Mac's Avoids Mindless Discounting

I was listening in on an analyst call with a major player in the marketing automation space. I'm not going to say who, because I've heard similar utterances from all vendors, so it wouldn't be fair to single these guys out. They were talking about a transit system that was using a mobile marketing application to know your likely stop. You log in when you get on (I guess to get loyalty program points—that part wasn't clear), and the system can push an “offer to the Starbucks at your usual stop to come in and get a dollar off a latte.” Look, that's no better than a big “$1 off” sign in the window. I realize the guy was just throwing out an example, except that this mindless discounting is actually what they were doing.

Let's not mindlessly discount—let's be more relevant and offer something that makes better business sense. If you got nothing else from the Cabela's example, at least recognize that we were able to make offers that were relevant without having to discount.

Compare that transit company's push offer to Mac's, a chain of convenience stores (like the transit company, located in Canada) that also sends a coupon to gasoline customers' cell phones while they're at the pump. The coupon is $1 off cash purchases. Note that convenience stores pay about 8 to 12 cents per gallon of gas when you use your credit card, so it only takes a few gallons paid for with cash to for Mac's to make money on this deal. Of course, when you go in the store to pay instead of paying at the pump, the chance of making additional sales goes way up (and sometimes they'll take that dollar-off offer and make it a free Coke with a hot dog or a free hot dog with a Coke). A trade-off like this isn't mindless discounting; the offer creates greater value for both.

You haven't even heard the best part yet. The best part is that when the offer is made via Wi-Fi, the customer can share it with friends. Mac's finds that the conversion rate actually takes off when the offer is a good one, giving them the old hockey-stick sales curve because the share rate will often outstrip the initial take-up rate. And yes, they use DCS to figure out what the good offers are.

That's our goal. Not so much that we get high share rates but rather that we increase customer acquisition and increase customer retention without giving away margin. Everybody wins because the customer gets an offer they like and Mac's gets additional revenue without loss in margin.

Decision Mapping

Mac's offers a good example of avoiding mindless discounting, one objective that all marketers share. To achieve that objective, we have to have a stronger vision of what the customer really wants. Otherwise, we are training the customer to respond to discounts, like Pavlov's dogs to the bell. And once you train them to buy only on discount, you find (as JC Penney did) that you can't go back to full price.

In Chapter 1, I presented a pair of charts that illustrated the concept of accelerated learning about an individual customer in order to respond with a relevant offer more quickly. Just as your salesperson can do in a conversation, you would like your marketing activity to know where a buyer is in the decision process or path to purchase and make the right offer to move her along.

That path to purchase is a series of small decisions. Do I need something? Do I need it now? What do I want? Where can I find it? How much will I pay? If we think of each engagement with a customer as if it were the customer seeking the answer to a question, we see that what we're really doing is engaging in a conversation with the customer. While we are progressively profiling based on what we learn, so is the customer.

Customer-level DCS is about learning in the moment and responding accordingly. For many marketers, customer communication is all too often treated as a one-off event and one-way to boot. Instead, consider customer conversations as a form of decision tree, mapped to match the customer's decision process.

That decision process was mapped by Aristotle centuries ago as being comprised of four stages, illustrated in Figure 7.1: attention, interest, desire, and action (sometimes conviction is placed between desire and action). What few realize is that this model was first expressed by Aristotle as a framework for understanding how rhetoric works in persuading large audiences. While it may also generally express stages that a consumer goes through, the challenge is operationalizing it.

Since Aristotle, many models have been built on how buyers make decisions. A more detailed model that is, I think, easier to operationalize is offered in Figure 7.2.

In this model, you see a circle of five steps. In general, the decision to purchase is thought to be like any other decision. There is a gap between the current state and the desired state. The decision is made that the gap can be filled by a product or service.

Need recognition seems to start the process, except that need recognition is a function of the ongoing evaluation process. You may have loved your car when you bought it, for example, but at some point you begin to realize that it is requiring more and more service, that there are safety features you'd like to have that it doesn't, and a host of other issues. So you decide you need a new car and then you move through the process.

Needs are then specified and translated into features. Based on those features, the buyer searches for products and evaluates them. Included in that evaluation are characteristics of the vendor—reputation for quality, service, and so forth. Then a product is selected, purchased, and used.

The beauty of this model is that it doesn't matter whether you sell to consumers or businesses, the model is just as applicable. Yes, there are situations where buyers short-circuit the process and skip feature specification and vendor/product evaluation. Thirsty? Visit the vending machine and buy a Coke. Not a lot of thought went into any stage.

Note that the vending machine exists because that's how you want to buy. You want convenience. If there is one truism in marketing, it's that there is competitive advantage to be gained in selling the way the customer wants to buy.

What's interesting to me is that we often put together channels based on meeting a need for how the customer wants to buy, but we don't necessarily think that way when putting our marketing materials together. When the customer downloads a white paper, what are the answers that he or she is hoping to find in that paper? Where is the buyer in the decision process?

While this model looks very rational and cognitive, we all know that many purchases are anything but rational. The purchase of an engagement ring; buying a convertible; selecting a vacation home—we may use rational thinking to evaluate features, but the decision of which features are important may be highly emotional in origin.

Sell to businesses? Emotions are a huge part of the decision. Don't get hung up on trying to find examples that don't fit the model; rather, focus on the nature in which one stage may be more or less important for some buyers in certain situations. If I'm thirsty, I may go from need recognition to purchase without much feature specification (other than I like Coke Zero). If I'm buying something complex, I may spend a great deal of time in each stage, exerting a lot of effort to make sure each of the small decisions is right.

Conversations and Big Data

If you are talking with a customer directly, your mind can process words at a rate of 600 to 800 per minute. That customer, however, is talking at a rate of about 120 words per minute. What you are doing with the extra brainpower, assuming you are paying attention, is processing nonverbals along with the verbal message in order to make sense of what you are being told. As you process information, you formulate a reply, and in the process carry out an intelligent conversation.

Your customer, whether speaking or acting, is trying to tell you something. The good news is that Big Data and Dynamic Customer Strategy enable us to carry on an intelligent conversation with the customer as she or he moves through this process. In each stage, and from stage to stage, decisions are being made. What do I want? Why? How do I intend to use it? Will I consider your product? How will I consider it? All of these questions represent decisions, the little decisions that move buyers through the process, and it's your job to give them the information they need to make the next decision.

Big Data offers the opportunity to create streaming insight—that is, as the data streams in, the operational systems analyze and respond accordingly, particularly at the consumer DCS level.¹ When you create models that can take the data the customer offers so that the right offer is made, your system is carrying on an intelligent conversation. First, though, you have to model the path to purchase so that you understand each of the small decisions along the way.

EarthLink's Simple Models

How badly did your eyes glaze over at the mention of propensity scoring? For many marketing managers, logistic regression models are like Nytol, the sleeping tablet. EarthLink, the high-speed Internet provider, serves small and midsize companies and residences throughout the United States. When they began to build a data-driven marketing practice, they used logistic regression and other fancy modeling techniques, but the marketers were uncomfortable with the analytics and didn't really use them.

Sam McPhaul, senior manager of business intelligence, thought a simpler approach was needed to help managers become comfortable with analytics in general. He turned to decision tree analysis, which simply uses a yes/no scoring system to analyze the data.

Essentially, the way decision tree analysis works is this. Let's say the question is, “Who is most likely to churn?” You ask that question so that you can engage in retention activities with those most likely to churn and not waste such activities on those most likely to stay.

For EarthLink, using the decision tree analysis in SAS Enterprise Miner yielded this finding: Customers who left were likely to have called in within 30 days prior to ask if they were eligible for an upgrade. That call indicated they were shopping around. (Figure 7.3 illustrates this analysis.)

But some would leave even without calling. So after breaking out the group who called, the analysis was redone and it was determined that the group who didn't use an EarthLink e-mail account were more likely to leave than those who did. Now EarthLink can call that group and take some action to retain them.

But what of those who had called to ask about the upgrade? Due to technical reasons, not all were eligible for an upgrade, but those who are eligible are given that opportunity. Turns out that if you run the analysis again on those who called but aren't eligible, the low-usage group (based on number of logins) is most likely to churn. Since those are profitable customers, targeting them for the most aggressive retention treatment yields the best results.

(Sidebar—I was once hired to do regression analysis on data about racehorses. The goal was to predict big career winners from data collected at age two in order to know which horses to buy. I couldn't build a regression model any better, or rather not even as good as, a decision tree model. For example, if a horse's sire was 17 years or older (a branch in the decision tree), then that horse was eliminated in a decision tree model while the age of a sire as a variable in the regression model didn't predict. Sometimes, simpler is better.)

Cascading Campaigns

An important point to recognize about the EarthLink or Meredith cases, and virtually every other effective use of DCS, is that DCS enables organizations to use Big Data for making offers that leverage customer needs and desires, not just price. We can end mindless discounting through the effective use of Big Data.

Meredith's responding to a customer's interest in outdoor cooking doesn't require offering a margin-eating special deal to win the business. EarthLink's responding to customer inquiries gave insights into their needs, insights that could be used to increase usage and avoid churn. Both simply require being in the right place and at the right time with the right offer.

Of course, knowing the right time, the right place, and the right offer is not so simple. If it were, everyone would have that right offer at the right place and voilà, problem solved. The reality is that our ability to map the path to purchase is but the first step (illustrated in Figure 7.4), progressive profiling in order to recognize who we're talking with is the second, and crafting offers that matter is the third.

That offer is whatever the buyer needs to move to the next step—the right offer is not always about making a purchase or offering a discount (heaven forbid!). Rather, think about the role that communication or marketing activity plays in the process of moving the customer through the path to purchase or decision cycle. In advertising terms, this is called the call to action; what action do you want the customer to take as a result of having received the communication? But you have to give them whatever they need to answer the call.

The result is what I call a cascading campaign, or series of marketing actions that create an intelligent conversation with the buyer and provide the buyer what is needed to move through the process, such as is illustrated in Figure 7.5, or a campaign that results in multiple paths to purchase. The one in the figure is an actual campaign created by my friends at the Pedowitz Group.

What's missing from this diagram is the call to action in each message. I prefer to include that information so that it is clearer what the message is about. In defense of Pedowitz, this particular example is an actual diagram, so they've pulled some of that information out. Note, however, that they have SLAs, or service level agreements, in there, meaning the marketing team has agreed to deliver on certain performance standards (and sales likewise) so that the system will work.

Some campaigns are pretty easy. It's back-to-school time, so you offer all of the parents in your database an offer based on kids going back to school. That's not hard. Or you're Discount Tire and your CRM system automatically sends an e-mail reminding customers to get a tire rotation (your operational model calculates their average daily mileage based on their last visit, which you then use to determine when the next rotation should occur). But what happens if that customer doesn't come in soon enough? Make another offer, and file each response away because that is part of the progressive profile.

Alternatively, what if you're in B2B and you have a complex product that has multiple people involved in the buying process across a couple of months? Or it takes months just to get them to budget for the purchase and another few months to get them to evaluate and purchase? You've got a lot of people to communicate with over time who need different pieces of information. White papers, trade shows, and other marketing activities can help so each “sale” or call to action is to take up the next activity.

To summarize, campaigns are created to match the wants of the buyer so that the right decision can be made to move them through the path to purchase. Thus each step of the communication process should align with the buying process. What this means from a DCS perspective is that each message is then a trial or test, and if the customer does what we hope, success! Otherwise, the cascading campaign may move that customer into a different group and a new sequence, or it may call for another message designed to reenter the process. That's part of the progressive profiling process—as the conversation continues, we learn more about the customer and are able to redirect the conversation appropriately.

Cascading Campaigns Accelerate Learning

Just as a reminder, for cascading campaigning to work, you've got to have data capture and automated models. Without those, your system will slow down and you'll miss sales opportunities.

Further, this process is about accelerating learning—what works and what doesn't. Conversion, or the rate at which buyers move to the next stage, can only be monitored when you have the data so that as buyers work their way through, we learn what works and what doesn't. Recall Microsoft's free offer (Chapter 3) and their ability to adjust seemingly on the fly—that was made possible because they had the data regarding conversions at each step in the campaign.

A challenge to attribution modeling is that most organizations fail to close the loop. Refer back to Figure 7.5. If a consumer visits your store but doesn't buy until later (perhaps over the Web), how will you know? You can't close the loop if you don't have both data capture and one data file with all of your customer data together.

What amazes me is that even with all of the automated systems now available, there's still an “us versus them” mentality between marketing and either sales or merchandising. The problem is worse when salespeople are involved, because we're so dependent on them for data. Marketing may provide leads but never know what happens to them. As a consequence, attribution modeling is impossible. Marketing can't tell the relative value between trade shows and Web-based marketing, for example, because there's no data provided back to tell whether customers were seen at the show or only visited the website.

You might assume that such a problem would be limited to smaller companies that use salespeople. However, we observed within a Fortune 5 organization (and no, the zeroes are not missing—that's Fortune 5, not Fortune 500) a major operating company that had no worthwhile data from salespeople in the CRM system.

One recent study estimated the cost of a sales call at about $300 to $500.³ Let's just say it is $150 and be conservative. If a trade-show-generated lead closes in two sales calls and a salesperson-generated sale closes in five sales calls, you saved $450 per sale for every lead identified at a trade show. Cut that to one call by inserting a white paper and a $25 chat session with an inside sales rep, and now you've saved another $125. Or if your cost per call is closer to $500, wow! You've really saved.

And here's the funny thing. Your customer is probably more satisfied with that process than the one where your rep started by knocking on a door. Which means, you guessed it—better customer experience, better customer loyalty. Remember, sell the way the buyer wants to buy.

One solution could be to simply compare the trade show lead list to the customer list and see who bought. Unfortunately, you then need to add in the sales activity data to figure out how many calls were made. That's relatively easy with trade show leads, but what about white paper leads? What about call center leads? The more complicated your campaign strategy, the more challenging the data requirements. But creating trackable campaigns is a necessary part of your Big Data strategy.

Accelerating the Process with Multifactorial Experimental Design

Back in Chapter 2, I introduced experiments as a way to learn. Now that we've also covered cascading campaigns, you can see how experiments can be implemented in each step of the campaign. Yet we really want to accelerate learning, and if you have to do A/B testing of each component, learning has been slowed to a crawl, especially if you want to test many components or factors. For example, if you are testing an e-mail campaign, you may want to test all of the following:

• Subject line. Should it refer to a sale or to a new product or something else entirely?
• Opening line. Should it be a salutation (“Dear Karen”) or an attention-getter (“Sale Ends Saturday!”) or…? And if it is an attention-getter, what's the right one?
• Product order. Which goes on the left, which goes on the right? Which gets left out and which gets put in?
• Price?

And that's before you even think about which customer type or types are best suited, when the e-mail should be sent, and a number of other questions.

One alternative is to do a full series of A/B tests. Overstock.com completed one study with 26 variables. If they had done complete A/B tests, they would have had 720 separate experiments running! Who has time for that?

The problem is that you have to control all of the possible causes. So if you have three versions of salutation, then you have three groups. But if you add first product as a category, you now need six groups. Add another variable such as three versions of the call to action and you now have 18 groups. If you remember our discussion of simple A/B tests in Chapter 2 (look back at Figure 2.5), then you can see how quickly these can become complicated. Relax, there are simpler methods for experimental design, though it takes a bit more up-front planning.

SAS and other high-end statistical packages have tools that can allow for the creation of multifactorial experimental designs. One method, Taguchi Block Design, is used in manufacturing and works well with fairly limited marketing examples. A more advanced version is Optimal Design, a class of experimental design techniques that enable the use of multiple independent variables that are manipulated experimentally. If you want to learn more about these procedures, check out SAS Institute, but for now the important thing to remember is that you can do significantly more complicated tests if you design for it up-front.

Summary

Turning Big Data into knowledge is only half the challenge—converting that knowledge into action and doing it quickly is the other half. If you can accelerate learning but you can't accelerate action, cashing in on opportunities is more difficult.

One mantra I repeat often is to sell the way the buyer wants to buy. But customers make a lot of little decisions along the way, so an important use of Big Data is to model the buyer's decision process, then offer the right information at the right time to help that buyer along the path to purchase. This intelligent conversation requires automated models that score buyers based on their actions so that the right material is put in front of them.

The result is a cascading campaign, a series of brief conversations that engage buyers where they are in the decision process. Then, based on buyer choices, the next appropriate step in the process takes place.

Cascading campaigns are also ongoing experiments—what's the right call to action, which salutation draws best—and similar research questions are constantly being tested. Using advanced experimental design, learning is accelerated as campaigns are honed.

In the B2B space, these campaigns nurture leads until they are “sales ready,” ready to receive a sales call. Sales cycles are shortened, costs are reduced, and customers actually like the process better. In the next chapter, we explore how Big Data can improve your customers' experiences with your company and products.