Probabilities and Statistics Pay-Off for Marketers
August 20, 2010 by Stics·
Leave a Comment
Conditioning Probabilities
When observing our customers, sometimes evidence can look pretty persuasive and sometimes it can be down-right compelling. And yet, without proper attention to underlying statistical properties in the customer data, you might form an entirely false impression of your customers. It is not uncommon for marketers to ask this question, at least once in their career.
“How did I make such a wrong conclusion from such compelling evidence”?
The answer lies in its numerical underpinnings, what statisticians call “conditioning”. Conditioning can cause you to reach the wrong conclusions and send lots of marketing dollars down the rabbit hole and take you to the Mad Hatters Tea Party.
The unhappy results of conditional probabilities exist in all facets of animal and human behavior. These are typically complex scenarios of layered assumptions, so I am using this simple example, of one type of casino player, to demonstrate how the application of incorrect assumptions can lead to lost revenue and profitability.
Casino Gambler Example
Let’s assume one percent of the casino gamers who played at your property were cheaters. That percentage might be high (Cheaters = 1%), but this is an example after all. Let’s also assume that when a player is a cheater, he or she will decline to fill out a loyalty Rewards Club card application about ninety nine percent of the time. (If cheater, 99% decline application.)
Now comes the interesting part. Let’s say that you see a suspicious player, approach him or her, ask the player to fill out a rewards application, and the player declines. What are the chances that that player is a cheater?
Intuition vs. Reality
Intuition would suggest a high likelihood of cheating. But the reality is different. There is only about a one in thirty-three percent chance that this individual casino player is a cheater. So consider the statistical probabilities before you decide that the player is not worthy of a revenue generating comp or a coupon.
Here’s a table that describes the situation exactly. You have total players showing up in green. In red are the 1% or 100 are cheaters. 99 of those 100 cheaters were not willing to fill out an application, so your test was very good.
Wrong Assumptions Equal Lost Profit
So here is the big question. If someone refuses to fill out an application (the red boxed group), should you refuse to serve them or market to them? The answer is No for the following reasons.
- First, there is only roughly a 1 in 33 chance (more precisely 99 of 3,399) that the player is a cheater.
- Second, there is a 3,330 out of 3,399 chance that they are a good player who will be profitable.
- And finally, these 3,300 players are likely to be even more profitable than reward applicants.Even when making generalizations about customers, it is essential to understand the effect of your assumptions on your conclusions. Especially when formulating policy, it really pays to know your probabilities and statistics.
If you found this interesting or want to learn more about the power of statistics for marketing, feel free to contact Stics. We are happy to help.
How to Tell What Your Customers Want
August 1, 2010 by Stics·
Leave a Comment
Past Behaviors: Misleading Facts
When we make guesses about the future, we usually go off past behaviors. But people change their minds frequently. So even if you know what your customers bought in the past, you cannot necessarily predict what they will buy in the future ― from their past behavior alone.
To illustrate that point, let’s look at Bob, Carol, Ted and Alice. We want to know if we should be marketing SUV’s to these people. From our customer records, we know they bought an SUV in the past.
As you can see from the chart, the traditional SQL (Structured Query Language) a
nswer correlates their past behavior with a recommendation to market to these people. But this approach falls short compared to the deeper statistical answer, because it only evaluates a limited set of the available data.
- For Bob, both the SQL and statistical answers are in agreement, telling us that he is worth marketing to because he is an SUV buyer.
- For Carol, the two answers lead to opposite conclusions. Perhaps she no longer needs one because her kids are grown. Maybe it’s something else that changed?
- Both Ted and Alice have some chance of buying an SUV having never done so previously.
- But Alice has a much higher probability of buying one. Perhaps she is starting a family or taken up a new sport that an SUV would be good for. Without statistical analysis, we would not know that Alice is worth marketing to.
Predicting the Future
The better way to make predictions about future behavior is to use a statistical model. A statistical model can take many complex inputs and produce outputs, like the probability of someone buying an SUV in the future.
The important difference here is that statistical models can respond to all the details within your data. This is superior compared to using generalities or segments of your data, like the “previously purchased” example shown above. By using statistical models rather than a traditional SQL approach, your will gain a better view of your customers and better refine your marketing efforts with increased accuracy and profitability.
This is one of the reasons why Stics statistical models improve marketing performance above other methods. Stics can provide customized statistical insight about your customers – and use less time than traditional segmentation models take to make. Plus, Stics has years of experience and the technological tools to quickly and reliably give you the information you need.
