Orders are the financial core of ecommerce and represent, at a macro level, the interaction between a buyer and an ecommerce site when value, usually money, is exchanged in a transaction. Products are bought within orders. Ecommerce order data, thus, represents the attributes and properties of the order itself (such as the total sales price, discounts used, and margin), the individual products in the order (such as their sales prices and margin), and the customer behavior that occurs before and after the order. At a technical level, ecommerce orders are transactions defined as a logical operation on data and are said to have atomicity, consistency, isolation, and durability (ACID) within databases (Wikipedia 2015). For the purpose of this chapter and in general within analytics, the transactions that we want to analyze are ecommerce orders for products (or services) that are captured, stored, and processed in ecommerce and analytical systems. It is useful for an analyst to have an understanding of technical details related to ecommerce transactional processing systems and related systems for order analysis. But business stakeholders don’t care very much about the database or the atomic-level detail captured about customer orders—that is, until the database doesn’t function as expected, doesn’t contain desired data, or can’t be used to provide data for answering a business question. In these cases, business stakeholders will want answers about why they can’t get the data they need. But for the most part business stakeholders will be more concerned about the financial impact and monetary measures of orders and the products. The analyst must help stakeholders ensure data is collected about products and orders via a data pipeline and governance/stewardship that results in clean and curated data that can be used for analysis. Analysts will explore and develop an understanding and perspective of the facts and dimensions about orders and products. They will apply this understanding to answering business questions, making recommendations, and describing insights derived from business questions about ecommerce orders and products.
The analysis of products and orders requires the analyst to think of the order as a transactional record in which detailed data is stored about not only the products in the order but also about the order itself. For example, an order has revenue generated in total, and that revenue can be broken down by product. The marketing channel that contributed to the transaction can be attributed. Promotions used, tax applied, and shipping charged can be analyzed within the boundaries of ecommerce order analysis. The customer can be tied back to the order, which may contain information about the gender, size, style, brand, and season of the product purchased. This rich transactional data set of order and product data and metadata can be mined and analyzed. By digging and mining the order-level detail in relation to the customer, marketing and advertising, and conversion, you can find and understand the data to answer business questions. You may want to look for trends, patterns, outliers, and anomalies in order data. Keep in mind that ecommerce order and product analytics also include data that you may not expect and that isn’t commonly considered for capture and usage. This data could include information about the privacy policy applied to the order, such as details or its version. Flags indicate whether the order was an outlier in some way, perhaps by the revenue amount, number of items ordered, possible fraudulent IP address, unusual device type, and more. The data model corresponding to an ecommerce order contains the data to analyze for different business purposes:
• Orders are important to analyze for all business stakeholders. Depending on the goals, questions, and seniority of the stakeholders, an analyst has a wealth of information to provide for answering each question.
• Senior leaders want to review summary information at the aggregate about orders and the products sold within them, such as changes in the number, pacing, revenue, trends in the purchase categories, and the impact of promotions and discounts on the revenue and profit.
• Marketing staff want to understand orders overall and the performance of products, categories, and brands as they relate to different marketing channels, campaigns, and promotions; high-value customers and cohorts; device types; and more.
• Technical staff want to understand the sufficiency of the data captured within existing systems and current data models to help define the road map and how the data is being used as it relates to scaling technologies and ensuring technical alignment in terms of resources and systems against business requirements.
• Buyer and merchandising staff want to understand analysis about the specific products purchased, the revenue and margins of specific products, brands and categories, as well as the impact of discounting and promotions on price and profit. Additionally, the merchandiser will want to understand (and test) the impact of visual merchandising in the ecommerce experience on the volume, size, and financial return on orders and products (Cutroni 2013).
An ecommerce order, from a business analytics perspective, represents a transaction for one or more product (or service) items. As with all data in analytics, creating, socializing, and approving definitions for order data requires data stewardship and governance. Order data will be represented by a data model that will have metadata that may not be properties of the transaction, but rather properties of the ecommerce site—such as total transaction time or measures of server load or latency, which can be helpful to include in order analysis.
At the summary level, an order could have these data elements in a conceptual data model:
• Order ID is the unique identifier associated with the order event however it is defined at your company.
• Revenue is the amount of money spent on buying the items in the order.
• Shipping is the cost of shipping the items bought in the order.
• Tax is the taxation on the cost of the items bought in the order.
• Promotion is the name of the promotional discount or code, if any, applied to the order.
• Affiliate is the source of the order if different from the core ecommerce experience.
• Channel is the marketing source, if any, that resulted in the order as identified from the attribution model applied to orders.
• Device is the type of machine used when the order occurred, such as a desktop, tablet, or mobile device.
• Customer ID is the unique identifier for the customer who bought the items in the order.
• Time is the time stamp at the time when the order occurred.
• Geography is the physical location of the customer who purchased the items—as semantically correct for your business. This location could be self-entered, derived from the shipping information, or assigned by resolving the IP address to a geography.
• Technical information is the IP address, various device or hardware IDs, and information available about the order in HTTP headers or via other data collection.
Related to and necessary for detailed order analysis are the individual products or items that are part of each transaction (notice how the transaction ID is the key to join the transactional data with the product data, and transaction ID can be joined with a customer ID):
• Transaction ID is the unique identifier associated with the order however it is defined at your company.
• Product ID is the unique identifier associated with a unique product in the shopping cart.
• Product name is the human-readable name for the product purchased.
• SKU is the “stock keeping unit” represented by a unique alphanumeric code for the product.
• Price is the price paid for the product excluding tax and shipping.
• Quantity is the number of units of the product purchased.
• Category is the qualitative descriptor that separates one item from another, often derived from a taxonomy.
• Brand is the name of the manufacturer or marketing label for the product.
• Style is a qualitative descriptor that indicates product distinctness in terms of fashion.
• Size summarizes the physical dimensions of the product along a set of known labels, such as small, medium, and large.
• Season describes the time of the year for companies selling goods specific to calendar activities and weather patterns.
From this order-level data and the product-level data, many different variables are available for analysis and visualization. An analyst can aggregate order and specific product revenue over time. The most popular products and SKUs can be derived. Assumptions and hypotheses about the customer base can be drawn based on the brand preference or style preference. Information on shipping, tax, and discounts, as well as marketing channel and affiliate data, can help with planning and operations. Models can be built from order and product variables to predict revenue and create recommendations. The best data model and facts and dimensions to use for orders and products will be custom to your business but will likely consist of at least some of the data we just reviewed.
You might be wondering what type of order data the analyst should analyze. In my experience, there is overall data you want to report and understand about all the orders on your site and data about the individual orders. Thus, measuring, tracking, reporting, and analyzing the following overall, aggregate data about all orders can be helpful: the number of total orders, units sold, total ordered revenue, total discounted revenue, average discount percentage, average order value, and average gross margin. Also, the analyst should analyze data about the specific, individual orders, including the order number, the order date, the products in the order, discounts used, discount percentage, the subtotal, the shipping cost, the tax cost, the total price, the gross margin, and net profit.
Key performance indicators (KPIs) are metrics and ratios that are measured and tracked over time against business goals and benchmarks in order to understand performance. KPIs provide a numeric indication of the positive and negative trends in important business data. They are used by ecommerce workers to make decisions about what has worked or not worked so they can make the right decisions about how to do their job. KPIs help to drive business planning, set goals, and when analyzed can help the company derive understanding of the business in a way that informs decision-making. Many metrics and KPIs can be created for understanding what’s happening with your ecommerce transactions and products. What follow are several KPIs that can be helpful for understanding orders. By analyzing the trends and movements in these KPIs over time and across periods, and by drilling-down and up and segmenting KPIs by dimensions, you may find insights. Here are some helpful KPIs for ecommerce orders:
• Total Revenue Versus Total Profit is a KPI that shows the top line and bottom line of the order. It can be drilled up or drilled down, segmented to infinity.
• Margin usually represents the contribution margin, gross margin, or net profit margin—all powerful KPIs to work with at the aggregate and in the detail of transactions.
• Median and Average Revenue per Order are KPIs that result from application of business statistics to revenue data. These KPIs can become benchmarks to calculate upper and lower control limits to find outliers.
• Orders per Customer is a KPI that when assessed across the customer universe enables segmentation to find the potentially highest-value customers based on their volume of orders.
• Visits to Order for an ecommerce digital experience is a KPI that begins to elucidate the customer journey. By finding out what the customer did in prior transactions or what influenced the prior visits, the ecommerce company can better respond to the customer and use the information to help other customers buy more quickly.
• Days to Order is like Visits to Order but instead denotes the number of calendar days between the first measurable and identifiable behavior by the customer and the date of the order. Again like Visits to Order, this KPI suggests that a customer journey occurs over a time period, and that some customers convert and place orders sooner than other customers. Thus, the period of time between when a customer first visits and places the first order can be determined—and effort can be placed in marketing and persuading customers to convert sooner.
• Revenue by Category is a helpful KPI for buyers, merchandisers, and planners because the distribution of revenue when mapped against business goals makes performance clear.
• Orders and Products Bought by Marketing Channel is a KPI that requires attribution to be made about the exposure of the customer to a marketing channel before the order.
• Events/Pages/Steps until Order are KPIs that can be used to denote a flow of events before an order. In ecommerce, the sequence of pages in the shopping cart flow leading to purchase or the steps to fill out a form can be tracked and measured to understand the impact on revenue.
• Top Shipping Methods is a list that helps an ecommerce company eliminate waste from offering fulfillment methods that are infrequently used.
• Top Promotions is a list that helps marketers understand current effective discounts in order to manage them and plan for future promotions.
• Top Products is a list that guides merchandisers on what products are most sold and/or have the best sell-through.
Order and product analysis for the ecommerce analyst can take many forms. As with all analytical activities, the work is most successful when done to answer a business question that can drive a decision that results in an action that achieves a business goal. With orders and products, there are many ways to analyze them:
• Financial analysis on revenue, cost, tax, shipping, and so on.
• Item analysis of specific products and their attributes.
• Promotional analysis about the impact of different promotions on revenue, profitability, and customer engagement.
• Category and brand analysis to understand the financial measures, customer behavior, and purchasing volumes of categories and brands of merchandise.
• Event and goal analysis to understand the specific actions and pre-identified goals that customers engage in within the ecommerce experience before purchase.
• Checkout path analysis to identify what’s working and not working during the checkout process, or across the path of linear steps within a digital ecommerce experience leading to a purchase.
• Funnel analysis to identify points of friction, abandonment, and fluidity in multistep processes that result in a user taking an action.
• Cluster analysis as a method for grouping orders and products together with similar characteristics to understand them as a group.
• Up-sell and cross-sell analytics to utilize recommendations to suggest items that are related in some way to the product being browsed.
• Next-best-product analysis from data science models that suggest products frequently purchased when other products are added to a cart or purchased.
In the next sections of this chapter, we’ll explore each of these types of order analysis in more detail.
Financial analysis concentrates on the monetary metrics related to an order. Measures and derivative calculations based on revenue, cost, taxes, margin, and shipping can be analyzed. The business questions to answer by doing financial analysis include the following:
• What orders have the highest value?
• What orders result in the lowest shipping cost?
• Do tax rates impact the quantity or amount of revenue in purchases?
• What might my future costs be for orders of this type?
• What will my margin be next month?
An order includes one or more products or items, which have attributes such as product name, promotions applied, size, style, brand, category, department, cost, revenue, margin, and more. Digging into the details of products that have been purchased can help you answer questions like these:
• What products are most frequently included and bought in orders?
• Do certain items provide a larger or smaller contribution to margin?
• What brands are bought when other brands are bought at the same time?
• What sizes should I reorder or order next season?
• What categories should receive more attention from merchandising?
Discounts and promotions in which something is offered as an incentive to purchase are very common in ecommerce environments. Promo-code aggregators, coupon sites, and sites that track discounts and promotional offers exist across the Web. These promotions are typically selected by marketing and merchandising, which base predicted financial return of these incentives using ecommerce data. Thus the analysts should help merchandising with promotional analysis in order to answer questions such as the following:
• What promotions generate the highest and lowest margins?
• Are promotions causing adverse user behavior, such that categories or brands aren’t being bought without promotional discounts?
• What promotion should I offer to drive the highest revenue, the most number of products in an order, or the highest margin?
• Do promotions have a net positive or a net negative impact on the overall business?
• What promotions should I always offer or never offer?
Orders include, of course, the purchase of products. These products are associated with a category and/or department, and they are manufactured or marketed by a particular brand. Taxonomies (and ontologies) are often used to define categories. For example, the brand Nike sells athletic apparel and shoes. The department into which a Nike brand may fall may be the men’s department, the footwear category, and the athletic shoe subcategory. This information becomes valuable to describe and predict orders when it is captured over time and analyzed to answer business questions like the following:
• What categories generate the highest revenue?
• What brands have the lowest cost?
• What brands and categories will be popular next year?
• What brands and categories are most and least popular?
• What brands and categories have the highest inventory turnover?
• What brands and categories have the most days in inventory?
Customers complete various actions when engaged in an ecommerce experience. They may search for a product, browse a category, examine a lookbook, add items to a cart, register for a newsletter, sign up for an account, buy products, and more. It is possible to measure and track user and customer events and user goals such as those just listed. These events and goals can be tracked against the customers who buy and the people who do not buy in order to understand which events and goals assist in generating orders and creating value. Some of the questions you can answer with event and goal analysis include these:
• What specific events did customers complete compared to noncustomers?
• How many instances of a specific event occurred?
• What events are likely to occur in the path to purchase?
• What events occur when people report positive or negative experiences?
• Do certain customer segments complete certain events?
The path to purchase describes the sequence of touchpoints that a customer has before buying. This approach to understanding what influences a person to buy examines all the touchpoints where a prospect could have been influenced before a purchase. For example, paid search clicks would be captured, but so would visits to product review sites and social conversation. Less concerned with collecting data about exposure and interaction with specific channels or ads, path-to-purchase analysis looks to explain the impact of all interactions before a purchase. That way, instead of prioritizing one way to interact with a customer, the ecommerce site could figure out how to maximize all interactions in the path to purchase (Kimelfeld 2013). Path-to-purchase analysis on transactions can reveal these details:
• What marketing channels are my customers engaging with before purchase?
• What sources outside of my control tend to influence customers who buy?
• What interactions do customers have with my brand or the products I sell before purchase?
• Are there common touchpoints within all purchases?
Funnel analysis is among the most common types of analysis that can be done on orders. If you consider a funnel, like the type used to capture water, the transaction point is where the consolidated water exits the funnel and enters the container. In this metaphor, the water is the customer and his money and the exit of the funnel is the conversion event when money changes hands between the customer and the ecommerce site. Funnels are composed of pre-identified steps that, when performed in a sequence, lead to a transaction. In ecommerce the series of steps, events, interactions, or pages required to buy a product after it is added to a shopping cart is the “funnel.” Each step that a prospective customer takes in your predefined funnel is measured to understand whether people proceed to the next step or drop off. Your ecommerce funnel will be specific to your company but will likely include the Add to Cart > View Cart > Start Checkout > Checkout Steps > End Checkout with a Conversion. Many of you understand this funnel metaphor innately; you know that a conversion rate is calculated by measuring the ratio of how many people start the funnel compared to how many complete it. A macro-conversion may begin with the first page in the session and end with an order (i.e., conversion as the ratio or orders to sessions). Or the conversion may be considered a “micro-conversion,” in which the starting point in the funnel begins within the site (such as when the product is added to the cart). Funnel analysis helps to answer business questions about transactions, including these:
• Do people start the sequence of steps on my ecommerce site in the order expected and where we expect?
• Do any of the pages or events in my funnel work well to propel people to the next step—or do not work well and cause people to abandon the funnel?
• What is my baseline in terms of total conversion, abandonment by step, and step-jumping for understanding how the steps in the funnel impact conversion?
• Are there particular points in the funnel that need to be addressed because they cause abandonment or aren’t working as expected?
A basic funnel analysis could include four simple steps and related KPIs:
1. Visits (or Sessions). The total number of people who visited the site.
• Conversion Rate. The percentage of orders from all visits.
2. Shoppers. The total number of people who viewed one or more products.
• Product View Rate. The percentage of visitors who viewed a product page.
3. Add to Cart. The total number of cart adds.
• Cart Add Rate. The percentage of people who added one or more products to their cart.
4. Orders. The total number of orders.
• Purchase Completion Rate. The percentage of people who completed the purchasing process and ordered.
Cluster analysis is a method for analyzing data by automatically categorizing or sorting it into groups based on similarities in the data. Data with the maximum similarity are clustered together separately from clusters with weaker or no similarity. Based on the attributes listed at the top of the chapter, cluster analysis is suitable to be performed on orders and their product data and metadata. Clustering isn’t a specific approach to analysis as much as it is a set of methods for grouping together like data. To do a cluster analysis, you need to prepare the data to support the approach chosen. For example, you might assign a value (vector) to a transaction to do a centroid-based clustering, or you may cluster based on the distributions of the data or values in your data using a clustering coefficient. The approach that’s best for you, of course, depends on your business questions and your analyst and the level of data science they want to apply. Clustering orders and product data from orders is often a start to investigating customer or marketing data. Cluster analysis can answer questions like these:
• What orders are similar to each other and how are they similar?
• What are the unusual orders that aren’t similar to other orders, and how?
• Are there relationships between the data about orders that I haven’t seen before?
• Are there particular aspects of high-value orders that can be better understood?
Selling another item to someone who purchased is called a cross-sell. Convincing someone to spend more money when he is buying a product is called an up-sell. Since every order is associated with one or more products or items purchased, it is possible to determine what else people who bought that product also bought. If you are capturing what products other people purchased when they didn’t purchase the product the current person is buying, then you have the data to up-sell. For example, if people buy shoes, you may be able to cross-sell them socks, or leather protector, or shoelaces. If people are buying one brand of boot and another more expensive brand of similar boot has been bought or looked at by other people, then the more expensive boot may be suggested as an up-sell. Use data about products in transactions to power analytically driven recommendations for up-selling and cross-selling.
After a transaction, people do something related. They may fill out a registration or warranty card, check for shipping tracker identification, check on shipping status, search for related items, buy more things, return items, come back to the site and browse, and more. By associating the after-purchase behavior with the previous order, you can do analysis to answer the following questions:
• What actions do people take after completing an order with these products?
• What actions should be suggested to the customer based on this order?
• Are there common or uncommon actions people take after ordering?
Many different approaches to product analysis are useful for ecommerce. Each type of product analysis answers different business questions for different business groups and stakeholders. For example product analysis for merchandising focuses on understanding inventory, creative design, offer detail, the optimal price, sales impact, suppliers, and promotional markdowns/discounts. A product analysis for marketing would be more concerned with marketing channels, the user experience within the ordering flow or purchase funnel, and what products caused customer frequency, retention, and loyalty.
From an analytical perspective, it is helpful to consider the buyer lifecycle to frame product analytics. The buyer lifecycle implicitly, if not directly, involves a person using an ecommerce experience where they are doing the following:
1. Starting with the selection of a product category or brand
2. Narrowing to a specific brand or category
3. Selecting a product and adding it to a cart
4. Completing the steps in the funnel to order the product
5. Optionally, having contact with customer service
6. Optionally, returning the purchased product
7. Optionally, providing feedback about the product
8. Using the product (if it is not returned)
9. Communicating socially about the product
Each step of the product within the buyer lifecycle presented here can be understood by product analysis. Use the buyer lifecycle to help merchandisers, buyers, planners, and other business owners better understand how products are found and ordered as part of the overall ecommerce customer experience.
Ecommerce experiences contain catalogs of products from brands. It’s important for a merchandiser and manager to understand which brands drive desired performance. I have seen cases in which less than 10% of brands drove 90% of the revenue. A Zipf distribution or power-law curve applies. You could hypothesize that 80% of an ecommerce site’s revenue is driven by 20% of its brands. Understanding the financial performance of the brands offered on ecommerce sites means measuring and calculating the revenue, margins, and turnover of brands. Brand analysis goes beyond finance into behavioral analysis to understand the influence of certain brands on purchasing patterns—for example, showcasing premium brands while also offering lower-cost alternatives. Consumer perceptions of brand values and attributes reflect onto the ecommerce brand, so it is important to research qualitative aspects about brands. Market research such as brand awareness, top-of-mind recall, and customer sentiment are part of brand analysis. Analyzing the brand requires an understanding of the qualitative descriptors, lifestyle, and feelings people have about brands and products, while aligning with quantitative research about prospect and customer acquisition, behavior, and conversion. Product brand analysis answers questions such as these:
• What brands are most popular and drive the largest share of revenue versus margins?
• How do the attributes signified by particular brands influence the perception of my own ecommerce site?
• Am I applying the right merchandising and resources to support brand partnerships?
• What products within brands are value drivers?
For ecommerce companies that sell products against more than one category, it is extremely important to understand the financial, acquisition, behavioral, conversion, retention, and loyalty data for each of the categories. What is a category? It’s the highest level in the ontology/taxonomy—or the hierarchy of terms and concepts—that appears on your site.
Categories are such things as shirts, jackets, pants, underwear, socks. When products are grouped under an often-complex categorization hierarchy, both employees and customers can more easily understand and find available products. Product categories may be defined by information and librarian scientists, based on the industry knowledge or expertise, or created to support merchandising conceptualization. Regardless of origin, the only way to really know whether your ecommerce site’s categorization scheme is optimal is to measure and benchmark performance and test it. The metrics for analyzing category performance include measures of financial performance, such as revenue, margin, and profit, and metrics, such as conversion rate, cost of customer acquisition, and lifetime value. By understanding how your customers perform on a category basis—separate from yet related to products, orders, and brands—you can make decisions that reduce the cost of categories and maximize the value. Some of the questions that can be answered through product category analysis include the following:
• What categories are most profitable and drive the highest lifetime value?
• Do particular categories influence product orders for other categories or are particular categories of products frequently bought together?
• Are particular categories viewed at higher or lower rates for engagement and for conversion?
• Which drive the most orders? Which have the highest margins? Which are most discounted?
• What categories drive the most customer acquisition? Does that differ from retention?
• Should I eliminate or redefine categories?
Customers will inevitably need help with their site usage, purchasing, shipping, transactions, and product usage. Ecommerce sites fulfill these pre- and post-sale customer needs through service capabilities. Many ecommerce sites enable customers to self-service—whether through FAQs; or via online capabilities to chat with agents who can diagnose, communicate, and resolve issues. Some ecommerce sites may even use expert systems and artificial intelligence for customer service. Human or robotic assistants may be offered to guide customers in getting service. Analytics can let you know when you have good or bad customer service. Fortunately, poor online ecommerce experiences can be turned around and made positive through excellent customer service, whether that customer service is delivered by machines or humans. On the other hand, positive online customer experiences that lead to the need for customer service can be diminished and negatively impacted by poor customer service. Thus, ecommerce companies that provide online features and capabilities for self-service and offline features and capabilities benefit from analytics around customer service. Business questions to answer around customer service include these:
• Has my company developed metrics to measure customer service efficiency and satisfaction?
• What customer service capabilities are most used and effective?
• How are our customer service capabilities impacting customer satisfaction?
• When are customers reaching out to customer service and what are the most frequent reasons?
• What do customers contact our company about and how does this impact our top and bottom lines or inform our user experience?
Customers sometimes return products bought in orders. For all ecommerce providers, it is nearly impossible to stop returns. Customers have buyer’s remorse, products don’t fit, the products don’t match the buyer’s perception of what was described, or people change their minds. A business can’t avoid returns unless they explicitly disallow them. And even then, there are manufacturer’s defects, shipping damage, and other issues that can’t be foreseen. Companies that care about retention and loyalty may not welcome returns, but they need to understand them. A company can determine problematic products and quality issues and also understand the revenue and marginal impact of product returns through analysis. Business questions that you should be able to answer for stakeholders include the following:
• What products, brands, and categories are returned most or least frequently over what rate and time period?
• How do returns impact profitability?
• Are there specific customer types or customers that return products?
• Are there identifiable behaviors that increase the likelihood of a return?
• What actions should we take to reduce the amount or frequency of returns?
Social media analytics is far beyond the scope of this book, and not possible in the page limit. It’s an important type of analysis highly resonant with most companies because it’s trendy and buzzworthy; people and your colleagues engage in it. Social media is also important because it can direct revenue and, perhaps more important, it influences people’s perceptions about your ecommerce brand. In that sense, it’s important to analyze the ecommerce impact of social media on the ecommerce brand overall, the products sold, direct response that lead to conversion/purchase of products in transactions, and about customer service, and the overall customer experience. Jim Sterne’s book, Social Media Metrics, is a good starting point. Business questions that are important to understand about social media by using social media analytics include these:
• Do people referred from social media behave differently and transact differently?
• What social channels are most effective for selling specific categories, brands, and products or building brand awareness?
• What’s the cost of customer acquisition and lifetime value of social-media-referred customers, and does that differ from other channels?
• How does social media impact the customer journey on the path to purchase?
Merchandising for ecommerce has its roots offline in physical stores. It’s the term used to describe work activities related to selling the products on the shelf, ensuring that those products are displayed effectively, and successfully offering promotional offers and discounts. To do so, merchandisers work closely with buyers, who purchase and manage the inventory, and finance, which helps to ensure that the monetary impact of merchandising decisions is net positive. Wikipedia describes merchandising as “any practice which contributes to the sale of products to a customer.... It refers to the variety of products available for sale and the display of those product in such a way that it stimulates interest and entices customers to buy.”
Merchandising and merchandisers are, of course, important for ecommerce businesses. They perform the activities defined previously for online retail; instead of the store, there is the site, mobile, or connected experience. In online environments merchandisers will participate and even lead efforts to optimize the experience (see Chapter 8, “Optimizing for Ecommerce Conversion and User Experience”). Doing so requires online merchandisers to understand the elements of user experience and design, and also comprehend the data around the customer (see Chapter 9, “Analyzing Ecommerce Customers”). Approving and participating in designing user experiences around merchandising necessitates knowledge of the customer who is buying the products, so data analysis is important. Merchandisers use tools and technology that are part of, connected to, or ancillary to the ecommerce platform (see Chapter 12, “What Is an Ecommerce Platform?”), so merchandising teams can require not only business support interpreting and reporting data, but also technical support. In some ecommerce companies merchandising may buy and order inventory. In large companies, merchandisers will work with buyers who find and buy inventory that merchandising will work with to merchandise.
Some of the techniques, technologies, and tools ecommerce merchandisers will use include data visualization, AB and multivariate testing tools, digital analytics, search, and other research tools for competitive intelligence, product comparisons, and social feedback such as ratings and rankings, and product reviews (wherever they exist). These tools are used by merchandisers to fulfill their goals, which can be measured only through analytics. Goals related to prospect and customer behavior, such as recency and frequency of purchases, may be targeted. But more commonly, merchandisers are incentivized by improvement to conversion rates, increases in average order size, the number of products in a transaction, and the margins resulting from orders.
Because merchandisers actively make decisions and are often empowered to change the site, it is important for analysts to work closely with merchandisers. Otherwise, merchandisers will simply use available data and analytical systems without guidance. The best merchandisers can be tough customers for ecommerce analytics teams because they have to act based on the data—and depending on ecommerce business, merchandising decisions can be made multiple times a day or even thousands of times a day in larger ecommerce sites. Setting up suitable analytics environments for merchandisers, which are self-service based, is necessary. Merchandisers can then come to the analytics team when they have larger concerns not satisfied through self-service systems, which require deeper analysis.
Creative testing and optimization is an activity that may be led, directed, or influenced heavily by the merchandising team. You can read about ecommerce optimization in Chapter 8. For merchandisers, the primary concern will be on category, brand, and product pages—and the home page. But other pages may also be in scope, such as shopping cart flow and purchasing pages. The testing that merchandising will need help with may be multivariate, but it is more common, in my experience, for merchandisers to want to perform AB testing. In other words, instead of testing multiple combinations of elements, such as creative text, font, offers, and images, it is preferred to test one thing against another (i.e., control versus test). In this way, the merchandisers can understand simply which test is working best to support goals, and worry less or get confused by trying to understand which elements that were tested were most important to the increase in performance. Some of the merchandising questions that can be answered with testing creative include the following:
• What creative text and conversion copywriting perform better for goals, such as conversion, repurchase, up-sell, cross-sell, and revenue per visitor?
• How should we rearrange or change the product search, product viewing, product selection, or product purchasing experience?
• Are there particular styles of typography, font, tone, and other stylistic design elements that when present or different improve performance?
• What impact do personalization features have on merchandising, discount usage, and promotions?
The analysis of inventory drives financial performance by controlling stock levels. The goal of inventory analysis is to ensure the appropriate amount of inventory, such that stock levels are minimized and overstocking is avoided. The idea is that by controlling inventory levels so that products are sold (i.e., they turn over), you maximize margin and profitability and also cash flows. As a result, inventory analysis can be largely driven by the finance team and, in companies that have a warehouse, also by the team that manages the warehouse. Finance will of course look at such concepts as revenue, margins, and profitability in relation to inventory—and also track metrics such as days inventory on hand, inventory turnover, and stock outs. The warehousing team will be responsible for managing and ensuring data collection about inventory, but doing so may require analytical and technical guidance. At the very least, inventory data needs to be governed so the data is synched such that stock availability on-site matches inventory availability. Merchandising will review the data and reports generated by these analytics and finance teams to curate and tune inventory items and levels. Again, the shared goals of analytics collaborating with merchandising and finance is to ensure the physical inventory, if any, for the ecommerce site supports business goals. Ecommerce inventory analysis helps to turn stock into cash while managing the inventory such that there are sell-throughs and turnovers, and so that there is a minimal amount of overstocking and sellouts.
Inventory analysis requires data not just about the inventory levels on hand or anticipated, but also about how that data is selling over time. As such, it’s a ripe area for using key performance indicators to understand the inventory, such as these proposed by inventory control expert Jason Sentell:
• Inventory Turnover, which refers to the number of times the inventory sells a year (annually). The more times an ecommerce site has turnover, or sells out of inventory, the more revenue is generated. Thus, the more inventory turnover, the better, assuming that it isn’t marked down below cost. Turnover can be a leading indicator for identifying how much inventory you need on hand to drive desired results. To calculate inventory turnover, you would divide the cost of goods sold from stock sales during the trailing 12 months by the total inventory investment during the same period. For those of you who studied accounting, the way you value inventory matters to this analysis. FIFO (first in first out) and LIFO (last in first out) accounting will impact the turnover levels, and is also why alignment with finance is important. Because inventory turnover takes a snapshot in time, it can be tracked on the frequency most applicable to the ecommerce site and aligned with financial concepts and compliance.
When doing an inventory turnover analysis, you have to consider many things. First, you must set goals for turnover. For typical ecommerce sites, I have seen this goal range from four to six turns per year. The lower your margins, the more turnover you want in order to maximize the revenues and profits. Higher-margin ecommerce sites need to turn over less to make the same money—but any company, regardless of margin, wants to turn over as much as possible. Inventory turnover must be calculated for all products, not just overall across the entire set of inventory. That way, faster- or lower-selling products can be understood in the context of revenue, profitability, and lifetime value. With the right analytics environment in place, it is even possible to calculate inventory turnover for customers from marketing channels. Keep in mind that inventory turnover is generally applicable only to products sold by the site, not via drop shipping or direct shipping or sales of items not in stock.
• Number of Stock Outs, which is a KPI used to measure the number of occurrences of backorders and the average duration to fill stock. When stock is sold out and people are ordering it, they will have a poor customer experience because they can’t buy what they want, and your revenue and margins will be negatively impacted; lifetime value may be at risk and stock outs could cause customer churn. To understand how to use this metric to guide merchandisers, consider that if you have many occurrences of out-of-stock items, you must spend money to correct stock levels. This can be due to underordering at the onset when the product is first bought, or it could be the result of suppliers taking too long to replenish you. In the first case, buy more. In the second case, find a new supplier or set appropriate customer expectations. If you are manufacturing your own items, you need to look at your supply chain and manufacturing operations to accelerate the availability of your products.
• Customer Service Level (CSL), which is a KPI that measures customer satisfaction based on meeting their demand and sending them what they have ordered on time to their expectations. To calculate CSL, you would divide the number of stocked products ordered by the number of those stocked products that you deliver by the date you said you would deliver them. CSL credits only transactions in which all the products are shipped and delivered. If products are only ordered (and not shipped/delivered), then no CSL credit is given to the order. As such, this customer-centric measure demonstrates how well you satisfy the customer. As with stock outs, you count only the inventory you stock on hand (and not drop shipments or direct-from-manufacturer orders for which the ecommerce site is the middleman). A perfect CSL score would be 100%, meaning that all orders shipped, with all products ordered were delivered on time (Sentell 2013). It’s unlikely a business could meet the 100% level. Since this KPI will be distributed between 0 and 100, you can plot it on a graph, visualize it, and determine what the appropriate level for your business is. You want to tie CSL to customer satisfaction.
Offer management is a type of merchandising analysis that speaks to the effectiveness of what was presented to prospects or customers, such that they accept or reject the offer. From this acceptance or rejection, the offer can be expanded or dialed down. An example of an offer might be “free shipping if you order more than $49” or “30% off on all orders over $200.” Other offers might be for additional features (buy a shirt, get a free monogram) or might represent other attributes of a product. For example, an ecommerce site may offer a personalized stylist at an additional cost to work with the customer.
Offers are managed and analyzed over time, as part of marketing campaigns, within the site on particular page types, and within advertising and other marketing communications. Offers will have attributes such as the offer name, the offer description, the channel in which the offer is made, and the page on which the offer is to display. Offers can have different versions depending on the amount of mass customization or one-to-one personalization. These versions may all be based on one template or a number of templates that may be tested. Certain customers will be targeted with offers, so analysis requires understanding how the customer got the offer and why, and then figuring out whether that offer was effective and targeted correctly. The metrics and KPIs to measure and manage offers are standard, including total revenue by offer, profitability by offer, conversion rate of offer, and so on.
What’s hard is not the task of defining the financial measures and other metrics for comparing offers to goals or against themselves, but rather the task of setting up the right data model, data collection, reporting, and visualization to analyze in order to provide guidance, insights, and recommendations. Offers will need to be coded with data collection so that they can be measured in the way required, and so that they can be tested in multivariate and AB testing tools.
Pricing analysis refers to the set of techniques that enable an ecommerce site to set the price that achieves the company’s objectives. Often, pricing is set to maximize revenue at the lowest possible cost. Economists reading this book will recognize that the optimal price is set when marginal revenue equals marginal cost, which can be plotted and visualized. Pricing may be set based on a customer manually agreeing to buy at the identified cost due to demand for the product/service (Uber’s surge pricing is an example) or dynamically based on a number of factors built into a statistical model and automated in real time (like normal Uber pricing, and of course surge pricing overall is dynamically suggested). Before doing pricing analysis, it is helpful to do the following:
1. Identify the pricing objective. An ecommerce site must determine why it wants to charge the money it charges. Maximizing shareholder value is a solid and common reason for public ecommerce sites. For private ecommerce business, pricing can be driven to maximize profitability or contribution margin, or to generate the highest possible revenue. When pricing is maximized for profitability, it is assumed that both demand and cost are understood and relatively fixed over time. Pricing for profitability ensures short-term financial health, but may sacrifice long-term health because prices may be too high compared to those of competitors. Other reasons for charging the highest price may be to ensure the highest rate of return on investment. But ROI can be challenging to control via pricing because you don’t know the amount of product returns you will have when you make the investment. Thus a better option may be to test prices over time and adjust them for estimated returns to maximize return. Many ecommerce sites do not generate profits, so in this case, pricing may be used to provide cash and ensure available cash flow so the company continues to exist (by providing money needed to pay bills). Survival may mean taking a loss on pricing, just to generate cash flow to survive and to continue operations in the short term as more capital is raised. Pricing may be used to take or improve market share by underpricing against competitors. The idea is to sell so much stock that it can be purchased less expensively (economies of scale) and generate higher future profits as more and more people buy from this company as opposed to competitors. Still other ecommerce sites may set pricing based on scarcity or even product quality. The more scarce the product, the higher the cost. The higher quality the product, the higher the price.
2. Estimate demand. Estimations are based on understanding the number of potential customers who want to buy the product, and also on how sensitive customers are to prices. Economic concepts like the Price Elasticity of Demand may be used to estimate the impact of the quantity of orders to changes in the price, by dividing the percentage change in demand by the percentage change in price. The resulting value will be between zero and infinity. Zero means that the price is perfectly inelastic, such that prices do impact demand. Between zero and one, the demand is said to inelastic, and demand changes by a smaller percentage than price. A value of one means that the demand changes at exactly the same percentage as the price changes (it’s called unitary elastic). A result of one to infinity means that demand is elastic such that the demand changes by a larger percentage than the price. Infinity means that the demand is perfectly elastic such that customers will buy at one price but not at the other (Banerjee 2015). Other methods for understanding demand may include statistical analysis to analyze past prices and demand and other factors. Some companies may test different prices with minimal statistical or economic rigor. Others may do customer research and ask consumers to identify or select what they want to pay, which is then used as a proxy to estimate demand at different price levels.
Demand estimation also involves understanding price sensitivity based on other cognitive or mind-set factors of the prospective customer set (Banerjee 2015). Demand can be impacted by the following:
• Shared cost, when the site is bearing some of the cost (like free shipping)
• Income levels, when the maximum price is set to some ratio of total income
• End benefit, which reflects that the customer may want to pay in increments, being concerned less with the total price and more with the monthly payment
• Sunk cost, when buyers may have more demand for products that complement existing products
• Comparison shopping, which causes sensitivity in the sense that some customers want the lowest possible price, but if they can’t find it, then they buy at the cost available
• Unique value, which describes how demand may increase if the product has some level of distinction that can’t be found elsewhere
• Substitution effect, in which, if there are not competing products or substitutes, demand may be higher
• Quality impact, when customers will pay more for items perceived to be of higher quality
• Inventory effect, in which the scarcity of inventory or places to purchase it impacts the sensitivity of the price charged
3. Estimate costs. Costs come in two forms: fixed costs and variable costs. Fixed costs don’t change; they are your overhead and do not vary with the amount of money a company makes. Variable costs change based on different factors, such as the cost of materials or the quantity of items ordered. Total cost is the sum of fixed and variable costs. For an ecommerce site it is absolutely critical to understand costs to target a contribution margin that you must achieve on average to protect the financial health of the business.
4. Do competitive intelligence. The analytics team can help merchandisers set product prices by researching what competitors who offer identical or similar products are charging for them. Market research on customers is necessary to understand if the product’s quality and function are superior to those of products in competitive offers. As a result, if the quality or function is identical, superior, or inferior, then the price should be set accordingly against competitors’ prices. If the product is superior, the price should be modeled higher, or if inferior, then lower. Competitive intelligence is covered in detail in my book Building a Digital Analytics Organization; this information also applies to ecommerce.
5. Choose a pricing method. Analysts will want to calculate both the floor (or bottom price) and the ceiling (or highest price) for the product’s price, and put that in the context of what competitors are charging. To do this work, the analyst must create or develop a cost function to set the floor and estimate demand to set the ceiling. A cost function requires creating a short-run or long-run average or total cost-curve using economic principles. The demand function could use an approach like the price elasticity discussed in step 2.
6. Set the final price. In ecommerce, the manufacturer may have recommended a sales price, which the pricing analyst may use without analyzing, or the company may follow the steps given previously to set the price. Small producers or entrepreneurs may prefer to use the methods described previously and test prices to understand demand and pricing sensitivities of their prospective customer base. In such cases, the final price requires the analyst to analyze all costs against demand, and other consumer psychological and competitive influences and factors to set a price that will sell product against sales goals.
The goal of excellent ecommerce merchandising is to sell product and drive profitable revenue. Merchandisers can take many different approaches to repositioning and reframing merchandising to sell it. To do this type of analysis, you need a control group that is merchandised normally to use as a baseline. As an analyst, you can use several modalities to analyze, deconstruct, and categorize merchandising activities, to compare them against the control:
• Function-based merchandising refers to analyzing the product in terms of how effectively the merchandising communicates or places the product against its functional use. Product features and attributes will be used, along with visual merchandising cues to drive sales. Tools or household items are often merchandised based on function. Compare this treatment to the control.
• Event-based merchandising refers to analyzing the product in terms of how it was merchandised to support a specific event, such as the Super Bowl or World Cup. The control merchandising is compared to the event-based merchandising.
• Holiday-based merchandising refers to analyzing the product in the context of a specific holiday, like Christmas, to determine how the offer, messaging, and price helped to drive sales against the control.
• Seasonality-based merchandising refers to understanding the merchandising impact on sales by analyzing the control against a seasonal categorization or display related to the current season, such as winter or fall.
• Popular culture–based merchandising puts a product within a specific cultural context—for example, selling clothes based on the lifestyle of a rapper or a popular music artist. The analyst compares the product sales of the control against the new merchandising treatment to understand the financial impact of a celebrity endorsement.
Entire books have been written about supply chain analysis in which systematic operations between producers, wholesalers, and third-party shipping and logistical providers are analyzed. The goal of supply chain analysis is to help the ecommerce company reduce the cost of the supply chain by being more efficient overall and in handling and transporting inventory—from producers to supplier to warehouse and delivery operations. Although it is beyond the scope of this book to cover the detail and intricacy of supply chain analysis, it is helpful for the ecommerce analyst to understand the supply chain and some of the analysis relevant to it, such as modeling product demand, tracking the delivery times to the warehouse or directly to customers, helping to improve inventory planning, and understanding warehouse delivery scheduling.
A key activity of merchandisers is creating sale events that mark down products, provide discounts, and communicate promotions in order to turn over and sell-through inventory. To analyze product promotions, do the following:
1. Understand the promotional plan. The analyst must learn the details of the promotion, such as when it will be rolled out, what products are impacted, what new treatments or experiences will be created on the site, and, of course, the suggested promotional discounts and pricing.
2. Assist with the customer targeting. The attributes of customers must be analyzed to understand, determine, and select which customers to target with the promotion. It may be that all customers see it, or only specific customers who log in or who respond to an e-mail, and so on. The point of targeting is to use the data to find the customers most open to the promotional offer.
3. Help set the promotional price. Based on analyzing past pricing and cost and demand models, the merchandising team may rely on the analysts to set or help to set the specific discount or promotional price.
4. Implement and ensure data collection. When new functionality, new user experiences, or new features are rolled out to support merchandising promotions, they must be tagged or instrumented to collect the relevant data. Campaigns related to the promotion need to be coded, and the discounts identified in a way that can be tracked to the purchase.
5. Analyze the promotional pages against the controls. You want to compare the key performance indicators you created for sales performance against those on the new promotional pages.
6. Analyze the category performance. Because products exist in categories, it is important to analyze how the financial performance of the category was impacted by the promotion of one or more products within it.
7. Understand the impact of the media mix. Analyzing the effectiveness of the marketing channels that were used to communicate the promotion and drive people to the site is important. Again, compare this performance to your baseline to assess it.
8. Determine the return on investment (ROI). Model and calculate the ROI by understanding cost, revenue, margin, and profit and/or loss so that you can communicate it to merchandising and management and explain the net financial impact of the promotional activity.
A merchandising analysis should allow the merchandising stakeholder and other ecommerce stakeholders to understand data about the category, brand, product, and, if used, SKUs. As reviewed in the discussion on data modeling in Chapter 5, “Ecommerce Analytics Data Model and Technology,” merchandising analysis may require the creation of new data models, with custom facts, measures, and dimensions. Important dimensions for merchandising analysis include category, brand, and product. At the macro level, data for merchandising analysis should include financial measures such as Total Merchandising Revenue, Average Product Price, Average Order Price, Average Discount, Average Discount Percentage, Average Gross Margin, and even the Price Range for all products. Other customer and product-centric merchandising measures at the macro level can include Total Shipping Costs, Number of Total Products Sold, Number of Discounted Products Sold, Number of Total Orders, and the Number of Discounted Orders.
For each merchandising or product dimension you create, such as category, product, and SKU, consider collecting and analyzing data such as the merchandising revenue by category, by brand, and by product and the percentage change since the last comparable period. For each dimension, you may want to analyze the Number of Orders, Quantity of Products, Units in Inventory at the Start of the Period and End of Period, the Weeks Supply On-Hand, the Sell-Through Rate, and the Average Unit Cost. Every product category that is merchandised should use similar data as presented previously available for each subcategory. Helpful categories and subcategories to analyze include the total percent of category revenue represented by that subcategory, including the product name, brand, the number of visits, views, visitors, shoppers, and orders. At the SKU-level, if used, all of the metrics and data previously presented can be analyzed by SKU code or other metadata and attributes.