AI-Driven Demand Forecasting

Founded in 1980, Forus is one of the most prominent Chilean companies in the wholesale and retail sale of leading clothing brands, with a total of 519 stores. It operates in six markets in the region, combining physical retail, e-commerce, and wholesale operations. The organization had major challenges ahead. Its business information was distributed across the ERP, store point-of-sale systems, the e-commerce platform, supplier systems, and multiple spreadsheets, without a common and reliable data layer that would allow different areas to work from the same version of the information. In the fashion retail industry, this situation has a direct impact on the business: overbuying means margin lost in liquidation processes, while buying below demand represents sales that are never recovered. Without a single, governed database, purchasing decisions largely depended on the judgment of the planner in charge, which also hindered the possibility of scaling artificial intelligence initiatives.

Ingenia designed and built a modern data architecture on BigQuery, with the data lake as a single source of truth. Through AI-assisted extraction and ingestion processes, all dispersed data was consolidated, and the platform was modeled so that the same base fueled both descriptive analytics and machine learning purchasing models. Forus consolidated all of its information into a data lake on BigQuery, which allowed it, for the first time, to plan its purchases using machine learning models. In turn, the internal team was put in charge of the platform, ensuring its autonomy over the solution.

Instead of replicating existing information silos, all sources were unified into a governed layer on BigQuery and object storage. Having a single source of truth allows for the enabling of BI and machine learning processes without duplicating pipelines or generating discussions about the validity of the data.

Given the heterogeneity and number of sources involved—ERP, POS, e-commerce and suppliers—artificial intelligence models were used to accelerate the extraction and normalization processes, reducing manual mapping time from several months to a few weeks and decreasing the cost of incorporating new sources.

The platform was designed with a focus on scalability: today it allows for answering both what happened, through descriptive analytics, and how much to buy, through machine learning demand models by country and brand, using the same database and without the need to rebuild the architecture.

With unified information, planning teams stopped working with spreadsheets and began operating with projections based on machine learning. Purchases were adjusted to the actual demand of each country and brand, which allowed Ingenia to lay the foundations to achieve: reduction of stockouts, decrease in liquidation due to overstock, margin recovery without increasing staff count, demand projection by country and brand, and a database prepared to scale toward new artificial intelligence use cases. Forus was positioned on a database prepared to support future artificial intelligence use cases, consolidating itself as a platform in constant evolution and not as a closed project.