Why Retailers Should Tread Lightly With Predictive Analytics
Was it 2011 or 2012 when we saw the advent of big data? Suddenly everyone was talking about the data explosion and predictive analytics. Every industry was jumping onto the bandwagon, including retail.
It started innocuously enough with A/B testing, data gathering and funnel mapping, but it didn’t stop there. The new-age, number-crunching, stat-spewing marketers and “data scientists” went on to add predictive analytics to the mix, making retail seem like a public sector or finance business, and in many cases, taking retailers on a wild goose chase after projected profits.
Make no mistake: I enjoy number crunching as much as any marketer worth his salt. My grievance is against complete disregard of everything other than raw numbers.
What Exactly is Predictive Analytics?
Simply put, predictive analytics establishes correlations between past and current data, and predicts future trends.
To borrow an explanation from the book "Predictive Analysis For Dummies":
"Data mining is the discovery of hidden patterns of data through machine learning — and sophisticated algorithms are the mining tools. Predictive analytics is the process of refining that data resource, using business knowledge to extract hidden value from those newly discovered patterns.
Data Mining + Business Knowledge = Predictive Analytics"
Let’s say you want to build a predictive model for a product and determine its sales performance for the next month/quarter/year. You sequence the data such that you find correlations between product, price, sales and other variables like location, demand, seasonality and so on. Based on that data, you can predict future demand for and sales of said product.
However, in an industry where change is the only constant, how can you compare results from previous months or quarters with future results?
Does it Provide a Rough Road Map?
Not always because there isn’t always enough data on all variables. Sometimes, there are external factors that can be interpreted in ambiguous ways.
For example, a rough road map for the Oct.-Dec. quarter would suggest that the sales for winter gear would rise because the past trends suggest the same. However, the past data wouldn't suggest that it would be the warmest winter in years, and hence your sales figures would be affected. Even if you do take weather predictions into consideration, you can’t measure the exact effect of weather on sales.
The Oct.-Dec. ’15 quarter didn’t bode well for most retailers. Macy’s and Best Buy reported low sales. However, surprisingly, J.C. Penney reported profit in the same months. And what do you make of the retail sales growth chart over the past few years?
There's no way you can predict the future reliably (unless you having night visions like Nostradamus). This is why a lot of subject experts have called B.S. on it. Most of the time, predictive analysis is nothing but cherry picking findings, p-hacking and overzealous reporting.
Want to know why?
P-Hacking in Retail
A food and beverages company used a reactive model to price its products. It depended on commodity prices, competitor pricing and price dictated by retail vendors. This approach led to a chaotic pricing architecture. Therefore, the company turned to predictive analytics to solve its pricing challenges and achieve competitive advantage by managing demand and elasticity with more control and precision.
The company went on to develop a demand model by conducting a regression analysis of all price fluctuations in products sold by it and its competitors in recent years. It analyzed demand for each brand to determine consumer sensitivity to price shifts. This data was fed into an optimization model, which took into account multiple variables, to determine the right pricing model for the company going forward.
The end result is that the company is expecting an annual revenue growth of approximately 1.5 percent and a subsequent profit growth of nearly 3 percent.
But guess what, Statista predicts global beverage sales will grow by 3.9 percent in 2016 compared to 2015. A revenue growth of 1.5 percent is nothing to write home about when the entire industry is booming.
The Right Approach to Predictive Analytics
Predictive analysis isn't always so ambiguous; it's great for a number of things like handling inventory, reducing overheads, minimizing fraud and reducing returns. Let's see how to do predictive analysis the right way.
1. Handling inventory and reducing overheads: Inventory management, especially running out of stock or being overstocked, has been the bane of retailers for ages. Imagine running out of plum cakes before Christmas or anti-allergy drugs during the flu season. The opposite is just as scary — imagine the fall-winter collection hogging hangers at a fashion store in March.
However, sophisticated inventory management systems take into account past trends, current promotions, demand forecasts, existing stock levels and movement to distribute products to multiple locations in such a way that they bring the best returns. For instance, Vend uses a wide range of variables and filters to drill down data and help you identify trends so you can make better-informed decisions about your purchasing, inventory and pricing for online as well as physical stores.
An inventory management system that relies on predictive analytics can generate stock orders and set dynamic reorder levels based on trends and seasonality. For example, if based on previous trends, reports show that flu season is starting and is going to be mild this year, a drugstore or pharmacy can fine-tune orders to ensure their anti-histamines stock neither runs out nor sits idle after the season is over.
2. Minimizing fraud and reducing returns: Predictive models help unearth patterns found in historical data to identify risks and opportunities. They take into consideration consumer behavior to assess the risk associated with a particular set of conditions. For instance, a leading footwear retailer deployed a software called Verify to spot fraudulent customers who would buy shoes, wear them and return them.
With Verify, the retailer was able to reduce its return rate by 5 percent and save over $1 million annually. It also found that a majority of the returns were done with the support and involvement of the staff. This knowledge helped them to improve staff quality and, thus, minimize fraud even further.
The Way Forward
Ever since the dawn of civilization, experimentation has been the mainstay, the very foundation of the progress. With the ability to save costs, make better business decisions and get more satisfied customers, it's easy to see why retailers want to use predictive analytics.
However, empirical evidence suggests that retailers that have fortified their brand, carved a niche for themselves, not relied just on numbers, not followed the competition, and done the unthinkable have consistently cut through the clutter.
And if not for these crazy, experimenting, path-breaking retailers, the world wouldn’t be the same, would it?
Rohan Ayyar is a project manager with India-based digital marketing agency E2M.
Related story: Prepare Your Supply Chain for Predictive Analytics
Rohan Ayyar is the regional marketing manager for India at SEMrush. His blog, The Marketing Mashup, covers digital marketing from the perspective of B2B, B2C, lead generation, mobile marketing, SEO, social media, content marketing, database marketing including predictive analytics, and conversion rate optimization. In addition, he'll look at emerging marketing technology and how marketers can use it. Reach Ayyar at firstname.lastname@example.org.