Optimization
Episode 09
Episode 09 Transcript
Hi, I’m John Colias, Senior Vice President of Decision Analyst’s Advanced Analytics Group, here today to talk about media mix modeling. And today we’re going to focus on optimization. Optimization happens at the end of the media mix modeling project. And in previous episodes we talked about the equation that’s produced by the media mix modeling exercise.
And this equation is used in optimization. It has inputs of the type of, the spending by type of media and by month, and it has an output of the net revenue (the revenue produced by selling the product) minus the media, (the cost of the media itself) to produce net revenue. And this net revenue has, becomes an objective, it becomes the objective function in the optimization exercise. And the objective function is maximized with an algorithm subject to certain constraints. And the first constraint would be (as you can see in the depiction here is) the total media budget. How much is being spent on the total media budget? And this constraint is important because the algorithm would basically explode and produce unrealistic numbers if we didn’t have some cap on the total media budget. .
The second type of constraint is a constraint on which media types can vary when doing the maximization of the objective function. The, by selecting the types of media that we want to optimize over, and which months we want to optimize those media types for, we’re actually constraining the other media types and those months to not vary, and that’s the constraint actually. And then there’s a realism constraint, a constraint on the upper bound of the percentage of the media budget that would be allocated to each type of media. For example, TV advertising, we might realistically never want it to be less than 10% of the total media budget for the year or no more than 60% of the total media budget for the year. And so with this objective function with these constraints are maximized by using a nonlinear programming algorithm. The nonlinear programming algorithm takes the objective function and actually maximizes it subject to these constraints and produces an output (which is the amount of spending by type of media) that varies for the particular months that we selected to vary, that maximizes the net revenue.
Now one caveat about this is that there are many nonlinear programming algorithms available, and some of these work well and some work, I won’t say badly, but not as well, with very complicated nonlinear functions in which most of these in the media mix modeling space would turn out to be a fairly complicated function to optimize. So it’s important to select a good algorithm that works well and is robust. For example, if the default solver algorithm that comes in Excel, which is a nonlinear programming algorithm actually, is probably not good enough for this exercise, it would typically stop before getting to the optimal solution. So it’s important to find the best nonlinear programming algorithm before engaging in this exercise. Thank you.
Thank you.
Presenter
John Colias, Ph.D.
Senior VP Research & Development
As a leader with both university teaching and business consulting experience, John focuses on predictive modeling, prescriptive analytics, and artificial intelligence. As Senior Vice President, Research & Development, at Decision Analyst, John combines academic and business interests to help analytics professionals by offering cutting-edge analytic solutions tempered by business realism. He holds a doctorate in economics from The University of Texas at Austin, with specializations in econometrics and mathematical modeling methods.
