ST. PETERSBURG, Fla. -- Wegmans Food Markets, Rochester, N.Y., is in the early stages of planning to use scan data to replenish its entire supply chain, including its stores' distribution centers.
These scan-based forecasts could also be shared with vendors, to be used as part of their planning and distribution systems that build finished-goods inventory.
Wegmans' focus right now is on initial tests with point-of-sale scan forecasting. POS scan forecasting is an evolution of current continuous replenishment programs, which involves the vendor managing a customers' inventory at the DC level.
"Since POS movement data provides the best measure of product demand, accurately forecasting it promises to reduce days of supply and increase service levels," said Tom Adelsberger, systems manager for inventory management at Wegmans. Adelsberger spoke at the Grocery Manufacturers of America's Information Systems and Logistics/Distribution Conference here.
Adelsberger said the average forecast accuracy of its test database, which contains more than 100,000 forecast records, is about 70%. Wegmans' current sale forecast accuracy, using warehouse movement data, is 45.7%. Wegmans defines forecast accuracy as a measure relating average forecast error and actual movement, expressed as a percentage.
The test database includes records on 1,900 items at 53 stores, in five different departments and 13 different categories, including paper towels, napkins, facial tissue, diapers, bath tissue, cereal, dog food, yogurt, dentifrice, batteries, juice concentrates and panty hose.
To create a POS forecast, Wegmans uses a technique called event-driven POS forecasting. This method breaks a forecast into two parts: baseline demand and incremental demand. Using forecasting software, the retailer forecasts the baseline demand using time-series techniques.
Wegmans has more than 104 weeks of movement history, by item and by store, and uses this information to create the baseline measurement.
Then, the retailer forecasts incremental demand using Wegmans' promotion events calendar and the promotional lift table. The promotional lift table relates lift value to lift causes based on previous history. Causes of promotional lift include price, ad strength and store display strength.
For the first 12 weeks of this weekly process, 20% of the items had forecast accuracy of between 85% and 90%. Overall the accuracy of the system came out above 70%.
The POS forecasts were compared to warehouse movement forecasts of 45.7%. For one item during a sale, 35,000 units were sold but 56,000 items were shipped. But forecasting warehouse shipments at 45% yields a warehouse movement forecast of 120,000.
"Where we expect to spend time in the future, when we start working on [POS scan forecasting] on a regular basis, is with forecast managers who would be involved with the exception activity, and working on items that have not shown good forecast accuracy," said Adelsberger.
Two forecast managers are in place and will ultimately be responsible for continuously monitoring and improving forecasts in grocery, dairy and frozens, as well as general merchandise and health and beauty care.
Ken Kappner, senior technology manager at the Global Logistics Center for Procter & Gamble, Cincinnati, who also spoke at the conference, said retailers typically plan their replenishment based on human count at the shelf.
This information is used to create a store-level order, which is placed with the retailer's DC. Manufacturers plan their production based on a statistical analysis of previous shipments to the retailer's DC, plus any available promotional information that can be obtained.