Working with clients, we often times encounter a lot of very specific questions around data collection and personalization: “What kind of data do I need?” “Do I have enough of it?” “How do I know if I have enough?” and/or “what data am I missing?”
To answer these questions, you need to first understand how your model, or algorithm, works. Algorithmic 1:1 personalization requires you to build a model that learns how previous users interacted with an experience to predict which experience is best for future users.
To build a good model, you need quality, relevant data. Why? Well, let’s depart for a second from the digital world and consider an experience we’ve all faced: our commute home from work.
Ask yourself: How long will it take you to get home today?
15 minutes? 20 minutes? An hour? Amazingly, you probably already have an answer to this question; even more amazingly, it’s probably correct within some small margin of error.
But have you ever stopped to think about how incredible it is that we can answer this question at all? That, without any conscious effort, your brain—the most complex learning machine on the planet—has created a model using past observations of your commute time to come up with a seemingly reasonable prediction of how long it will take you to get home today.
More interestingly, your brain doesn’t always come up with the same prediction. Consider the following questions and ask yourself if you have a sense of whether your predicted commute time goes up or down depending on the answer?
- What day of the week is it? Monday or Friday? Is it the weekend or work week?
- What time are you going to leave the office? 1:30, 4:30, 5:30, 5:45, 6:00, 6:30, or 7:30?
- What time of year is it? Is it dark or light when you leave?
- What is the weather outside? Is it sunny, cloudy, raining, snowing?
- Is there temporary construction?
- Is there a car accident?
- Are you in a different location than your typical day?
- How long have you been using this same route?
Not only does your brain have a good idea of the average time it takes, but it can use a lot of contextual data to make the predictions much more accurate depending on numerous factors.
Imagine how much your prediction would change if you didn’t know what day of the week it was or the time of your departure, the weather, accidents, or road closures. It would get worse. In fact, it could be so unstable that it might not be useful at all!
Fortunately, we do collect this data. Otherwise, our lives would be very unpredictable, and our significant others would inevitably be unhappy with us. That’s also why data is critical to personalization.
Put simply, if your personalization algorithm has no, or irrelevant, contextual data then it cannot make strong, stable predictions.
Just like your brain, you need to supply your personalization algorithms with as much relevant data as possible so it can be used to make stronger predictions about the outcomes you care about. This is also why simply turning on a black box technology is not going to be enough to drive ROI.
And contextual data is important regardless of whether you’re doing algorithmic-based personalization or segment-based campaigns. Most targeted (1: many) campaigns utilize high-aggregation data attributes, such as new and returning users, but the more contextual data you can build about your customers the more micro-targeted (1: few) you can become.
Okay so, what data do I need for personalization? And how do I know if I have the right data already?
Back to those pesky questions. If you read our white paper on personalization strategy, you’ll recall that personalization should be centered on your customers and understanding why different customers need different experiences.
With this in mind, the data you collect should help quantify all of these differences about customers, so the algorithm can determine which solution is best for which customer. By using the strategy exercise found in our whitepaper, you can determine if you have the necessary data to segment your customers in order to support your goals.
Since selecting the right data points is highly dependent on your business, it’s hard to give a specific recommendation. However, here are a few buckets we generally recommend:
While the raw data you collect is not very useful, the true value comes when you transform this data to be more descriptive characteristics of your customer.
It’s true that some personalization software and technologies can help you begin the process of transforming your data and creating a unified view of the customer, but this is really where the expertise of data scientists is required to help you create the best contextual data for your specific needs.
Again, the specific contextual elements to consider depend greatly on your customers’ needs, your business and your goals, but here’s where we generally start:
Customer Type: Who are they? How do they compare to other customers?
- Value – How much have they spent? How much do they typically spend?
- Loyalty – Is the customer new? Returning? How often do they visit? How long since the last visit?
- Firmographics – What size business? Industry? What is their potential?
- Demographics – What is their age? Gender?
- Life Events – What stage are they in (recently moved, married, the homebuyer, new hire)?
Behavioral / Transaction: What are they doing now? What have then done in the past?
- Action Indicator – Did they perform the action or not?
- Action Counts – How many times did they perform the action in a given time period?
- Action Sequence – Did they perform an action but not an expected subsequent action?
Preferences: What options do they prefer?
- Content / Product Affinity – What content or product types have they engaged with or purchased?
- Attribute Affinity – What descriptives attributes of the content or products do they gravitate towards?
- Price Point Affinity – What relative price points have they bought or browsed?
Environmental: What is the state of the environment?
- Time Indicators – What is the time of day, day of the week, and/or month of the year?
- Geographics – Where are they? Is it always the same or different?
- Weather – What is the current weather? Forecasted weather? Weather events?
By no means is this a definitive list of all the data you should build about your customers, but hopefully, it helps you start thinking about where you have gaps in your current state.
A Quick Note on CDPs, Profiles, Stitching, and Identity Resolution
While there is plenty of opportunity for personalization within a single channel, the stated goal of most organizations is to create multi-channel or omnichannel experiences. This fundamentally requires that each customer is identified as an individual customer across each channel and the data from one channel can be linked, or stitched, to the data in another channel.
The most important thing to know is all of these technologies have one core goal: continuing to build a better set of contextual data that can be used to provide the most relevant experience possible.
At Brooks Bell, we have years of experience helping brands increase revenue through personalization, whether that’s through maturing customer data, building the necessary technical foundation, or developing personalization strategies and launching campaigns.