By Justin Sutton
The Curse of Dimensionality! The Law of Unintended Consequences! Simpson’s Paradox!
Despite what their mysterious names might suggest, these are not the next releases of Fortnite or an upcoming Harry Potter series, but rather important data science principles to be considered when conducting a data analysis and evaluating implications of those results.
When I jumped into my first data science project two years ago, I had equal knowledge on each of these subjects—which is to say, minimal at best. As the lowly “account guy” who spent most of his career focused on integrated marketing campaigns, I’d always associated data science with complex formulas and endless spreadsheets, but had little understanding of how powerful this discipline could be in solving branding and marketing problems for my clients.
Data science is clearly having something of a moment right now. Demand for data scientists is on the rise and data science methods are playing an integral part in evolving AI and Machine Learning technology. Yet despite its ‘buzziness’ in fields across corporate America, data science practices and the opportunities they offer still seem opaque and intangible to many marketers. By better understanding the immediate and actionable ways that data science can support their plans, brand leaders can quickly develop innovative new approaches to building their brand and driving growth for their business.
“A problem well-stated is a problem half-solved.” – Charles Kettering.
On the surface, there may not be an obvious connection between a breakthrough creative idea and, say, a complex, data-rich model. But the connective thread from powerful integrated marketing to effective data science work is clear (and a constant charge for our team), a foundational and deep understanding of the business problem. What are we setting out to prove or disprove? What connections can be made from these insights to outside variables? What are the implications of our results within the organization? What other factors might be a risk to these insights and observations?
With a commitment to understanding (or, in some cases, redefining) the business problem and its surrounding context, along with an understanding of some basic principles and methodologies, the data science world that once seemed cryptic and dull can quickly transform into a powerful tool for generating insight and informing new strategies. Here are a few examples:
Mining new insight from old research.
Many organizations have an abundance of proprietary research data that sits on the shelf and never gets picked back up—But by applying a relatively new methodology called Bayesian Belief Network modelling, marketers can not only define new consumer segments, but also establish what factors are most important to them, and therefore what messages to prioritize in future communication. This approach is especially beneficial for marketing leaders with limited budgets and even limited data. By deriving new insights from existing research, the value of the original investment increases exponentially and those limited budgets can be repurposed toward new executions, tests and opportunities to optimize.
Evaluating effectiveness and efficiency.
One critical task marketing leaders face is in evaluating marketing and media investments. Where am I over-investing? What channels or products are ‘punching above their weight?’ How might I shift dollars to become more efficient with my budget? Rather than solely measuring the impact of marketing and media investments with impressions or reach, a robust marketing mix analysis can evaluate return on those investments by their impact on the bottom line: revenue. Furthermore, by using a multi-year data set, interesting patterns can emerge and additional analysis can be sparked. For example, a halo and cannibalization study can highlight those investments that not only have a positive (or negative) impact on sales, but also a significant impact on other specific products or services in the portfolio.
Planning: From research studies to planning tools.
Many traditional research studies carry long lag times and limited nuance in terms of their implications on planning decisions. By combining custom technology with data science methods, marketers can evolve their research efforts from static studies that report on the past, to tools that futurecast and project immediate results based on real planning scenarios they might be considering. Fueled by predictive models, these tools help forecast future sales within a very small margin of error. This approach is suited perfectly for organizations with a substantial amount of data, across a variety of classification types and over multiple years. While building a custom product can prove more expensive, by adhering to Agile methods of product development, marketing leaders can ensure their budget is maximized and the tool they are getting is useful early on, and at every stage in the process.
Defining a competitive strategy.
By focusing solely on share leaders in their category or market, marketers can sometimes overlook opportunities for significant wins against smaller, but more vulnerable players. For example, by comparing a calculation of consumer loyalty data, marketers can identify competitors that appear most vulnerable in a market, relative to their brand, product or service. With a vulnerable competitor identified, the focus can then turn to understanding what that vulnerability is and how to leverage it through a sharp strategy and focused, hard-hitting creative. This approach has proven to be particularly effective for retail clients in highly-competitive, heavily-saturated markets, such as quick-serve restaurants or brick-and-mortar retailers. The methodology is predicated on the idea that it is a zero-sum game and new customers must come from the competition. And as the wise Sun Tzu declared, “in war, the way is to avoid what is strong, and strike what is weak”!
As new data sources continue to proliferate organizations across the country, brand leaders have a huge opportunity to embrace more advanced data science methods to uncover powerful insights that traditional regression models may not pick up. These insights, while sometimes nuanced, can identify new customers, inform new messages, elicit new plans and ultimately drive new revenue and brand growth.
Justin Sutton | Director, Client Engagement
Since joining VSA in 2014, Justin has led engagements for clients facing unique business challenges across a number of industries. From developing consumer marketing campaigns and managing field marketing programs for a national QSR brand, to building a state-of-the-art predictive analytics tool for another global QSR leader, Justin has managed teams across VSA disciplines to help solve business problems in a lasting and powerful way. Prior to joining VSA, Justin spent seven years managing engagements with Safeway, Brinker Restaurants, Hotels.com and Bloomin’ Brands (formerly OSI Restaurant Partners) at DDB, Energy BBDO and Y&R Chicago, respectively. Reach out to Justin at email@example.com.