This article will closely examine three significant categories of data-driven marketing analytics: descriptive, predictive, and prescriptive.
What is data-driven analytics?
In the broadest sense, data-driven marketing means deploying quantitative analytic methods to analyze and extract actionable meaning from data effectively. This allows marketing analysts and stakeholders to develop informed marketing choices and deliverables. We live in a renaissance of data, where enormous amounts previously unavailable to analysts are now at our fingertips. This phenomenon shows no signs of stopping in the future, so data-driven marketing analytics is more crucial than ever. According to a Fortune Business Insights report published in July 2022, the global big data analytics market in 2021 reached $240.56 billion in 2021 and is expected to rise to $655.53 billion by 2029. Depending on the industry, a 2015 McKinsey brief found that big data and advanced analytics yielded anywhere from a one percent to 60 percent increase in revenue and cost improvement.
In numerous cases over the past decade, companies have centered their work on more broadly-targeted methods to derive novel information from big data and advanced analytics. Enterprising data scientists have helmed this work, and analytic vendors, are keen on diving into the data and creatively running analyses to unearth nuggets of knowledge and insights. Depending on the workflow stage(s) and data analytics requirements, there are three major categories of data-driven marketing analytics: descriptive analytics, predictive analytics, and prescriptive analytics. The categories encapsulate an array of analytic techniques, which encapsulate but are not limited to statistical modeling, machine learning, data mining, and AI.
What is descriptive analytics?
Descriptive analytics is a category of data-driven analytics that engages in interpretative and analytic work on historical data. Descriptive analytics’ primary purpose is to locate past patterns and trends retroactively. This is extremely useful for businesses because this analysis helps uncover future opportunities or potential issues. A 2020 research article in Nature Machine Intelligence analyzed combining procedural content generation (PCG) from the gaming industry and machine learning (ML) methods. Using PCG, an older technology that algorithmically generates a game’s content, along with ML, allows researchers and analysts to randomize problem parameters during analysis. This version of descriptive analytics analyzes past issues with machine learning models being too specific and will enable PCG to be married to ML to open up new opportunities.
Another paper published in 2019 by the Journal of Marketing Research proposed an analytic methodology to examine co-occurrences of products in consumer carts. They focused on contemporary advances in ML and natural language processing (NLP). Their process is already highly adaptable and scalable for the retail industry because data from online checkout and cart systems are readily available. When paired with an analysis of cross-product co-occurrence, complementarity, and intra-category substitution, this methodology is a promising approach to descriptive analytics by analyzing past trends of co-occurrent products.
What is predictive analytics?
Predictive analytics is a category of data-driven analytics that analyzes current and/or historical data using statistical methods. Predictive analytics’ primary purpose is, as the name suggests, to analyze the likelihood of an event(s) happening in the future. By focusing on the nature of the event and the potentially time-sensitive occurrence of the event, predictive analytics is helpful for several prediction questions in marketing analytics. These problems include credit scoring, customer churn and turnover, customer acquisition, lifetime value, advertising click rates, purchase frequency, and product recommendations. Predictive analytics can assist in optimizing and improving many analytic areas, including efficiency, risk management and reduction, and customer service.
For example, a 2021 article published in ACM Computing Surveys developed a meta-analysis of current methodologies for predicting user responses in the online advertising industry. By creating a taxonomy of user-response prediction methods, focusing on ML methods, their research can be used to grapple with predictive analytics in other contexts.
One of the most promising methods for predictive analytics in marketing is deep learning (DL). DL is a subfield of machine learning that focuses on algorithms that take their structure and operations from neural networks in the brain. These artificial neural networks deploy multiple processing layers to derive increasingly more complex features from large amounts of data. This gives DL a reasonably strong prediction accuracy compared to other analytic methods, as it is algorithmically designed to model how the human brain processes certain types of knowledge. Like human neural networks, DL can concurrently analyze multiple data types, such as images, audio, location, clickstream data, app data, and social network data. This ability to process disparate data separates it from other analytic methods. Still, DL is also distinct from the human neural networks based on the sheer scale of data it can process, which is well beyond what the human brain can handle.
However, one of the major complaints about current predictive analytics methods is that they are too opaque, with little possibility of discerning what happens during processing after the initial input. Therefore, it’s essential to always keep in mind the potential limitations of predictive analytics, mainly when relying on algorithms to streamline predictions. For example, a 2022 article in the Journal of Business Research combined a systematic literature review and interviews with ML specialists to identify algorithmic bias in ML-based marketing analytics. The article highlights an array of algorithmic biases that can pervade ML-based marketing analytics, including weak algorithm designs, unrepresentative datasets, ineffective models, or historical human tendencies. The latter problem is especially problematic, as it can lead predictive analytics to replicate stereotypes and biases against marginalized consumers. For example, a 2021 article found that algorithmic models designed to guide clinical decision-making during the COVID-19 pandemic were rushed and biased against minority populations. A 2020 article found that Uber and Lyft’s algorithms set higher prices for destinations in predominantly African-American people. These algorithmic biases weren’t limited to race- a 2019 article published in Management Science found that Facebook’s targeted ad algorithms showed fewer ads advertising STEM jobs to women than men, despite initially intending to be gender-neutral in its delivery. The examples above underscore the importance of balancing predictive accuracy with transparency when working with predictive analytic algorithms for marketing.
What is prescriptive analytics?
Prescriptive analytics is a category of data-driven analytics that applies testing to derive which outcome will yield the most promising marketing results in a given scenario. Prescriptive analytics’ primary purpose is determining why a potential product will occur or making the best decisions in a particular method. Unlike predictive analytics, prescriptive analytics focuses on the actions needed to attain predicted outcomes and the relational impacts of each step. In that sense, prescriptive analytics encapsulates predictive analytics.
Predictive analytics commonly utilizes randomized experimental techniques, such as A/B testing, to determine the best marketing outcomes, such as price point, ad copy, strategy, demographics, etc. Randomized experiments work by randomly distributing the experimental variables across different treatments and identifying the treatment that produces the best outcome. For example, a randomized experiment on ad copy would randomly allocate specific messaging across other conditions, such as consumer age ranges or platforms, and determine which states produced the most effective outcome for the ad copy. Randomization in data-driven analytics allows analysts to reduce bias by equalizing other variables that weren’t directly considered in the experiment’s design. However, randomized experimentation is not always practical for most businesses, as it can be costly and time-consuming. As a result, much of prescriptive analytics is concerned with yielding dynamic choices using outcomes-based predictive analytics and quasi-experimental and observational data.
In a 2021 article published in Marketing Science, researchers focus on “moment marketing,” which allows analysts to synchronize online advertising with offline events in real time. The researchers conducted a causal estimation study by examining significant shifts in (offline) TV advertising costs for a major U.S. fast-food brand over four months in 2017. Using statistical methodologies, the researchers found that advertising focusing on moment-based TV adverts could be used to optimize online sponsored search advertising. This synchronization would produce cross-channel advertising effects. For example, TV advertising can alter the content and platform of online searches, such as how and where consumers search, shortly after a TV ad airs. As a result, the average consumer conducting online investigations reacts differently to search results found “at the moment.” An example of this cross-channel advertising is a Superbowl TV ad spurring viewers to look up a brand or a product in a particular context shortly after the ad airs on television. “Moment marketing” is a generative space for prescriptive analytics, as it allows analysts to understand better how marketing decisions interact and relate to one another.
Looking Ahead
In the era of big data, companies are increasingly leveraging data-driven analytics to elucidate information that can facilitate business and marketing strategy, deliver better products and services, and target online experiences to consumers. The three major data-driven analytic approaches—descriptive analytics, predictive analytics, and prescriptive analytics— are valuable for businesses working with data. These approaches all complement and ideally occur in tandem with each other.
However, even though analytic models frequently outperform humans, context, reasoning, and final decision-making are ultimately in the hands of analysts. Data-driven analytical approaches are essential tools and methodologies for analysts to rely on. Still, given algorithms’ relative opacity, analysts must continually check their methods for potential biases.
Finally, recent consumer concerns about privacy in tandem with current laws regulating data privacy have spurred companies to examine how they can make their data more transparent and open to consumers. The future of data-driven analytics in marketing will have to contend with this new expectation of data transparency.