Chapter 11

Making Marketing Improvements

IN THIS CHAPTER

check Uncovering popular use cases

check Mapping and scoring your sales-and-marketing channels

check Getting hip to modern data privacy requirements

check Rolling your own marketing mix models

Each and every customer has their own, unique set of preferences, needs, and wants. The better you can align your business and its offerings with the most intimate desires of your customers, the easier it is to convert leads into new customers and make repeat sales with existing customers. Luckily for you, data science, when combined with data analytics, is the perfect vehicle by which to offer the personalization that today’s tech-savvy consumer craves. In this chapter, I have you look at several ways to go about using data science in marketing to improve the profitability of your company’s marketing efforts.

Exploring Popular Use Cases for Data Science in Marketing

You can use data science in a lot of ways to increase your marketing return on investment. These are the most common marketing use cases for data science:

  • Recommendation engines: Whether you build them yourself or use a recommender application instead, recommendation systems use filtering to segment customers according to shared characteristics. It’s useful to segment customers in this way so that you can target offers for your customers’ known preferences, in order to upsell and cross-sell your brand’s offerings.
  • Churn reduction: Customer churn describes the loss, or churn, of existing customers. Customer churn analysis is a set of analytical techniques that are designed to identify, monitor, and issue alerts on indicators that signify when customers are likely to churn. With the information that’s generated in customer churn analysis, businesses can take preemptive measures to retain at-risk customers.
  • Content creation: Though the use of AI for content creation isn’t exactly new, it has become a lot more accessible lately. Natural language processing (NLP) is used in copy-writing applications as well as in applications that can read your copy back to you after you’ve written it. This is in addition to the NLP-generated text suggestions that we all have come to know and love inside of word processing applications like Word or Google Docs. Recently, even newer AI tools have emerged that can create new original content from a prewritten sample — GPT-3, for example, which is able to collect text on a single topic from many different sources, consolidate that to a single research finding and then write it out as copy that can, in some cases, pass for human-written copy. Refer back to Chapter 10 for more about GPT-3.
  • Lifetime value forecasting: Modern lifetime value forecasting (LTV) involves using machine learning to predict the lifetime value — the total monetary value of a customer over their lifetime with your company, in other words — from existing customer data. Traditional approaches for LTV calculations used basic averages and groups to make predictions, but developments in data science have made it possible to generate much more accurate, reliable predictions of LTV using machine learning.

  • Hypertargeted advertising: Hypertargeted advertising is a direct product of data science and machine learning. Social media platforms, like Facebook, have been utilizing its users’ behavioral and interest data as well as data from its data partners to make predictions about which users are most likely to convert to customers, which are likely to opt in to become leads, and which are likely to become likely to opt in if they’re made more aware of the problem solved and solutions offered by your company.

    Remember Facebook takes all the heat, but if we consumers are truly being honest with ourselves, all the social platforms, search engines, and mobile apps out there are part of the race to monetize their platforms from either targeting ads or selling user data by way of data partnerships (a partnership between companies wherein data access is bought and sold). If you find yourself angered by this situation, just realize that there truly is no such thing as a free lunch. Someone has to pay to develop and maintain these free behemoth telecommunication tools we consumers are all so fortunate to use. If you’re unwilling to be advertised to, you should be willing to pay for your usage. And if you’re willing to be advertised to, what do you prefer — an ad that’s extremely interesting to you about a product or service you didn’t know existed and would love to purchase? Or an irrelevant ad spamming you about lame products you don’t even like or want to know about? As someone who lived through the dotcom bubble, let me tell you: Nontargeted ads are spammy and no fun to experience.

  • Dynamic pricing: In this form of real-time pricing optimization, machine learning models use clickstream data (and customer data, where available) to generate predictions for the optimal price point at which a customer will buy. A dynamic price is set in real-time, while the user is on the website, in order to maximize the likelihood of a sale.
  • Channel scoring: Channel scoring helps you identify the best channels to use for the specific campaigns you’re working on. By analyzing your current marketing strategy, you can identify where your customers are coming from and then assign a score to each channel. You can also identify channels that are performing better than others — as well as those that are putting a drag on your overall strategy. (There's a five-step process for how to do on channel scoring later in this chapter.)
  • Lead scoring: Lead scoring is the practice of first assimilating data you have on a lead and then calculating out a score for each lead to indicate the stage of a lead — anywhere from cold to piping hot or (hopefully) already converted. Having this score on hand is helpful in categorizing your leads and predicting which type of marketing campaign will be most effective for each particular group in terms of converting people in those groups from leads to paying customers.
  • Marketing mix modeling: Marketing mix modeling involves taking historical sales-and-marketing time series data and using statistical machine learning methods to uncover relationships between your core product and marketing features and sales made of those products. After you determine the nature of those relationships, you can predict future sales and tweak your company’s marketing plans accordingly. (I cover marketing mix modeling in greater detail later in this chapter.)
  • Market basket analysis: Market basket analysis involves analyzing transactional data to identify products that are commonly purchased together in pairs or small groups within a single checkout transaction. After uncovering these product pairings, retailers can then use that information in their product placement strategy to increase the number of items sold during a single customer visit. (This analysis is widely used to support in-store product placement.) Product placements should be done in such a way that the items frequently bought together are displayed next to each other so that customers are encouraged to buy them, resulting in a boost in sales.

Tip For some coding demonstrations showing you how to build recommenders in both R and Python, be sure to check out the free training resources I’ve left for you over on http://businessgrowth.ai.

Tip For a demonstration showing you how to do it in Excel, check out the free market basket analysis demonstration I’ve also left for you over on http://businessgrowth.ai.

Although I’d be delighted to write an entire book on the marketing data science topic, I have only a few pages to cover it within this chapter. For that reason, I’ve decided to do a deep dive into some of the more accessible methodologies — the ones that don’t require you to purchase new software to get started, in other words. Those freebie methodologies are

  • Omnichannel analytics
  • Channel scoring
  • Marketing mix modeling

I discuss each methodology in greater detail in later sections.

Turning Web Analytics into Dollars and Sense

Do you know which marketing and sales channels are driving the greatest number of new customers into your business? If your company is like most, the answer is no. In this section, I give you the lowdown on omnichannel marketing, channel analytics, and how to increase your marketing ROI with a simple 5-step approach to channel scoring that even the most novice data analyst could implement.

Web analytics is a generic term that represents the practice of tracking and evaluating website data in order to increase leads, make sales, and deliver excellent customer experiences. The use of visitor location and demographic data is the most basic form of web analytics, but basic won’t get you very far — so let’s focus on omnichannel analytics instead. It’s a type of web analytics that has the highest likelihood of getting you traction — fast!

Getting acquainted with omnichannel analytics

Omnichannel analytics are a form of data analytics that clarifies your customer interests and expectations along each of your sales-and-marketing channels in the hope of directly increasing the number of leads and sales that are generated from your company’s marketing endeavors.

An omnichannel approach assumes that you have several channels through which you market to customers and that you have data resources you can use to identify what your customers want to see on each channel. Through omnichannel marketing, you can provide your customers with different experiences on a channel-by-channel basis; thus allowing you to cater to your customers' different expectations and personalize their experience with your company based on where they’re interacting with it and at what stage along their customer journey they find themselves.

Mapping your channels

To make sense of your customer interactions (often called “touchpoints” in marketing speak), you first need to map out where those “customer touches” take place online or in the physical world. With respect to omnichannel analytics, the term channel refers to either sales channels or marketing channels, or both. You’ll want to start your journey in omnichannel analytics by first mapping out all your company’s sales-and-marketing channels.

Sales channels are the channels through which your company generates sales and distributes its products and services to the buyer. Your company’s sales channels might include these elements:

  • Company website: If your company offers products or services that are available for purchase online, then the web pages where sales are generated and the product or service is distributed is considered a sales channel.
  • Company email list: Technically, sales aren’t made directly within email messages, but emails can be used as sales assets that directly trigger sales. It's often the case that, in the ecommerce industry, the distribution of the product or service happens by email, so consider your company’s email list as a potential sales channel.
  • Brick-and-mortar store: If your company has a physical store where people enter and purchase tangible products or services, then each physical, brick-and-mortar store is considered a sales channel.
  • Sales calls: If your company offers sales calls that directly trigger new sales, the sales call is considered a sales channel.
  • Live events: Lots of companies host or sponsor live events as a way to directly trigger sales from attendees. If your company participates in an event with the intention of making sales, you should treat that live event as a sales channel.

Turning now to marketing channels — those channels through which your company generates leads as well as sales for your company — the most common ones are described in this list:

  • Paid ads: Paid ads are one of the most commonly used ways to increase brand awareness, generate leads, and drive sales. Each ads channel should be treated as a marketing channel.
  • Website traffic from search engine optimization (SEO): Search engines, like Google and Bing, are more than just tools for finding fast answers to your burning questions, like “Which came first — the chicken or the egg?” From a business and marketing perspective, in fact, search engines are useful tools that should be utilized and optimized in your company’s favor to increase brand awareness, leads, and sales. Within omnichannel analytics, SEO traffic sources are definitely treated as a marketing channel.
  • Website traffic from social media: Your company’s social media accounts are a primary touchpoint where brand representatives can interact with customers, create brand awareness, drive leads and even make sales. Within omnichannel analytics, your company’s LinkedIn pages, Instagram pages, Facebook pages, and the rest — should each be treated as a separate marketing channel.
  • Live events: Most companies use live events as a way to increase brand awareness and word-of-mouth marketing as well as to generate a fresh set of warm leads. Regardless of whether you treat them as sales channels, all live events should be treated as marketing channels.
  • Referral sites: If your company offers engaging, free blog content on its website, if it has affiliate partnerships, or if it throws an occasional sales event, you should see some referral sites — referral traffic that’s coming from other websites on the Internet — within your web analytics. These referral sites should be treated as marketing channels subject to omnichannel analytics.

Building analytics around channel performance

To build an effective omnichannel analytics strategy, you need access to data from a variety of sources. For sales channels, you need financial data that reflects the total dollar amounts of sales as well as details about where those sales were generated. Keep in mind that, if your company is running ads, the ad costs should be subtracted from the revenues generated along each channel in order to paint a more accurate picture of channel performance with respect to leads and sales per dollar invested into ads.

For marketing channels, you can rely heavily on Google Analytics for tracking how your online marketing channels are performing with respect to the behavior of website visitors, and how well each social or referral network is generating leads and sales. Unfortunately, evaluating your marketing channels isn’t as simple as just looking at your Google Analytics data. You also need to lean on your social media analytics that are provided by each of the social platforms separately, or a third-party vendor that provides the intel to you. Lastly, for any live marketing events, you have to gather follow-up statistics from the marketing personnel in charge of the event.

Remember The purpose of omnichannel analytics is to support your company in providing your customers with different experiences on a channel-by-channel basis. The idea here is to allow you to cater to your customers' expectations and personalize their experience with your company. This type of personalization goes a long way in terms of creating brand trust.

Scoring your company’s channels

Analyzing your current sales-and-marketing strategy means you’ll be able to identify where your customers are coming from and then assign a score to each channel based on that channel's success. The scorecard you create then serves as a visual representation of the current importance of your various channels, relative to one another. The scores also help you identify and improve underperforming customer touchpoints. Armed with this knowledge, you can optimize and improve your marketing and sales strategy. Your overall strategy will not only make more sense but also provide a data-informed foundation for improving the ROI of your future multichannel marketing and sales campaigns.

You can go about scoring your sales channels in a number of ways, but I’ve created a simplified 5-step approach, just to give you a quick snapshot of what's involved. Here goes:

  1. Map your channels.

    Itemize all channels through which sales are generated. I discuss this step in the earlier section “Mapping your channels.”

  2. Score your channels.

    Evaluate each of those channels and score them out, based on the characteristics of customers they generate. Important metrics to look at here may include these:

    1. Customer lifetime value

      Machine learning comes to the rescue when generating accurate LTV forecasts.

    2. Customer reviews and satisfaction metrics
    3. Upsell, downsell, and subscription renewal rates
    4. Ticket volume
    5. Customer profitability

    These metrics help you understand the quality of the customers that are coming through each of your sales channels. Just looking at these metrics alone can help you identify areas of opportunity, as well as areas to avoid, if you want to attract better customers for your business.

  3. Create a channel scorecard.

    Summarize your findings for each of the metrics by creating a scorecard for each channel.

  4. Define a customer avatar for each channel.

    A customer avatar is a data-infused description of a company’s ideal client — it describes both their demographic and their psychographic. Define your customer avatar by using your scorecard alongside behavioral analytics for each of your channels, making sure to also consider known attributes of existing customers within these channels, to make some educated guesses about what types of customers fit into each of the channels in your channel portfolio.

  5. Tweak your sales-and-marketing strategy.

    Looking at this customer avatar along with each channel scorecard, decide which changes you can make to improve a particular channel's performance so that it better supports your company's overall sales strategy and goals.

Figure 11-1 shows a sample of what your channel scorecard might look like after you’ve completed this 5-step process.

Schematic illustration of a channel scorecard.

FIGURE 11-1: A channel scorecard.

Unfortunately, there is no exact, cut-and-dried formula to use for assigning a score to a particular channel. You really need to dig into your channel numbers and determine why particular channels are generating the most leads and sales — these metrics should be weighted in importance. Then look to see how that success is reflected in the channel data in terms of customer engagement statistics with your channels. Based on the numbers you collect for each channel, you then need to assign a relative score for all your sales-and-marketing channels.

Tip I’ve left a handy Channel Scorecard Calculator template you can use to get started over on the companion website for this book at http://businessgrowth.ai.

What can scoring your company’s channels in this way do to improve your company's sales-and-marketing strategy (and thus your company’s bottom line)? I'm glad you asked. Here are the main advantages:

  • Customer acquisition: When you fine-tune your marketing strategy so that it aligns better with your customer desires and expectations along each channel, your marketing ROI immediately increases, as does the revenue associated with each new customer acquisition of new customer acquisitions from leads you’re nurturing along each of those channels.
  • Customer retention: Fine-tuning your sales-and-marketing strategy so that your company keeps on pulse with changes and the evolution of its customer desires helps keep your existing customers coming back for more — driving an uptick in repeat purchases and word-of-mouth marketing.
  • New product or service development: By scoring your channels in the way I’ve just discussed, you have a much more granular view of your customer and their preferences. This perspective is, of course, helpful in designing products and services that your customers need, want, and adore.

Building Data Products That Increase Sales-and-Marketing ROI

Marketing optimization is one of the more popular use cases for data products. Unlike 30 years ago when 99 percent of products were tangible goods, today the word product has come to represent a wide variety of assets, including these:

  • Tangible goods: A tangible good is a physical item that is bought and sold on the open market.
  • Software: This one includes stand-alone applications, Software as a Service (SaaS) products, and even templates that are built on top of SaaS products.
  • Software features: In this context, a software feature is a functional unit within a software package that satisfies a business or user requirement. If you have product management experience, you’re already aware that a feature is the product when you’re managing software products.
  • Digital products: This category includes all forms of information products that come in electronic format, such as customized calculators, dashboards, courses, e-books, PDF documents, and slide decks.

When I talk about data products, I’m referring to monetizable products that are derived directly from any combination of data resources, technologies, and skill sets. Data products are a popular way to monetize data — a topic I cover in greater detail in Chapter 14. To name a few types of data products out there:

  • Software and software features: Any sort of data-intensive software would qualify as a data product. This of course includes software packages that are exclusively designed for machine learning, data visualization, data clean-up, and so on. Most software these days includes at least some predictive AI or data visualization components. Regardless of what the rest of the software does, these features would themselves qualify as data products.
  • Digital products and certain tangible goods: This includes all forms of digital products that rely heavily on data resources, technologies, or skill sets.

Remember A data product manager is someone who uses data science to generate predictive insights that better inform decision-making around product design, development, launch, and strategy. Because a data product manager uses data expertise to improve products in this way, you can get away with calling the products they manage data products regardless of the data resources and technologies that play a part in the product development and delivery.

Popular marketing data products include the obvious ones like Google Analytics and Adobe Analytics, but also less-obvious data products like marketing dashboards, data-intensive custom marketing strategies and plans, and even the social analytics features inside your social media accounts that tell you how your posts are performing. If you’re curious how to go about using data and data expertise to build better marketing products and services, pay close attention to the first three chapters in Part 4.

Tip In my own business, I use Google Data Studio and Segmetrics, both powerful yet affordable marketing analytics products. To make a free copy of the Google Data Studio templates I recommend, be sure to check out the Google Data Studio Template Recommendations on this book’s website at http://businessgrowth.ai.

Increasing Profit Margins with Marketing Mix Modeling

You could say that marketing mix modeling (MMM) involves, in its simplest form, using machine learning methods on sales-and-marketing data to predict what combination of product features and marketing methods will directly result in the most sales. MMM is one of the most powerful marketing data science use cases today, but it ain’t cheap.

Unless you hire a data scientist that specializes in MMM, or learn to do it yourself, the benefits of MMM are constrained mostly to larger companies with bigger marketing budgets that have sufficient allowances for costly vendor solutions. MMM’s high costs are caused by several factors, the largest of which is the amount of manual data science work that’s required. Several solution providers have built software solutions to help minimize the amount of work, but MMM is simply too complex to expect a SaaS solution to come in and do all the analytical work for you. My plan for this section is to give you a head start on figuring out how to do MMM in-house for your company.

Collecting data on the four Ps

When I talk about optimizing a marketing mix for more sales, what I’m talking about is identifying the exact mixture of product, place, promotion, and price — otherwise known in marketing as the four Ps — that are responsible for producing the most sales. The main thing you need to know about the marketing mix is that it should contain a mixture of features that are reflective of both your product and your marketing methods and that those features should be directly relevant to how well that product sells. Because the four Ps will comprise the variable in your machine learning models, it pays to look at each of them a little closer.

Inspecting important product features

The product here is, of course, the product you want to sell. In analyzing your product, you want to make sure it’s high-quality and easy to use and exceeds the buyer's expectations. If your product is no good, it doesn’t matter how you price or promote it — your buyers will be unhappy, sales will falter, and selling it will eventually tarnish your company’s brand.

Tip If you’re selling services, you can technically use a service package as the product here. In that case, you’ll probably want to extend to a seven Ps approach by adding variables that represent process, people, and physical evidence.

Playing with the price aspect

Price here is simply the price at which you sell your product. You can find lots of different pricing strategies out there, but the main thing to remember here is that you don’t want to be in a race to the bottom. The price should reflect the value that the product provides its buyer as well as how much supply exists to meet its demand. Rather than lower prices, look for ways to increase the value of the product and improve your marketing messaging to enhance the product’s positioning. Generally, as prices increase, sales volumes decrease — so your distribution numbers (as represented by your place variables) would decrease. Simple math can tell you, however, that you can still end up seeing an increase in sales revenues anyway. That’s one of the reasons it’s important to include both price and distribution in the marketing mix.

Remember From my experience, and the experience of the countless entrepreneurs who’ve mentored me, when you drop prices, you tend to attract more sales with fewer marketing requirements, but the buyers are generally higher-maintenance customers. Under-pricing your offer erodes the profitability of the product because you draw more customers who all tend to require more support services — which has to be paid for out of the operations budget. In this case, the distribution numbers (as represented by your place variables) can go up, but the price would go down and the overall profitability of this product for your company would suffer. Obviously, this is a situation you want to avoid.

Placing your product

With respect to the place variable, I’m talking about the place where the sale was made and the place where the product is distributed to its buyer. So, if you have a digital business, the place is nearly equivalent to the sales channel. But if you’re in retail and have a brick-and-mortar store in addition to an e-commerce store, place would designate the actual location where the sales and product distributions are made.

Promoting your offer

Promotion technically refers to how your company makes potential customers aware that the product is available for purchase. This often includes avenues such as organic marketing, paid ads, press releases, and how your brand appears in search engine results. Promotion is the vehicle by which these tactics are communicated to customers in order to produce an increase in sales.

Implementing marketing mix modeling

Simply put, machine learning for MMM involves deploying both linear and nonlinear regression methods on product, marketing, and sales time-series. As with any other machine learning method, you start by selecting appropriate variables. (For more on regression methods, see Chapter 4.)

You use a regression method, so you need to select response and explanatory variables. Good options for response variables in MMM include these:

  • Number of sales, represented as a count
  • Sales revenue, in dollar amount

When it comes to explanatory variables, you need to select features that truly represent how each of the four Ps behaves. (To keep this description concise, I’ve listed these features in order of decreasing impact on total sales.)

Here are four good ways to represent product in a marketing mix model:

  • Product quality in terms of the quality of components that comprise it, represented as a percentage (%) of a desirable attribute
  • Product quality in terms of durability, represented as a count of the product life span in days
  • Product quality in terms of conformance to manufacturing requirements, represented with a risk priority number (RPN) — a numerical score that represents the risk associated with a manufacturing process and the steps that comprise it. These processes should conform to manufacturing specifications. When they don’t, you get a higher RPN, and a greater risk of product defects.
  • Product newness, represented as a count of the number of days on the market

The place variable generally refers to distribution, so you’d want to explore representing it in any of the following ways:

  • Distribution volume, represented as the total number of units distributed and available for sale per location or per unit time — for example, per day, month, or quarter, depending on the time interval you’d like to use to represent the distribution volume.
  • Distribution in terms of the number of units purchased total, represented as a count
  • Distribution in terms of the number of units purchased per location, represented as a count per location

To represent price within the model, here are a few features to explore:

  • Unit cost, in dollar amount
  • Spending per customer, in dollar amount per sale
  • Product discount, in dollar amount or percentage (%)

And lastly, to represent promotion within the model, you need to home in on the activities your company engages in to increase product awareness and sales. Here are some features you might include:

  • Number of promotions
  • Cost per traditional promotion, in dollar amount, for TV ad spend, print ads, or outdoor campaigns, for example
  • Cost per digital promotion, in dollar amount, for Facebook or Instagram ads, Twitter ads, or paid search ads, for example

Unfortunately, if you want to take a single online course or read a comprehensive book on how to implement MMM, you won’t find any yet. The good news is that you can learn for free if you’re willing to use Google Search to track down a set of free online resources, like the one hosted over on the open-source R documentation website at https://rpubs.com/nihil0/mmm01.

Tip I’ve compiled an entire listing of MMM learning resources for both R and Python. You can access that listing over on the website for this book, at http://businessgrowth.ai.

Increasing profitability with MMM

Because the entire point of MMM is to quantify direct relationships between your marketing mix and sales for your business, not much room remains for conflating the profitability issue. Once you’ve identified statistically (and economically) significant relationships between your marketing mix and actual sales for the business, you can reliably predict what marketing mix will produce even more sales — and then adjust your company’s future marketing plans for improved ROI.