Data Science for the Marketing and Advertising Industry

An enterprise or individual can have a mind-blowing product or service, but without the proper marketing and advertising, attracting customers will be extremely difficult. Marketing research isn’t a new concept. However, when compared to data gathering in the early 20th century, market researchers are now swamped with data. Between Twitter, Facebook, Google, and other social media channels, we can now collect an exponential amount of information about consumer preferences and other trends that may impact marketing efforts. As a matter of fact, digital advertising comprises a majority of the incoming cash flow for the likes of Twitter (86%), Facebook (99%), and Google (71%). Whether we are conscious of it or not, we are being “sold to” at almost every point throughout our daily lives.

Enter the marketing data scientist who is armed with an arsenal of predictive algorithms and the analytical acumen to parse through a massive number of data points. Granted, the title “marketing data scientist” hasn’t quite materialized in terms of employers explicitly stating that they are looking for this type of data scientist. At the current time, most data scientists wear many departmental hats, meaning they work inter-departmentally to build predictive models for a variety of business objectives: marketing, sales, human resources, risk mitigation, finance, robotics, cyber security, etc. But, data science is still evolving, and more distinct specializations are likely to emerge.

For example, natural language processing (also known as text analytics) is a particular discipline where the focus is on processing and analyzing human language. Computers are intrinsically mathematical. When compared to straightforward measurements (e.g., height, weight, test scores), human language has far more nuances and interdependencies that don’t fit neatly into a numerically measurable classification. Language conveys emotion and intention, and even human beings tend to struggle with accurately interpreting what the other person is trying to communicate. This is an important aspect of being a marketing data scientist as marketing and advertising are essentially a conveying of emotion and intention.

What Does a Marketing Data Scientist Do?

As a marketing data scientist, you’ll be using the usual tools: SQL, R, Python, and a preferred data visualization approach (usually this means Tableau). Depending on the size of the enterprise you’re working for, you may also have survey and questionnaire construction responsibilities — which is both an art and a science unto itself. You’ll also likely be tasked with analyzing consumer responses, sales call logs, customer service logs, in addition to external data sources such as social media mentions and interactions with your employer’s brand. But, that’s only the beginning.

Companies want to know what their competitors are and aren’t doing so they can target an expansion of their market share for a particular consumer segment. Thus, pulling together and analyzing data about competitors, including pricing, news, and consumer sentiment, will likely be an additional requirement. Data scientists will take these activities a step further and build predictive and prescriptive models to automate the process of deriving actionable insights and producing a recommended course of action.

A Data Science Use Case: The Marketing Funnel

While there are mixed views about the utility of the Marketing Funnel, it is a handy classification system for explaining data science use cases as applied to the world of marketing and advertising. Summarily, the Marketing Funnel is separated into three meta-categories: Lead Generation, Lead Nurturing, and Sales (which is at the bottom of the funnel). Each component is discussed in further detail below along with an example of how data science can be used to guide decision making in terms of improving the movement of customers towards a sales conversion.

Lead Generation: Websites and Blog Posts

Lead generation is about creating a broader consumer awareness that your product or service exists. One of the main channels for awareness creation is your website. For instance, if a website isn’t consumer friendly (and a data scientist will help to figure out precisely why this is the case), then this impacts consumer conversion rates. The website bounce rate will climb while the sales have a high likelihood of decreasing. A marketing data scientist can help improve the consumer experience via a website by performing A/B testing, creating a chatbot (NLP is particularly crucial for this aspect), and/ or building an advanced recommendation system. There are many approaches available to resolve this issue. It’s the data scientist’s job to apply their advanced statistical and mathematical capabilities and advise on which course of action to take, then build and maintain the right model to capture the right data that produces either a prescriptive or predictive evaluation.

Content is still king. Well written blogs and articles that inform customers without pushing a hard sell will increase website traffic along with brand awareness. Although machine learning and AI aren’t quite to the point where they can auto-generate content that mirrors human writing, predictive models can be used to gauge which type of content is likely prompting your brand messaging to go “viral.” Additional data collected from social media blasts — where your product or service is mentioned, or there are related images — can be ingested, cleaned, and incorporated into the predictive model. Are there particular customer segments that prefer “edutaining” blogs? If so, what are the components of these blog types, e.g., sentiment, emotion, intention, keywords, etc.? Is there a correlation between the time of day and the amount of traffic your blog posts generate? Are there specific problems that the blogs help customers resolve? If so, what are they and how can this information be leveraged to create new products or services that your competitors haven’t addressed as of yet?

Lead Nurturing: Emails and Free Trials

Data collection “walls” are pervasive. If you’ve ever tried to download a white paper or a free trial from a vendor, you’re probably familiar with having to send a certain amount of data in return for the “freebie,” e.g., email, name, telephone number (in some cases), etc. Not too soon after you give this info, you’ll most assuredly receive marketing emails. Why? Because this is still one of the most effective lead nurturing categories of digital marketing.

With regard to how data science applies to email marketing campaigns, a savvy data scientist can analyze the text in an email campaign and devise a predictive algorithm for keywords, images, and sentiment that receive higher response rates based on prior outcomes such as the potential customer completing a transaction. Drilling down a bit further, by gathering demographic data and, when available, social media interaction data, machine learning algorithms can be applied to intensify the email’s personalization (beyond just auto inserting your first name into a general, “Hi there! Buy from us!” template.

Consumers demand personalization. They don’t want to sense that they’re being viewed as just another sale. One consumer may find a humorous email much more compelling than a “Prices Slashed! 50% off until midnight tonight!” message. For example, not all women who are runners between the ages of 18 and 35 will respond to the same email messaging. Lead nurturing benefits from even more granularity. Where are they located? What are their other purchasing and social media habits? Are there pain points that are specific to the individual that you can address in an individualized email campaign? How can this be automated?

The exact same questions and processes can be applied to offering free trials except with a few more considerations. Inserting a “would you recommend” pop up when their free trial has expired that provides a rating (Net Promoter Score) and a selection of choices for why they would or wouldn’t recommend the product will provide additional diagnostic and predictive data.

Combined with the demographic, social media, and free trial outcome data (did they purchase a subscription or allow the free trial to lapse into oblivion?) this can increase actionable insights that lead to quickly identifying where this lead generation component has failed vs. succeeded. Is there a correlation between job title and the lead nurturing outcome? Are individuals or enterprises more likely to subscribe? If so, what are the cofactors, and are they impactful enough that you can make accurate predictions about which free trial — as well as consumer — features have a higher likelihood of leading to a sale?

Sales: Transaction and Customer Service Data

The completion of a sale does not end the “nurturing” aspect of the enterprise-consumer relationship. Certainly, one of the main business objectives is to attract new customers. However, if current customers are neglected, then you’re going to increase your customer acquisition costs as you try to bring in new customers to replace the loss in cash flow. At this level of the Marketing Funnel, transaction data can contribute to a more complete picture of who your customers are and identify their needs. Is there a strong association between the lead generation approach and the customer finalizing a transaction? Did they need technical support right away? If you’ve included a survey about the transaction process (recommended), do the responses have a robust correlation to product pricing, demographic indicators, the length of time and/or the number of emails, or a particular social media advertising blast? Answers to these questions spiral back around to building an accurate prediction model.

Even when all of your other factors in the Marketing Funnel are on point, if your customer service or technical support are lacking, this can spell disaster for customer retention. Customer service and technical calls or ticket logs can be data mined for keywords that specify the most frequent complaints (or compliments). Also, customer service and technical support responses can provide a wealth of information as to potential trigger words and statements that quickly, and satisfactorily, solve the customer’s issue. Are there customer service or technical support representatives who have a higher rate of resolution and customer satisfaction? Are there particular keywords or statements used by the customers that can signal how the call is likely to end? What are the most frequent words or statements used by customer service and technical support that guide the customer towards a continuing relationship? Most online support services — especially from major enterprises — now request that customers complete a short survey of their experience, even if it’s another “would you recommend?” pop up; this type of data is also essential for honing accurate analytics.

Conclusion

There is an adage in data science (and statistics): not all data is good data. The suggestions briefly examined in the above sections only touch the surface of the data science cycle and predictive model construction. For instance, it’s far easier to say “you should data mine for keywords and statements that may be associated with customer retention” than it is to actually go through the process of sifting through the data, cleaning it, and deciding how to approach this aspect of NLP. Trying to predict sentiments, intentions, and emotions based on an algorithm (or even a collection of them working in tandem), is an additional challenge that researchers are continuing to explore regarding applicability and accuracy.

Furthermore, marketing and advertising data are replete with unstructured and semi-structured data, not all of which has a direct measurement (e.g., categorical data vs. discrete and continuous). Each business problem is likely to present you with a different dataset which may derail any standardized data science processes you’ve created.

Thus, there is more to effective marketing and advertising than merely posting a blog and launching a website. A marketing data scientist can help to greatly improve the customer experience and ensure that you’re targeting the individual consumer with the right messaging at precisely the most opportune time. For those interested in entering this data science sector, coursework in marketing, digital marketing, NLP, consumer and behavioral psychology, computational linguistics, and UX/UI design are recommended. Data science is a perpetual learning process regardless of the specialty. So, aspiring data scientists should be prepared for ongoing learning as there is no one size fits all algorithm (or model) for predicting human behavior.