Case studies / Retail
Case studyReplenishment forecasting that store planners trust
A home and garden retailer replaced habit-based store ordering with forecasts that explain themselves, cutting stock-outs and overstock while planners kept the final say.
Overview
A home and garden retailer with 118 stores across the UK and Ireland, supplied from one central distribution centre, with sharply seasonal trade and a heavy promotional calendar.
Scale. Around 22,000 active products and more than 40,000 product-store combinations replenished weekly, rising in spring when garden ranges arrive.
The challenge
Each store manager placed their own orders, mostly by habit. They looked at the shelf, remembered last year and added a margin for comfort. Some were very good at it. Many were not, and almost none had time to think about it carefully during the spring rush.
The result was a familiar paradox: stores ran out of compost and barbecue charcoal on sunny weekends while back rooms filled with patio furniture that would be marked down in August. Promotions made it worse. Managers over-ordered for a deal, then under-ordered the following week while the surplus sold through, and the distribution centre saw wild swings it could not plan for.
The head of supply chain had tried a forecasting tool before. Planners ignored it because it gave numbers without reasons, and when it was wrong nobody could tell why. Any new approach had to earn their confidence, not bypass them.
The solution
Every week the system produces a demand forecast for each product in each store, drawing on recent sales, seasonal patterns, the promotion calendar and known events such as bank holidays. It turns those forecasts into suggested orders, taking account of current stock, deliveries already on the way, pack sizes and shelf capacity.
Planners see each suggestion with its reasons laid out simply, so they can judge it quickly. They can override any line, and the system asks why: local event, display change, quality problem, supplier issue. Those reasons are reviewed and, where a pattern emerges, used to adjust the forecasts.
Accuracy is monitored by product group, region and store, with alerts when a forecast drifts. The distribution centre receives a consolidated view of upcoming store demand several weeks ahead, which has made labour and inbound planning far less reactive. Store managers now spend their time on the shop floor rather than on order sheets.
The engagement
- Eight months
- Seven specialists
- Data preparation, demand forecasting, promotion and seasonality handling, suggested-order design, planner workflow, override capture and accuracy monitoring.
Integrations
- Point-of-sale system
- Warehouse management system
- Supplier ordering
- Promotion calendar
- Data warehouse
How we delivered it
- 01
Prepare history and define accuracy
Six weeks
We assembled three years of sales, stock, deliveries and promotions at product-store level and cleaned out the noise: stock-out weeks where sales were capped, one-off clearances and store refits. With the supply chain team we agreed how accuracy would be measured and which product groups mattered most.
- 02
Build forecasts and test them against past ordering
Eight weeks
Forecasts were built for every product-store combination, with separate handling for seasonal curves, weather-sensitive lines and promotional uplift. We replayed the previous season and compared what the forecasts would have ordered with what managers actually ordered, line by line.
- 03
Show planners suggestions with reasons
Ten weeks
In a pilot region, central planners received suggested orders with a short explanation for each: base demand, seasonal lift, promotion effect, current stock. They could accept, change or reject each line, and every override was recorded with a reason code.
- 04
Extend store by store
Ten weeks
Stores joined in groups once accuracy in comparable stores had held for several weeks. Override reasons were reviewed fortnightly and fed back into the forecasts, so recurring corrections became part of the model rather than repeated by hand.
Results
- Fewer stock-outs on core and weather-sensitive lines
- Less overstock and lower markdown at season end
- Planners kept control and focused on exceptions
- Steadier demand signal for the distribution centre
Stock-outs on core lines fell noticeably in the stores using the forecasts, most visibly on high-volume weekends when the weather turned. At the same time, less excess stock was left in stores at the end of each season, and markdown spend dropped over the first full year.
Planners kept control. Most suggestions are now accepted as they stand, and planners spend their time on the exceptions that need local knowledge: a new housing estate near one store, a competitor closing near another. Their overrides have become one of the most useful inputs to the model.
The distribution centre gained a steadier, earlier picture of demand. Promotional weeks no longer arrive as surprises, and inbound deliveries can be booked with suppliers further in advance.
The difference this time was that it told us why. When we disagreed, we said so, and a few weeks later the forecast had learned from it.
What made the difference
Planner overrides were captured and used to improve the forecasts. Treating corrections as data, rather than as resistance, made the forecasts better and gave planners a reason to engage.
