Procurement and supply chain professionals who use data based on customer purchasing habits to recommend operational decisions for warehouse stock could be unwittingly generating a cascade of inaccuracies and mounting inefficiencies in inventory management. This is because they lack a suitable analytical method to separate the data ‘chaff’ from the wheat, an expert has claimed.
Huanan Zhang, assistant professor of industrial engineering at Penn State University, aims to build a new ‘smart learning’ algorithm to achieve cost reductions and boost profits through improved warehouse inventory forecasting.
At present, operational inventory decisions are based on historical data concerning customer purchasing preferences.
However, this is a crude instrument, Zhang maintains, unsuited to high-dimensional inventory systems such as warehouses.
He said that implementing a misrepresented operational policy that is based on unchanged historical data may result in a “spiral-down effect” where the quality of the data being collected and operational decisions decline over time.
Zhang added that it is a difficult area to analyse due to the amount of data that is available and the amount of possible outcomes there are.
His new algorithm will analyse customer transaction data to craft a more accurate picture of prospective customer purchasing behaviour, automatically improving decisions by ‘learning’ and modifying forecasts as it crunches the numbers.
Ambiguities arise, for example, when companies use urban warehouses for the convenience of same-day shipping.
These are typically located in big metropolitan cities and are significantly smaller than conventional warehouses.
However, an item stored in one of these centres that generates lots of sales may be doing so simply because of the location and rapid shipping.
There is no way currently of determining whether consumers enjoy that item or want more of it.
However, existing algorithms tend to skew the data in favour of stocking more of that item based on historical sales, even though it may not subsequently be in high demand.
Zhang’s algorithm will not be hoodwinked by ambiguous data because it will ‘learn smartly’ to help reduce product waste and boost commercial revenues.