For illustration, let’s make the search space a bit more complex. Let’s assume that RCA is interested in offering two different television sets (one with a price $375 or lower, and one with a price $380 or greater). What are the best two offerings to bring to the market to optimize revenue for RCA, again under the previous assumptions of competition and market size?
If we employ an interpolation step value of 10, there are 31 unique prices to investigate over the $300-$450 range. With two products to search simultaneously (and if we didn't restrict the prices for the two offerings), the total possible combinations is equal to (3x3x2x2x31)2 = 1,245,456. With speedy simulation methods like Share of Preference or First Choice, this would still take about 1 hour to search using Exhaustive Search.
Speed and IIA Concerns
Although Exhaustive Search guarantees finding the global optimum, when it takes too long to be feasible in practice, much faster heuristic searches such as Grid or Genetic may be used. Although Grid and Genetic do not guarantee the globally optimal result, they often still find the optimal result or one extremely close to it.
Another consideration with this example is that two products provided by RCA would likely cannibalize themselves in the real world. Simulation methods like Share of Preference are not as preferred as Randomized First Choice when cannibalization issues (also red-bus/blue-bus, or IIA) are a concern.
Randomized First Choice is much slower than the First Choice or Share of Preference methods, so we'll use the following settings for this example to keep things running quickly for illustration:
•Randomized First Choice with just 20 iterations per respondent
•Grid Search
Specifying the Products
Product #1 |
Product #2 |
Product #3 |
Product #4 |
Product #5 |
Product #6 |
|---|---|---|---|---|---|
JVC 25" screen Mono sound No blockout No PIP $300 |
JVC 26" screen Stereo sound Blockout No PIP $350 |
Sony 26" screen Stereo sound No blockout PIP $350 |
Sony 27" screen Surround sound Blockout PIP $450 |
RCA 1-3 1-3 1-2 1-2 =Range(300,375,5) |
RCA 1-3 1-3 1-2 1-2 =Range(380,450,5) |
Simulation Method: Randomized First Choice (click the 
icon to specify that 20 iterations should be used per respondent)
Range Behavior: Search - Grid, optimizing Revenue (click the icon to specify Revenue as the single objective)
The optimal 2 products found in terms of maximizing revenue for RCA (net revenue across both products) are as follows:
Product #5 |
Product #6 |
|---|---|
RCA 27" screen Stereo sound Blockout PIP $315 |
RCA 27" screen Surround sound Blockout PIP $380 |
We can see that as with the one-product optimization solution, an RCA at about $380 with all the best features for screen size, sound quality, channel blockout and PIP is found (product #6 above). In addition, a bit more revenue might be derived if a very similar television were simultaneously offered with the lesser quality Stereo Sound for $65 less (product #5 above), rather than with Surround Sound.
Recall that the best one-product simulation (Search Example #2) achieved revenue of about $169MM, whereas the revenue for this two-product offering from RCA is about $193MM. (We've assumed a market size of 1MM units sold).
All the simulations conducted this point have lacked the important element of cost information. We don’t know whether RCA could stay in business manufacturing and selling these two television sets. Even though we have specified a revenue maximization strategy, RCA may very well lose money on every unit sold if the total costs for these television sets exceed the prices indicated.
As we’ve emphasized before, having cost information is key to obtaining the most value from product optimization searches. Without key cost data, the search results often return trivial answers that are not in the best interest of a company. In the next example, we’ll extend this example by including cost information and conducting a profitability search.