Read our article on

.

Managing stock levels effectively is an ongoing challenge for retailers, both on and offline. But getting it right (or wrong) can have a direct impact on overall profits. Bobby Gray, head of analytics and data marketing at Artefact looks at the role that machine learning (which extends far beyond traditional prediction techniques) plays in evolving demand forecasting.

Last year’s sales should not equal this year’s predictions

Inventory management is an essential part of retail life – and has a direct line to overall revenue. It’s also a perpetual challenge.

Traditionally sales data from previous years is used to predict what will be sold in the period ahead, but this is often overly simplistic.

Let’s take a product with poor sales figures; the obvious conclusion is that it’s not popular. But rather than the low numbers reflecting buyers’ disinterest, it might indicate the opposite – rapid purchase of the item has caused it to sell out more quickly than predicted, leaving a gap on the shelves. Lack of availability results in low sales preventing it fulfilling its sales potential – a cycle that can continue indefinitely.

This is a wasted revenue opportunity – but because predicting sales is still a fairly inexact science it’s quite common, along with many other sales-related situations that can adversely affect profits.

It’s a particularly critical issue for retailers that have thousands of product lines and/or need to manage perishable inventory efficiently; excess stock is as bad for business as having too few items. (Estimates suggest that poor inventory management costs US retailers close to $2bn per year.)

The variations in purchasing patterns

Purchases are affected by many factors: weather, shopping trends, regulation, new products, buying behaviors, seasonal events, promotions, competitor activity, a pandemic… Predictions based on previously recorded data don’t necessarily factor in specific events, making monthly sales appear evenly distributed when this is unlikely.

Taking seasonality into account when predicting future sales is not fool-proof; a date being a weekday or a weekend causes fluctuations in the figures. Consumer behavior can also be influenced by events such as national holidays, cultural festivals and major sporting tournaments. Even making forecasts around annual certainties is not a guarantee of success – Easter, for example, is a different date each year, with sales of some products varying widely depending on whether it’s early or late.

Price level signals can also make top-line sales figures misleading; an in-store promotion can markedly affect the sales of a product from a given category, and even make the store as a whole more attractive to shoppers.

Meanwhile, a product might show as ‘in stock’, but be unavailable, with big-box retailers often struggling to refill shelves in real-time. This opens a ‘sales downtime’ window, particularly for popular items, which move quickly as soon as they’re available for purchase.

These are real issues faced by retailers. But significant incremental profit can be unlocked through effective order and inventory management, which also uses the marketing budget more effectively, and increases average selling prices thanks to the reduced need for promotions to clear excess stock. Easy to flag, this presents a major challenge, not least because it requires the processing of data from vast numbers of stock items.

Advanced technology and smart inventory management

Machine learning models can forecast sales months in advance; they extend far beyond traditional prediction techniques, which tend to rely on the standard factors of day, product and store. Buying behavior is far, far more complex, requiring sensitivity to things such as seasonality, consumption trends, price levels, one-off events, etc.

This calls for retailers to collect and analyze huge amounts of data from different sources and in varying formats; big data tools process the information into the clean and readable format required for predictive modeling.

Armed with this detail, retailers can leverage readily-accessible information to resolve common problems. For example, it’s almost impossible for employees to continually monitor stock availability and reorder immediately. But real-time (or nearly real-time, such as daily) sales data at the item level (and models that analyze the usual sales flow of a product) can detect when an item has run out, or is about to do so. Deviations from the normal time period between each sale in a given market (offline and online) can be flagged and human intervention used to review and rectify.

Advanced technology is increasingly putting many win-win scenarios within the reach of retailers, such as smart inventory management increasing incremental revenue opportunities. Machine learning can also assist with fine-tuning operations; activities such as optimizing product assortments, offering more enticing and profitable promotions and setting prices are all revenue-generating options that become feasible with the right information to hand.

The right data and tools result in prompt and accurate reports, making once time-consuming, complex and potentially impossible tasks more straightforward. Viewed from that perspective, machine learning has an important role in the evolution of improved retail sales planning and revenue growth, whether in-store or online.

Thanks a lot for reading! Please feel free to reach out if you wish to contribute to the package development or have any improvement ideas. In the meantime, you can visit the Artefact blog for more information about our data projects.

Read our article on

.

Artefact Newsletter

Interested in Data Consulting | Data & Digital Marketing | Digital Commerce ?
Read our monthly newsletter to get actionable advice, insights, business cases, from all our data experts around the world!

Newsletter Sign Up