At any given time, a myriad of forces are at work influencing our consumer perceptions, affecting the attitudes and preferences that ultimately determine our purchase outcomes and future consumption patterns. Factors that affect our purchase decisions vary widely from one industry to the next, because firms differ in product offerings, regulatory considerations, and competitive scenarios.
The consumer decision-making process
The process a consumer goes through during the decision-making process is remarkably complex. Consumer scientists have identified five distinct steps involved in purchase behavior (see Resources). The first stage is one of need recognition, where a potential buyer recognizes an imbalance between actual and preferred states. This recognition can result from both internal and external stimuli, where a consumer might notice the gap on his or her own or through some external force such as marketing, advertising, or simply viewing others’ consumption habits and desiring to have the same.
After the initial product need is realized, that consumer embarks on an information search. Depending on the consumer’s level of interest in the product and personal attitude toward risk, he or she will spend either a short or long amount of time gathering and assembling the information needed to make an informed purchase decision. Next, deliberation occurs, where an identification and evaluation of alternatives is performed. During this third stage, a consumer analyzes product attributes, determines thresholds, and ranks product attributes by personal importance. It is in this third stage that price considerations and availability of alternatives are stressed. Finally a purchase occurs, then postpurchase and the process of cognitive dissonance, when a consumer mulls over his or her actions and wonders if he or she made a good decision, whether he or she got the right product as well as good value, and mentally adjusts attitudes to bring this into balance.
Purchase drivers: Channels
A lot of research has been devoted to understanding the drivers of purchase in a retail environment. It is important to note that these drivers also differ based on channel of sale. In the case of purchases made in retail stores (see Resources), purchases involve face-to face interaction with service personnel. In-store aspects like store display and presentation, ambiance, customer treatment, store layout, discounting, and promotions all enhance or detract from the non-virtual shopping experience, whereas e-commerce site experience factors include site design, site performance and reliability, security, and customer service. Regardless of channel, pricing is a key factor related to consumer choice.
Some studies have found that consumers’ price sensitivity tends to be less evident when shopping online and through mobile channels. One study (see Resources) found the primary impact on consumer choice to be channel (store, catalog, Internet) and price of the product, noting a distinct segment of customers preferring to transact through the Internet channel.
Retail represents one of the most complex industries in terms of the number of products and channels involved. A supermarket may offer tens of thousands of products, with roughly 50 percent of them perishable. A home improvement store also offers an infinite and ever-changing supply of options. Consider limited-product companies such as electric power and insurance companies. These firms traditionally offer only a limited number of products. Utilities may offer a selection of billing, bundling, metering, self-management of usage, rate structure, and prepaid options; however, consumer choice is essentially limited to a single product: electricity, water, or natural gas. Although insurance companies offer a comparatively larger selection of products, including auto, health, life, property, renters, pet, and financial products, that list is still finite compared to the complex array of product categories and alternatives that retail companies offer prospective customers.
Identifying the determinants of demand for firms with a limited number of products is therefore also relatively less complex, whereas retail represents an industry in which channel and consumer data collection is extremely rich and frequent. The retail industry is in a constant state of flux, with increasing competition and experiencing a seismic technological shift toward mobile as a channel that provides real-time alternative product pricing, feature-comparison capabilities, and one-step purchasing. Retailers that employ predictive analytics to better understand their customer and prospect behavior at a micro-segment level will be better situated to make faster data-driven decisions and have a deeper grasp on customer demand and preferences.
Evaluating product performance
Given all this complexity across a multichannel series of product categories, brand selection, packaging, inventory, pricing, discounting, and display options, how do retailers effectively evaluate performance at the product category level? At large retailers, this is done through a retail and supply chain management approach called category management, where the ranges of products are assembled into broad-level groups categorized by similarities. Each group is then run as a separate business unit. Each category manager is responsible for the direction and performance of his or her product category business unit, and each develops his or her own profitability targets and business strategy (see Resources for more information about category management).
Traditional metrics around sales per selling space or product category dominate the industry, and these key performance indicators (KPIs) are tracked in real time in enterprise reporting and inventory management systems and viewed by managers through reporting dashboards. Another important evaluation approach originating from the marketing discipline is to evaluate the customer lifetime value (LTV) of the customer base for the product category and understand how that audience interacts with the brand from a cross-product category perspective.
Let’s assume, for our purposes, that traditional aggregate KPIs like sales per square foot and sales per employee metrics are already tracked and stored in the database to support additional advanced analytics modeling initiatives. Consider what information about consumers at the individual level might be useful in terms of understanding the overall potential of an existing customer base for a product category, and consider how this information could be used to increase sales and manage inventory. The variable list at the consumer and household level available for modeling is virtually infinite—and overwhelming—from a data perspective. Recent technological strides have addressed traditional Structured Query Language (SQL) databases and commercially available statistical tools and now new NoSQL and cloud-based systems are emerging as common components of the go-to scalable data warehouse architecture revamps. Databases such as Netezza, Cassandra, and Pentaho combine with systems like Apache Hadoop and MapReduce, which allow access to auto-classify functionality and filter by sample the petabytes of consumer data in an interactive way not possible with legacy systems. The proliferation of social, mobile, web, video, and picture data associated with consumers’ interaction with the brand represents some of the most vital and untapped data in the retail industry. This information provides savvy retailers with an opportunity to gain a competitive advantage as they find new ways to derive individual and product-level insights by combining new data and approaches with industry-standard transaction, bar code-level, and survey-based data.
Retail performance is ultimately determined by sales numbers, and pricing strategy is an integral component of the health of a business unit and the enterprise overall. Gathering information about the consumer in terms of individual and household preferences, volume, interval of usage, bundling of purchases, complementary and alternative brands and products, and price sensitivity at the product category level allows insight from a bottom-up perspective. This insight will support micro-segmentation of customers for profile development that drives pricing and product design strategy and provides data to support bottom-up level sales forecasts and inventory supply forecasts. One of the most significant consumer-level variables as an input to both the segmentation and the sales forecasting models is price sensitivity. The remainder of this article focuses on the measurement and usage of this key consumer price-sensitivity input into the model.
Price sensitivity is the marketing term for the product- and consumer-level metrics that economists refer to as price elasticity of demand. Basic economics (see Resources) teaches us that all consumers are not created equal. One of the first concepts introduced is consumer preference, where consumers must choose bundles of goods, and the allocation subject to their budget constraint is represented by indifference curves. The rate by which they substitute some of one good to gain more of the second good is called the marginal rate of substitution.
Price elasticity of demand
Another concept introduced early on is price elasticity of demand (see Resources), a sensitivity measure representing the impact of a price change on quantity demanded. If quantity and price are represented by Q and P, price elasticity of demand is represented by the following expression:
Ed = (∆Q / Q) ч (∆P / P)
This indicates the percentage change in quantity that results from a one-percent increase in the price of that good. This price elasticity is affected by three main factors:
- First is the availability of substitutes and, generally speaking, the more alternatives consumers have to a product that has become more expensive, the more likely they are to shift to a similar product; in this case, demand is elastic.
- A second consideration is the time frame consumers have to adjust to the price change. Over time, more information on other alternatives results in demand becoming more elastic over time.
- Finally, there’s a question of how much of a household budget is allocated for this good: The larger the budget share, the more elastic the demand.
Two-stage estimation model
What is proposed here is a two-stage estimation model. In the first series of predictive models, price sensitivity measures at the individual and household level are determined. In the second estimation step, these price sensitivity inputs generated in the first stage of modeling become inputs into the second predictive product-demand estimation model.
The approach presented here involves an initial estimation of individual levels of price sensitivity at the product category level as a base. It is recommended that moving average and lag variables that capture sensitivity changes also be derived from a time series of purchase data and other indicators of developing or diminishing price sensitivity that might be stemming from the onset or easement of budgetary constraints (for example, job loss, student off to college, divorce, illness, other negative macroeconomic indicators affecting consumer confidence on the negative side; or salary increase, child finished with college, marriage, and other positive economic indicators). Other variables to facilitate a cross-price elasticity of demand representation (how price sensitivity for a product changes in relation to a price change in either a complementary or substitute product) should be collected.
Measuring price sensitivity
Price sensitivity, like customer loyalty, can be measured in a variety of ways and from a great deal of different consumer data. One measure used at the outside of new product launch pricing strategy is called the van Westendorp Price Sensitivity Analysis (PSA; see Resources), an approach that uses consumer survey feedback data to determine a range of acceptable and optimal pricing. This data is useful in determining whether a price-skimming (see Resources) strategy is feasible.
Survey-level data that is available at the household level may be added to the model. Keep in mind this will be a sample representing only a small percentage of households; if this data is added to the sensitivity estimation model, this data issue can be managed in two ways. First, a look-alike model can be contracted to score the customer database with expected responses. The second approach is to overlay the survey data on top of either a micro-segmentation or propensity model and assume that all others within a segment or a probability threshold have similar responses. The Resources section offers a link providing an example where Taco Bell was able to identify price sensitivity through surveys.
Social media provides a rich and untapped source of semantic data that can be valuable in terms of price-sensitivity prediction. It is necessary to collect customer social media IDs to do this modeling at a household level, and increasingly more firms are collecting Twitter and Facebook IDs that can be merged with customer transaction and survey data. If this linkage variable is not available, specific brand and product category price sensitivity can be derived from anonymous consumer comments and aggregated for other sensitivity analysis validation purposes. Briefly, consumers who use keywords such as expensive, cheap, a steal, coupon, and so forth tend to be price conscious, especially when comment share is at a higher-than-average rate for the brand in relation to quality, packaging, service, and other brand category comments.
After a company has identified household levels of price sensitivity at the category level, this data can be used and updated to improve strategy and processes throughout the enterprise. One such specific example is pricing strategy and specific households to target for new product offerings that will have a higher propensity to pay a premium during the early-introduction phase of the product launch. In a case where promotions or price decreases are planned, predictive models can predict the new demand scenarios, and these projections directly relate to inventory control to meet increased or decreased expected demand. Additionally, the price-sensitivity variables can be useful in terms of messaging at a segment level.
Currently, integration continues to happen in terms of pulling together all of the customers’ touch points and transactions; incorporating these with how brands are interacting with customers across social channels like Twitter and Facebook and how consumers are influencing one another’s purchase decisions. The retail industry promises to become only more complex and full of rich new data sources to drive competitive advantage to those brands that put predictive analytics into practice to leverage all of it.