Marketing Success Sans ‘Every Breath They Take, Every Move They Make’
Last month, I talked about how to measure success when there are many conflicting goals and available metrics flying around (refer to "Marketing Success Metrics: Response or Dollars?"). This time, let’s start thinking about how to act on data and intelligence that we’ve gathered. And that means we get to touch different kinds of advanced analytics.
But before we get into boring analytics talk, citing words like “predictive analytics” and “segmentation,” let’s talk about what kind of data are required to make predictions better and more accurate. After all, no data, no analytics.
I often get questions like what the “best” kind of data are. And my answer is, to the inquirer’s disappointment, “it depends.” It really depends on what you are trying to predict, or ultimately, do. If you would like to have an accurate forecast of futures sales, such an effort calls for a past sales history (but not necessarily on an individual or transactional level); past and current marcom spending by channel; web and other channel traffic data; and environmental data, such as economic indicators, just to start off.
Conversely, if you’d like to predict an individual’s product affinity, preferred offer types or likelihood to respond to certain promotion types, such predictive modeling requires data about the past behavior of the target. And that word “behavior” may evoke different responses, even among seasoned marketers. Yes, we are all reflections of our past behavior, but what does that mean? Every breath you take, every move you make?
Thanks to the Big Data hype a few years back, many now believe that we should just collect anything and everything about everybody. Surely, cost for data collection, storage and maintenance has decreased quite a bit over the years, but that doesn’t mean that we should just hoard data mindlessly. Because you may be deferring inevitable data hygiene, standardization, categorization and consolidation to future users — or machines — who must sort out unorganized and unrefined data and provide applicable insights.
So, going back to that question of what makes up data about human behavior, let’s define what that means in a categorical fashion. With proliferation of digital data collection and analytics, the definition of behavioral data has expanded considerably.
In short, what people casually refer to as “behavioral data” may include this to measure success:
- Online Behavior: Web data regarding click, view and other shopping behavior.
- Purchase: Transactional data, made of who, what, when, how much and through what channel.
- Response: Response history, in relation to specific promotions, covering open, click-through, opt-out, view, shopping basket, conversion/transaction. Offline response may be as simple as product purchase.
- Channel: Channel usage data, not necessarily limited to shopping behavior.
- Payment: Payment and related delinquent history — essential for credit purchases and continuity and subscription businesses.
- Communication: Call, chat or other communication log data, positive or negative in nature.
- Movement: Physical proximity or movement data, in store or store area, for example.
- Survey: Responses to various surveys.
- Opt-in/Opt-out: Sign-up specific 2-way communications and channel specific opt-out requests.
- Social Media: Product review, social media posting and product/service-related sentiment data.
I am sure some will think of more categories. But before we create an exhaustive list of data types, let’s pause and think about what we are trying to do here.
First off, all of these data traceable to a person are being collected for one major reason (at least for marketers): To sell more things to them. If the goal is to predict the who, what, when and why of buying behavior, do we really need all of this?
The ‘Who’ of Buying Behavior
In the prediction business, predicting “who” (as in “who will buy this product?”) is the simplest kind of action. We’d need some PII (personally identifiable information) that can link to buying behaviors of the target. After all, the whole modeling technique was invented to rank target individuals and set up contact priority — in that order. Like sending expensive catalogs only to high-score individuals, in terms of “likely to respond,” or sales teams contacting high “likely to convert” targets as priorities in B2B businesses.
The ‘What’ of Buying Behavior
The next difficulty level lies with the prediction of “what” (as in “what is that target individual going to buy next?”). This type of prediction is generally a hit-or-miss, so even mighty Amazon displays multiple product offers at the end of a successful transaction, by saying “Customers who purchased this item are also interested in these products.” Such a gentle push, based on collaborative filtering, requires massive purchase history by many buyers to be effective. But, provided with ample amounts of data, it is not terribly difficult, and the risk of being wrong is relatively low. Pinpointing the very next product for 1:1 messaging can be challenging, but product basket analysis can easily lead to popular combinations of products, at the minimum.
The ‘When’ of Buying Behavior
Then there is the question of “when,” and such a prediction is not easy. For instance, you may be able to predict if a certain family is very likely to go on a luxury cruise “at some point.” Maybe their demographics and past behavior data are very revealing in that regard. But will they go on a cruise “this year”? Now, that is a very different question. You may need to procure costly “trigger” or real-time data (such as, someone visited travel sites and browsed a series of luxury cruise options). If such explicit data are not available, they you may have to put some educated guesses in based on the “date of last vacation,” and “average days between vacations” for families with similar profiles. Difficult, but not impossible.
The ‘Why’ of Buying Behavior
The most difficult prediction would be “why.” Why does anyone do anything? Heck, we often have a hard time figuring out why even a close friend does anything. We can be total snoops and watch for signs and plow through every comment, but is “figuring out everyone” even feasible on a mass scale? Even so, do you believe every Facebook posting? We may get a glimpse of such data on a social media, but how do we get to use them without looking totally creepy and triggering many alarms along the way? If the goal is to set up the direction of new product development, yes, social data analysis can be very helpful. But that kind of study doesn’t have to be done on an individual level.
I personally gave up on that “why” part a long time ago, and settled on “high correlation.” Yes, correlation does not always equate to causality, but if the goal is to sell things to them, do we really need to know “why” someone is shopping for anything? If we get to know any strong correlation of known behavior to certain product purchases, just rejoice for such findings and send out a gentle nudge to the target individual. Don’t ask “why” every time, unless she volunteers such information in a survey.
Like you see here, each type of question calls for a different type of data.
Know What You Want to Find
And that has been the central theme of this series — that we must define clear goals before we dig into data and analytics. We should not initiate massive data collection and employ cutting-edge machine learning, just because such techniques are available to us.
A while back, a bunch of geeky analysts had this philosophical discussion. If the Genie of the Lamp shows up in front of you and promises to grant you one marketing superpower between two choices, which one should you pick?
- Choice No. 1 is that you will obtain super vision, as if you are sitting on every shopper’s shoulder — online or offline for argument’s sake — and be able to see everything they see. Let’s call that “cameras on shoppers’ shoulders.”
- Choice No. 2 is that you will have access to the complete purchase history data of all shoppers.
Now, if such a question were posed to me, my answer would still be “it depends.” (Hey, at least I am consistent.)
- If the goal were to design the optimal store traffic, the obvious choice would be option No. 1.
- If the goal were to predict each person’s product affinity, No. 2 would be the obvious choice. For something like that, we do not need every piece of movement data (such as online click-stream data), if we have the purchase history data.
We are all reflections of our past behavior, and what we look at does not always equate to what we actually purchase. Maybe a shopper just loves to look at fancy shoes, with no ability or intention of purchasing them? For many, shopping — online or offline — is not just the act of purchasing, but a form of entertainment, as well. Movement can be a good directional indicator, but not the best predictor of future purchase. The shopper’s wallet is more closely related to her past purchases.
Of course, click-stream data are useful, if purchase history is not available. After all, such a massive data collection is the reason why analytical vendors created the concept of Big Data. But why dig through all of those “indicational” data of possible purchases, if actual transaction history can be used? Not me.
Always set the goals first by which to measure success, and then look at the data list (included in this article). Do not think that you must stalk every customer and obtain every piece of data to get into the prediction business for “selling more things.” “Every breath they take, every move they make” is not a good starting point for data strategy.
Stephen H. Yu is a world-class database marketer. He has a proven track record in comprehensive strategic planning and tactical execution, effectively bridging the gap between the marketing and technology world with a balanced view obtained from more than 30 years of experience in best practices of database marketing. Currently, Yu is president and chief consultant at Willow Data Strategy. Previously, he was the head of analytics and insights at eClerx, and VP, Data Strategy & Analytics at Infogroup. Prior to that, Yu was the founding CTO of I-Behavior Inc., which pioneered the use of SKU-level behavioral data. “As a long-time data player with plenty of battle experiences, I would like to share my thoughts and knowledge that I obtained from being a bridge person between the marketing world and the technology world. In the end, data and analytics are just tools for decision-makers; let’s think about what we should be (or shouldn’t be) doing with them first. And the tools must be wielded properly to meet the goals, so let me share some useful tricks in database design, data refinement process and analytics.” Reach him at firstname.lastname@example.org.