this is applied in InGenius, let's wrap up with a few final thoughts. There's a variety of common extensions to conjoint experiments and simulations. Maybe one of the most direct ones, and not included in the InGenius module, is directly incorporating people's preference measurements of competing offerings. In other words, we conduct our conjoint experiment simulating the nine or more versions of the cooler that we would want people to evaluate, but we also intersperse existing competitive offerings and see how people evaluate those individually. So we would now have individualized measures of preference. The other nice thing about this approach would be that we can recognize that branding matters. Even if the product of a competing offering is technically the same in terms of all the attributes and levels, there might be some sort of branding effect that changes people's preference for that product. By measuring preference directly, rather than inferred from market share information, we could get a better understanding. Another common extension to conjoint experiments is that we have to account for what's called non-compensatory attributes. A common example of this might be, I will never buy a car without heated seats. In other words, if somebody gets a non-heated seat version of the car, it literally does not matter what other combinations of features we present to them. They simply are not willing to make a trade-off. Non-compensatory attributes exist throughout the marketing space. Conjoint studies, although not the ingenious version, can absolutely account for these non-compensatory attributes. Another example of a non-compensatory attribute might be me when I'm shopping for a fanny pack. I absolutely demand that it must also have a beer holster. Common extension to conjoint experiments is we want to use these results to make revenue or profit forecasts. Quickly we realize that not every person in the market purchases with the same frequency. In other words, their inner purchase time varies. To solve this, we usually also measure someone's inner purchase frequency and simply multiply them by whatever the baseline assumed default inner purchase time is in the market. So if we presume that most people buy a cooler once every year, people who buy coolers twice a year or buy two coolers a year would be simply multiplied by two. Also, if we want to take the results of a conjoint experiment simulation and map it onto what will really happen in the market in terms of market share, we have to relax the assumption that promotion and distribution will not be 100%. In other words, in the baseline simulation results of a conjoint experiment, we assume that all competitors are fully known, they're always 100% available, and our new product that we're bringing to market will have 100% awareness and 100% distribution. In some cases this might be reasonable, especially for online shoppers, but in other cases this might simply be impractical. As a simple assumption, we might assume that if our new product is only going to be in half of all retail outlets in the first year, we simply cut our market share assumption by about half. Another common extension to conjoint experiments is the use of what's called choice-based conjoint or CBC. In fact, this might not even be so much an extension as more of a prevailing paradigm in conjoint experiments. In choice-based conjoint, you actually present a series of the different options simultaneously to the respondent and you ask them to pick which one they most prefer. The example on the bottom left here for golf balls illustrates this. Notice in the example that we did in this lecture, we actually had people directly score each option one at a time. Choice-based conjoint, on the other hand, has people select from a series of alternatives. Another common extension to conjoint analysis is the use of market segmentation. This can go in both directions. Sometimes, marketers first perform needs-based segmentation of the market, something like k-means. Then, within each of the identified needs-based segments, they conduct a conjoint analysis. This can be useful because hopefully, each one of those needs-based groups would be similar in their preference evaluations within the segments but be different across. As an alternative, conjoint analysis results can be directly used as the inputs into a segmentation model. Keep in mind that conjoint analysis has been considered one of the great analytical successes of marketing research. As sometimes people say, the proof is in the pudding. Despite the many limitations that can come along with conjoint experiments, they've been deployed for over 40 years for marketing research, and their continued use and track record of success show that marketers will continue to be using them in the future. Soft2 Software is probably the industry leader in specializing in conjoint experimental design, implementation, and analysis, but there's a variety of other research firms that provide conjoint analysis service as well. Or, with this video, maybe you can just do it yourself.