We see a paradox in two important analytics trends. The most recent results from The CMO Survey conducted by Duke University’s Fuqua School of Business and sponsored by Deloitte LLP and the American Marketing Association reports that the percentage of marketing budgets companies plan to allocate to analytics over the next three years will increase from 5.8% to 17.3%—a whopping 198% increase. These increases are expected despite the fact that top marketers report that the effect of analytics on company-wide performance remains modest, with an average performance score of 4.1 on a seven-point scale, where 1=not at all effective and 7=highly effective. More importantly, this performance impact has shown little increase over the last five years, when it was rated 3.8 on the same scale.
How can it be that firms have not seen any increase in how analytics contribute to company performance, but are nonetheless planning to increase spending so dramatically? Based on our work with member companies at the Marketing Science Institute, two competing forces explain this discrepancy—the data used in analytics and the analyst talent producing it. We discuss how each force has inhibited organizations from realizing the full potential of marketing analytics and offer specific prescriptions to better align analytics outcomes with increased spending.
The Data Challenge
Data are becoming ubiquitous, so at first blush it would appear that analytics should be able to deliver on its promise of value creation. However, data grows on its own terms, and this growth is often driven by IT investments, rather than by coherent marketing goals. As a result, data libraries often look like the proverbial cluttered closet, where it is hard to separate the insights from the junk.Insight Center
In most companies, data is not integrated. Data collected by different systems is disjointed, lacking variables to match the data, and using different coding schemes. For example, data from mobile devices and data from PCs might indicate similar browsing paths, but if the consumer data and the data on pages browsed cannot be matched, it is hard to determine browsing behavior. That’s why understanding how data will ultimately be integrated and measured should be considered prior to collecting the data, precisely because it will lower the cost of matching.
What’s more, most companies have huge amounts of data, making it hard to process in a timely manner. Merging data across a vast number of customers and interactions involves “translating” code, systems, and dictionaries. Once cohered, vast amounts of information can overwhelm processing power and algorithms. Many approaches exist to scale analytics, but collecting data that cannot be analyzed is inefficient.
An irony of having too much data is that you often have too little information. The more data and fields collected, the less they overlap, creating “holes” in the data. For example, two customers with the same level of transactions could have very different shares of wallet. While one represents a selling opportunity, the other might offer little potential gain. Data should be designed with an eye towards imputation — so the holes in the data can filled as needed to drive strategy.
Perhaps worst of all, data is often not causal. For example, while it is true that search advertising can be correlated with purchase because customers are in a motivated state to buy, it does not follow that ads caused sales. Even if the firm did not advertise, consumers are motivated to buy, so how does one know whether the ads were effective? Worse, as data grows, these problems compound. Without the right analytic approach, no amount of investment will translate to insights.
Companies should do two things to harness the power of analytics in their marketing functions. First, rather than create data and then decide what to do with it, firms should decide what to do first, and then which data they need to do it. This means better integrating marketing and IT, and developing systems around the information needs of the senior management team instead of creating a culture of “capture data and pray.”
Second, companies should create an integrated 360-degree view of the customer that considers every customer behavior from the time the alarm rings in the morning until they go to bed in the evening. Every potential engagement point, for both communication and purchase, should be captured. Only then can firms completely understand their customers via analytics, and develop customized experiences to delight them. The CMO Survey we referenced above shows that firms’ performance on this type of integration has not improved over the last five years, challenging companies’ ability to answer the most important questions about their customers.
The Data Analyst Challenge
The CMO Survey also found that only 1.9% of marketing leaders reported that their companies have the right talent to leverage marketing analytics. Good data analysts, like good data, are hard to find. Sadly, the overall rating on a seven-point scale, where 1 is “does not have the right talent” and 7 is “has the right talent,” has not changed between the first time the question was asked in 2013 (Mean 3.4, SD =1.7) and 2017 (Mean 3.7, SD =1.7).
The gap between the promise and the reality of analytics points to a disconnect that needs resolution. Companies need to better align their data strategy and data analyst talent to realize the potential that analytics can bring to marketing managers. In the absence of talent, even great data can lie fallow and prevent a firm from harnessing the full potential of the data. What are some of the characteristics that companies should look for in good data scientists? They should:
Clearly define the business problem. Managers who rely on data scientists to know what might be possible to do with the data often find great value in simply having that person help define the problem. For example, a marketer coming to a data analyst asking questions about driving conversions might not realize that there’s also data at the top of the purchase funnel that might be even more germane to driving long-term sales. Rather than taking requests as they are stated, data analysts should take requests as they should be asked, integrating advice tightly with the needs of the company. For example, a request to assess how marketing promotions affect sales should also account for the effect of promotions on brand equity.
Understand how algorithms and data map to business problems. Companies will see more effective data analytics if teams are clear on firm objectives, informed of the strategy, sensitive to organizational structure, and exposed to customers. To enable this understanding, data analysts should spend physical time outside of data analytics, perhaps visiting customers to give them an understanding of market requirements, attending market planning meetings to better appreciate the company’s goals, and helping to ensure data (IT), data analytics, and marketing are all aligned.
Understand the company’s goals. Data analytics is beset by multiple requests, like a waiter serving too many customers. A clear recognition of a firm’s goals enables data analysts to prioritize projects and allocate time to those that are the most important (those that have the highest marginal value to a firm). Requests should be centralized, and then prioritized by a) whether the findings have the potential to change the way things are done and b) the economic consequences of such changes. Several companies develop standardized forms to ensure requests are assessed on an equal footing. An attendant benefit of this process is that it mitigates the potential for opportunistic research clients to approach analysts asking them to conduct a study to support a preconceived strategy for political reasons, instead of deciding between strategies that are in the best interests of the firm.
Communicate insights, not facts. Communication theory tells us that the transmitter and receiver of information must share a common domain of knowledge for information to be transmitted. This means analysts need to understand what the firm’s managers can understand. Small font sizes, complex figures and equations, the use of jargon, and an emphasis on the modeling process instead of insights and explanations are common errors when presenting analyses. Why should one use a complicated model to present information when a simple infographic would suffice? Presentations should be organized around insights, rather than analytic approaches. This is another reason it is critical for analysts to connect externally with customers and internally with the managers using their work. Plus, instead of reporting a “parameter estimate,” an analyst should communicate how results point to tangible strategic actions. This requires analysts to structure their analysis in a decision framework that helps managers assess best and worst case scenarios.
Develop an instinct for mapping the variation in the data to the business questions. That means two things. First, analysts need a comprehensive understanding of all the relevant drivers (e.g., marketing and environmental factors) and outcomes (e.g., purchase funnel metrics). For example, to ascertain the effect of advertising on sales, one would need to recognize that concurrent changes in product design can affect sales, lest one misattribute the effect of product changes to the advertising that announces them. Second, analysts must have a means to ensure that drivers lead to outcomes instead of outcomes leading to drivers. Once again, this requires the analyst to understand the nature of the markets being analyzed. Regarding the latter, no complicated model that purports to control for missing information can ever compensate fully for lack of causal variation. Likes drive sales and sales drive likes. However, disentangling the two means having some factor that can independently manipulate one and not the other.
Identify the best tool for the problem. On the analytics side, it goes without saying that years of training and practice are necessary. One cannot play an instrument without learning it, and the same is true for analysts. Most important is knowing which tool, of the many available, is best for which problem. At a very granular level, experimental methods are especially adept at assessing causality; supervised machine learning excels at prediction where non-supervised machine learning can decompose non-numerical stimuli into tags or attributes for further analysis. Economics and psychology afford deep insights into the nature of consumer behavior, and statistics can help us excel at inference. A strong understanding of marketing grounds all of these tools and disciplines in the business context necessary to produce effective advice.
Span skill boundaries. Some marketing analysts excel at math and coding, and some excel at framing issues, developing explanations, and connecting to business implications. A far smaller set excel at both. Companies either need to wrap these variegated skills into one person through training and accumulating different types of experiences, or, more likely, assemble a team that is sufficiently facile with the techniques that they can interact productively, ensuring that there is some mechanism to match the approach (and the analyst) to the problem. This match requires senior talent, with the breadth of perspective to align analytical resources and business problems.
In light of the exponential growth in customer, competitor, and marketplace information, companies face an unprecedented opportunity to delight their customers by delivering the right products and services to the right people at the right time and the right format, location, devices, and channels. Realizing that potential, however, requires a proactive and strategic approach to marketing analytics. Companies need to invest in the right mix of data, systems, and people to realize these gains.