Data-driven marketing is all the rage. Why shouldn't marketers use big data to increase marketing performance? With predictive analytics it’s a triple win: your employer can see revenue liftoff 30% or more, your customers will have a better experience and as a marketer, you will gain visibility and respect within the organization.
So what could go wrong?
Here are the three most common pitfalls of using big data for marketing projects:
- Dirty data
- One time models
- Data without actions
Let’s look at each of them in a little more detail.
There is a saying in predictive analytics: “garbage in, garbage out.” If you feed predictive algorithms the wrong facts or incomplete facts, you’ll get the wrong recommendations for your customers.
Important: Sending customers the wrong recommendation is much worse than not sending any recommendations at all.
Data scientists will tell you that data preparation before analysis can make up 95% of all the work. Here are some of the things that could go wrong if you have dirty data:
- Mix up a person’s gender
- Misspell a person’s name
- Undelivered email or direct mail (typos, wrong address, person moved)
- Recommending the wrong store
- Confuse household members
- Send multiple mailing to the same household (also expensive)
Important: Having incomplete information about a customer can be just as bad and lead to the wrong conclusion about customer profitability.
Another common error is perceiving a customer in a linear way and not recognizing for the value that they really stand for. For example, perhaps a customer browses your website frequently but always ends up buying in the store. If you don’t reconcile these data sources you may mistake the web customers as “low value.” In a different scenario - a customer buys a lot but returns items just as frequently or makes very frequent calls into your call center. This behavior also influences profitability.
At a minimum you should link customer data from the following sources:
- Email behavior
- Web behavior
- Online transactions
- In-store transaction
- Loyalty program interaction
- Call center interaction
- Returns and complaints
The data may be controlled by different parts of the organization and physically reside in different databases on different servers. Nobody said it would be easy! The topic of dirty data is so big and important that we’ll dedicate a separate blog post to this topic very soon.
After collecting a complete and accurate profile of every customer, it’s time to apply predictive models. For an overview of predictive models marketers should care about see our previous post on, “The Definitive Guide to Predictive Analytics Models for Marketing.”
Important: plan for ongoing maintenance of predictive models (preferably daily) or they will not produce accurate recommendations.
Many marketers make the mistake of hiring consultants to build a one-time model, but don’t plan for ongoing revision of these models. The problem is that predictive analytics models for marketing (and any other predictive models) require constant tuning. The models will not continue to produce accurate recommendations unless you revise them in small or big ways over time. At AgilOne, we refresh both the customer data and models on a daily basis to ensure ongoing accuracy. If you build models yourself, make sure to plan for the resources that will maintain the models over time.
Data without actions
Perhaps the most important sin of all is to collect customer data and apply predictive analytics, but not take any action. Here’s the dirty secret of big data: data is great, but unless you’re using customer data and customer predictions in your daily marketing campaigns, you won’t see any returns.
Important: unless you use customer predictions in your day-to-day marketing programs you will see zero return.
This problem is not easy to solve and data scientists cannot help you solve it. You’ll need system integration and be able to connect your customer database and predictions database directly with your marketing apps such as your call center, email service provider, direct mail house and website to get a 360 degree accurate reading on your customers. If you use modern applications, they should have open API’s and allow for connectivity, but the process of building connectors can be tedious. So make sure to ask your analytics vendor about their support for your marketing apps.
AgilOne can help you every step of the way to make sure you don’t fall down any potential predictive analytics pitfalls. We’ll make sure you reach your goal of seamless, integrated predictive analytics for your data-driven marketing efforts. Again, it’s a triple win: your employer will see revenue liftoff, your customers will have a better shopping experience and as a marketer, you will gain visibility and respect within the organization.