|
|
Media Coverage
"Optimize
Your Next Big Marketing Play"
iMedia
May
24,
2007
Optimization is a powerful tool for improving the performance
of your marketing programs. The good news is it really
works.
What does it mean to say we "optimize" online marketing? To many, it connotes some sort of improvement, but it is far more than a fancy word. Improvement is a linear, single step in making something better, while optimization is a mathematically rigorous evaluation of the full range of practical options leading to the selection of the best available choice.
Here is a good example: Painting a house is not the same as optimizing its resale value (though it helps). If no attempt is made to evaluate the full range of alternatives for spending the time and effort on the house, you're merely improving (without optimizing) the price your house will bring.
What are we optimizing?
We need to define an optimization objective in terms of something measurable. An objective function describes this goal mathematically, in terms of the factors we can vary or control.
With the objective function defined, optimization is the maximization or minimization of that function. In plainer English: We are either trying to capture as much of something good as we can or avoid something painful or costly as much as we can.
What success metric are we trying to optimize in online marketing? Here are some candidates (there are many others):
- Clicks
- Page views
- Revenue
- eCPM - effective cost-per-acquisition
- VPM - value produced per thousand impressions
- Ad revenue
- Incremental profit from up-selling or cross-selling existing customers
- Member involvement (uploads, responses, postings, et cetera)
Optimization is a technique for making decisions involving complex trade-offs. In deciding how and where to apply it, it is useful to think of it as a way of converting on latent demand, which is hidden or hampered by:
- Wide variation in your audiences' product knowledge, needs, desires, or cultures
- Too large a number of products/offers to display all at once
- Offer eligibility rules, localized versions
- Geographically-driven pricing
If you only have 3-4 seconds of a customer's attention, then it is critical to show them the right offer when you get an opportunity. To optimize this interaction, you really need to have:
- A range of products/offers that are materially different
- An audience containing subgroups that differ in terms of what they need and what they respond to.
It takes data to optimize (and scale and time)
A statistically rigorous approach to optimization is inappropriate for cases where:
- The desired outcome is too infrequent to provide enough history to model
- There are no measurements of the predictors or factors that drive the outcome
- The campaign will not be consistent or stable long enough to apply learnings
Real optimization requires the collection of sufficient history to support statistically valid conclusions linking what you can control to what you want to happen. Anything less is merely a hopeful stab at improvement.
The sky is NOT the limit
We can't always unleash optimization on a simple objective function and get a practical result. Real-world optimization problems frequently involve constraints that need to be included in how we set up our objective. Some examples are limited budgets, multiple product marketing teams that expect their offers to get equal time and space, offers that are mutually exclusive, products that should never be presented together, or products that should always be presented together.
To measure the effectiveness of an optimization program, you need to compare the performance of your optimized program to a relevant reference point. You might be tempted to simply compare pre-optimization performance to optimization performance, but you're taking a huge chance if you do and could wind up mistaking the natural variations or seasonal ups and downs in your business for the effects of your program. This is a common optimization rookie error: don't do it. Some better options would be:
- Run a portion of your impressions to serve as a random control
- Run a portion of your impressions according to your pre-optimized targeting rules to serve as a "Business As Usual" reference point.
- Run your best-performing creative/offer/product to serve as a "Current Champion" reference point
Conclusion
Optimization is a powerful tool for improving the performance of your online marketing programs. It's not magic and requires some commitment. It also doesn't work for all programs. You need scale, offer choice, a heterogeneous audience and a measurable objective. The good news is it gets easier as you master it, and it really works.
<< Return to Previous Page |
Privacy
Site Map
Home
![[x+1]](../images/logo_xplus_main.gif){kind=logo}
