Farmers is a nationwide Department store with 60 stores and over 450,000 SKUs, including
a broad portfolio of men's and women's fashion, lingerie, beauty, and homewares.
Increasingly volatile seasonal consumer buying patterns are an ongoing challenge, which in recent years has been compounded by supply chain disruption.
Farmers' strategy for growth in a highly contested market is to focus on customer
experience with the best range and stock availability. To mitigate the risk of over or under-
supply, Farmers require pinpoint accurate demand profiles for each store. The retail
footprint, customer catchment, and large SKU count represent a complex forecasting and
replenishment cadence. Existing spreadsheets currently used for planning purposes cannot
provide this level of insight and granularity, are very time consuming, and often inaccurate
due to the obsolete forecasting technologies used.
Farmers wanted new planning tools that could save their planners time by automating many
business-as-usual replenishment forecasting activities within their business, alerting only
significant variance for intervention by their planners and allowing the planning and
merchandising team time to focus on other more high-value tasks.
Higher revenue per store from this approach due to fewer stock-outs (and obsolescence)
with corresponding higher stock turn was also required.
“Farmers wanted new planning tools that could save their planners time by automating many business-as-usual replenishment forecasting activities.”
Weekly, QU ingests 102 million lines of data from Farmers. Customer data includes daily
sales and stock on hand as part of over 200 product attributes per line for use in QU’s
proprietary forecasting models.
Near-term forecasts are produced weekly for the replenishment planning and review cycle and,
once validated, are transferred into Farmers ERP. This means the highest possible
percentage of SKUs can be systemically replenished based on the actual SKU location
demand.
Where QU supports management of the forecasts by planners, it is by exception using QU’s
alerting function, which highlights variances to parameters set by the customer.
QU’s architecture uses AWS storage and services with a process flow that begins with the
raw data landing in an S3 bucket. Crawlers are then used to infer schema and add to the
Data catalog and Glue jobs are run to clean the data. Quantiful’s data scientists use
Sagemaker to create features and apply machine learning with the forecast, which is then
hosted in an RDS for use by the QU web app.
The web app is implemented with React/Redux and Lambda backend handlers.
Commenting on the design-build, Quantiful Data Engineer Drew Hentz pointed out that
“AWS Infrastructure as Code is a powerful enabler of CI/CD methodologies and reproducible
environments and while there are many ways to implement the same functionality, finding
the best fit up front is quite valuable”.
QU is demonstrating increasing bottom-line business value by;
David Lean, CTO of Farmers, commented that apart from improved forecast accuracy, one
of the primary reasons for implementing QU was to enhance the productivity of the
planning team. David was particularly delighted with QU's forecast disaggregation
capability, which uses machine learning to automatically reset products and SKU forecasts at
a lower level in the hierarchy if a change to forecasts was made at higher levels.