Retail Technology
| Log in | Subscribe



Subscribe | Log in
Retail Technology
Subscribe

Retail inventory optimisation expert Matthias Steinberg argues that forecasting technology has to date failed to adequately deal with slow-moving products even though they constitutes a large part of a retailerÂ’s portfolio

Retail inventory optimisation expert Matthias Steinberg argues that forecasting technology has to date failed to adequately deal with slow-moving products even though they constitutes a large part of a retailer’s portfolio

 

Sales forecasting in retail is considered a mature technology, which is based on a statistical foundation that has been known and verified for several decades.

 

Add in the fact that product availability and supply chain efficiency is paramount in the retail industry, one would expect retailers to be largely satisfied with the performance of their forecasting technology. But Matthias Steinberg, chief executive of cloud-based data forecasting services provider Lokad considers if is this really the case?

 

“Even though more than half the inventory of a typical food retailer is in the stores, we cannot identify a single retail chain in Europe that is performing advanced forecasts on a store level (beyond simple moving averages or simple rules),” pointed out Steinberg.

 

He continued: “Last year we participated in an extensive forecasting trial, as one of three vendors competing to provide daily point of sale forecasting capability to one of Europe’s largest grocery retailer. We all failed to meet the mark.

 

“Ask any retailer, wholesaler or e-commerce merchant with a large product portfolio how their forecasting systems perform on the slow movers. These are products that sell infrequently and which constitute the often significant ‘long tail’ of a product portfolio. I wonder how many millions have been independently invested in forecasting solution from the largest vendors; chances are you will receive pained looks.”

 

Managing sparse data

 

According to Steinberg, all these businesses suffer from the same failing: the inability of existing forecasting technology to deal with sparse data, i.e. data that is caused by infrequent sales in low numbers. Unfortunately, this type of sales data is an important part of the product portfolio in retail.

 

“Realising that the industry, including us, has been unable to solve this problem was puzzling enough. What we discovered then was even more surprising (or rather shocking),” he declared. “When forecasting sparse sales, the most fundamental concept of sales forecasting and safety stock analysis which we all have been using without challenge is completely unsuitable: The assumption that demand and forecasting error is distributed following a normal curve.”

 

He added: “In a true case of not seeing the forest for all the trees, we had missed the obvious. Fast forward a few months, and we are introducing “quantile forecasting” to the market. By applying the concept of quantiles to inventory optimisation, we not only solve the problem caused by sparse data, but also significantly improved the forecasting accuracy compared to classic methods.”

 

The concept of quantile forecasting (also called quantile regression) has been known for decades in academic circles. And it has been applied successfully in other trades, for example financial analysts have been extensively using quantiles for Value at Risk (VaR) analysis since the late 1980s.

 

Harnessing power of quantile forecasting

 

Steinberg said quantile forecasting was particularly suited to certain applications. These include those that require high service levels of 90% and above; intermittent demand (or slow movers), or bulk orders (with spiky demand).

 

“Benchmarks against our classic forecasting technology with retail clients in food, non-food, hardware, luxury and spare parts show that quantile forecasts typically bring a performance improvement of over 20%, that is either 20% less inventory or 20% less stock outs. In our opinion, quantile forecasting not only solve the problem of forecasting sparse data, but also makes classic forecasts plain obsolete for inventory optimisation in retail, wholesale and e-commerce.”

 

Steinberg added that, until Lokad’s breakthrough with quantiles, the industry was not lacking statistical insight, but rather insights into “the profound relationship between quantiles and inventory optimisation”. However, he also admits that quantile forecasting models typically require about 10x more processing power compared to classic forecasting models.

 

So, he added: “Without cloud computing, quantile forecasting is hardly possible.” Lokad’s business is based on cloud-driven analytics and forecasting technologies.

 

Steinberg concluded by predicting that quantile forecasting will make classic forecasting obsolete in the next 10 years by “not only forecasting slow movers correctly, but also by taking forecasting performance in retail to a whole new level,” he said.