The Problem With Predictive Analytics

Let's pretend you need to close 30 more customers to hit your sales target.

Luckily, you just so happen to have a whitepaper with a customer close rate of 30%. You might assume that if you get 100 more people to download that whitepaper, you’ll get 30 more customers and everythingwill be gravy -- because based on its past performance, that's the outcome you can expect.

If you're thinking like that, you're using predictive analytics. Predictive analytics are a way of making decisions by taking the data of past performance, and using that data to expect future performance. If you've ever worked in the actuarial side of insurance, financial services, even healthcare, there's a good chance you've come across it before.

如今，还有另一个行业正在利用它：营销。问题是，我们可能使用预测分析都是错误的。

What's the Problem?

好吧，第一个问题是营销本身 - 这不是一门硬科学，因此，结果可能远远超出了预期结果的历史范围。让我们以白皮书为例。我们的假设是，我们可以关闭30个客户可能是一个可怕的假设 - 因为在该人群中有可能发生一些变化，而我们预计可能会导致我们关闭或超过预期的30个。

Here's a real world example of predictive analytics gone awry that you might recognize. A lot of people have used predictive analytics as investment tools to build mortgage backed securities. Using traditional statistics developed for hard sciences, they predicted a risk of default, and their models told them it was statistically impossible for everyone to default on their mortgage at the same time. But that failed to take into account certain elements of the housing market. And so what happened? Everyone depended on those models being correct, but their failure to take unknownsinto accountresulted in, well, some unknowns that broke the model. The unanticipated result was a lot of people defaulting on their mortgage, and a huge economic collapse in 2008.

Accounting for Unknowns

您可能会说好，我只需要考虑未知数即可。但是关于未知数的事情就是 - 他们unknown。您可以尝试预测它们，但是您无法解释所有这些。这就是黑天鹅理论的进来。

You see, people actually started to try to account for those variables by building up models for it -- and then the models got so complicated that people wereso surethey were correct because ... well ... because there were a lot of models and variables and they were really complicated. They had to be right ... right?

好吧，人们在这些模型中建立的许多东西都是known未知。但是，未知数的现实实际上更像是墨菲定律。大型会议的那天，您知道会发生不好的事情，您只是不知道会发生什么。(Thanks for the analogy,Katie Burke）

让我们以天鹅为例。在16世纪的伦敦，人们认为所有天鹅都是白色的，因为所有天鹅的文件都表示它们有白色的羽毛。然后在1697年，一位荷兰探险家在澳大利亚发现了一条黑天鹅 - 换句话说，曾经被认为不可能的事物被一个未知所拒绝了。

简而言之，这是黑天鹅理论Nassim Nicholas Taleb。这是您甚至不知道的事情可以改变您的预期结果。因此，如果您构建了试图考虑黑天鹅的模型 - 好吧，您不能。这才是重点。您不能预料到每个未知的人。而且您必须考虑that当您倾向于预测分析时。

问题是，没有足够的人这样做。这就是营销中预测分析的问题。

Predictive Analytics Still Have Their Place

I've made predictive analytics sound pretty dubious thus far, but we shouldn't throw the baby out with the bathwater. Using predictive analytics is flawed when you use it to build up complex models, and then base your marketing decisions off of those models with the expectation of 100% accuracy. But I asked around to other marketers who still say there's still a lot of cool stuff you can do with it.

Matt Wainwright, Director of Marketing atAttend.com, says he uses predictive analytics to determine which pieces of content he should repromote. "What I don't do is make guarantees of specific results based on numbers like a 20% close rate vs. a 5% close rate. What I做做做出明确的决定，即20％的近距离率可能比5％的近距离要好，即使可能仍然有些事情可能会影响我无法解释的近距离。”

In other words, you can't divorce yourself from common sense.That's why probabilistic analytics are a better bet than predictive analytics. You account for the variables you can anticipate, use that data to make smart decisions, but don't guarantee specific results.

如果您无法抗拒研究预测分析并根据过去的表现做出具体预测的冲动，请负责任地做到这一点。预测分析不应用于对大型赌博进行预测 - 只有低风险。幸运的是，营销通常比您可能遇到这种方法的许多其他行业要低的赌注。请记住，当P毫无疑问，当您开始将它们视为实际预测时，您也可以将它们视为预言，这无疑是有帮助的。