MaxAB, founded in 2018 and based in Egypt, offers an app-based and web channel ordering system for grocery store owners. Small grocery stores make up the majority of this type of retailer across Egypt and Morocco. Each store is usually dependent on deliveries from individual suppliers, making stock replenishment slow. Using MaxAB, orders can be placed on a daily basis, with deliveries received in just 24 hours.

The company provides grocery stores with stock from its network of 30 warehouses across both countries, with 3,000 product lines in dry goods, chilled goods, and beverages. End-product costs are further reduced by aggregating orders and buying warehouse stock in bulk. Fast warehouse product turnover also means fresher products and improved customer cash flow.

MaxAB had been developing its own demand forecasting models in several areas. These were primarily to improve predictions for customer sales demand, manage product pricing levels, offer special pricing for retailers, and optimize warehouse stock levels and product holding costs. The company needed to streamline model development and iterate at faster speeds. “Our data team was using AWS for our demand forecasting models, but we needed them to be more accurate,” says Amr Faisal, chief technology officer (CTO) at MaxAB. “Working closely with AWS helped us to refine and speed development, and realize the full potential of what we could achieve.”

When the MaxAB data team began working on its demand forecasting models using AWS, it developed them manually in silos. This meant it was difficult to collaborate on model development, resulting in slow model iteration speeds and a lack of model interdependency.

MaxAB benefited from a regional initiative, the AWS Prototyping Program, to help it build a new model development structure using Amazon SageMaker Studio, the fully integrated development environment for machine learning (ML). Gaining access to targeted AWS support helped the company streamline its processes. “Working closely with AWS on this program was a big enabler for us,” says Faisal. “It allowed our data team to understand which of the AWS services we already used would best support the building and optimization of all demand forecasting models in development.”

Using Amazon SageMaker Studio, the company has reduced demand forecasting model completion times by over 94 percent, taking it down from 9 hours to just 30 minutes. To pull in data from multiple sources across the company and support model training, the company uses Amazon Athena, which allows it to query data instantly. The company has cut manual data ingestion processes by linking Athena to Amazon Simple Storage Service (Amazon S3), allowing it to retrieve any amount of data from anywhere.

MaxAB has increased model accuracy from 50 to 80 percent—a 60 percent improvement—by building on AWS. The company has also been able to increase model iteration speeds by a factor of 8. This means that its team can spend more time on model development by applying different model features and algorithms to further improve accuracy rates. “By freeing up time, it’s much easier to experiment,” says Faisal. “We can continually enhance demand forecasting accuracy on and get better results—it’s not just a one-time thing.”

Using AWS, the company has developed its first operational model to track product sales spikes on the same day. This allows its marketplace team to understand why spikes have occurred, revisit pricing structures, and limit sales of specific products to protect supply. With an almost immediate understanding of changes in product demand, the company can create special sales’ pricing offers for its retailer customers.

Daily demand predictions also help in the use of on-demand workers. “Along with full-time staff, we also employ temporary workers for our warehouses to fulfil customer orders,” says Faisal. “Using AWS means we can more accurately predict how many people we need on a given day and lower our employment costs.”

MaxAB has also developed two additional demand forecasting models that can provide demand predictions for either the next-day or for 7 days ahead, which it plans to begin using soon. This will allow it to connect warehouses and make demand predictions on a regional level and connect each model to further improve demand-and-supply forecasts.

The company’s data team is now primed to develop its demand forecasting capability further and support future growth using AWS. “The entire experience of working closely with AWS to build our new demand forecasting development structure really energized the entire data team,” says Faisal. “We felt our vision and goals were fully supported throughout the whole process, and now we’re ready to grow on a much smarter and better-informed basis.”