Demanding Solutions

Demanding Solutions helps data scientists in mid-size companies forecast retail demand, using the same kind of power currently only available to the largest organizations.  We make it easy for them to get started, to apply cutting edge techniques to their data, and to customize as much as they want.

Forecasting product demand accurately is critical for supply chain management in retailers and product companies.  Poor forecasts mean items are unavailable or wasted, so improving forecast accuracy by just 1% could mean a 2-4% improvement in profit.  The largest businesses employ armies of Ph.D's to constantly tune and improve their forecasts, creating customized tools.  Mid-size companies with $100 Million -$1 Billion in revenue represent 1/3 of US GDP and don't usually have the luxury of big overhead staffs, but they are starting to hire small data science teams, charged with improving results.

However, the legacy forecasting tools date from another era, when forecasting was done by Finance.  These tools are often inflexible, with black box forecast models, making it difficult for data scientists to actually apply their skills.  These data scientists know there are better techniques being published every day that they could try on their data.  But investing the time to try each new approach is hard to justify for these small teams, with no certainty that a particular tool will work well for their specific data.  They also have to forecast across thousands of SKUs with different characteristics every month, or even week, so it's also difficult to customize to bring their domain knowledge to bear.

We created a fully functioning web app to solve these problems, implementing four very different forecasting approaches, including cutting-edge tools from Facebook and Amazon released only in the last few months, and tested it on real-world data for a major mid-market product manufacturer using data on thousands of SKUs across 4 years.  The company's forecasts benefit from a mature process and years of optimization.  Despite that, we were able to deliver performance that beat their best results by several percentage points, demonstrating the practical and financial value of our product to its target audience.