Sigma Healthcare has used several machine learning models to consolidate gains around its demand forecasting for medication, helping it optimise inventory and availability.
The full-line wholesale and retail pharmacy company previously made gains in its forecast accuracy of between five and 10 percent after implementing the response and supply planning module of SAP integrated business planning (IBP).
Use of this module curtailed “inefficiencies” in demand and supply planning processes, replacing manual spreadsheets and hours of work per day for supply planners to review and process purchase orders.
The module cut the time needed for supply planners to process orders from five-to-six hours a day to one-to-two hours, with further reductions occurring over time.
“With that workload freed up, there was a lot more time to actually focus on forecast accuracy, so we were able to increase [accuracy] by five-to-10 percent,” demand planning and systems manager Marcus Williams told the recent SAP NOW AI summit in Melbourne.
Forecast accuracy has been further improved by using machine learning forecast models in IBP.
“Over the last couple of years, I’ve led a number of projects inside the operations planning team at Sigma in regards to making use of the full features in IBP demand planning, especially the machine learning forecast models like extreme gradient boosting, standard gradient boosting, and also auto outlier correction,” Williams said.
Extreme gradient boosting is useful on “large and complex datasets”, according to SAP technical documentation, helping demand planners “forecast sales or identify problem areas in inventory or delivery”.
Automatic outlier correction, meanwhile, prevents “a data entry error or a one-time event that affected the sales results” from skewing forecast calculations.
Williams said the combined impact of these machine learning models had been an additional improvement in forecast accuracy of “at least another 10 percent, on top of the five-to-10 percent we got initially” from implementing the response and supply planning module of IBP.
Williams predicted additional optimisations would come from Sigma Healthcare starting to use SAP Joule, the name given to the generative AI ‘copilot’ that works with SAP systems.
“We’re really keen to start looking at Joule because I think generative AI within IBP for our planning team will be extremely beneficial,” he said.
“Rather than spend 30 minutes trying to solve a problem, [Joule] can assist.
“They might see an out-of-tolerance forecast or safety stock too high; they’re the types of things I’m hoping generative AI can really assist with.”
An out-of-tolerance event is similar to an outlier — something atypical that impacts demand or supply planning.
Safety stock, meanwhile, is extra inventory that is typically held to meet unexpected demand for a product or other inventory availability issues.
Williams predicted it would be important to keep pace with quarterly upgrades of IBP, and with the pace of AI in general.
“[That’s] going to put us in good stead and continue to give Sigma Healthcare a real competitive advantage.”
Sigma Healthcare runs both SAP S/4HANA and IBP, and partnered with EY to optimise the IBP environment.
The company is perhaps best known for its retail chains, which include Amcal and – as of this year – Chemist Warehouse.
