Our Story

Our founding team has had a unique vantage point – witnessing the evolution of supply chain planning over multiple decades – by teaching cutting-edge theory and working with industry. The world has changed significantly, and it is becoming increasingly evident that the capabilities of available supply chain planning solutions has not caught up:

1. Globalization has led to a dynamic business world with increasing risk factors such as supply chain disruptions. From the global chip shortage to the ship stuck in the Suez Canal, this past year has demonstrated the need for agile and robust supply chains. In order to achieve this goal, planners need tools, which currently do not exist, that can help them anticipate and quantify the impact of future scenarios, evaluate strategies for mitigating risk, and adapt to changes in supply chain structure.
2. There has been an explosion in the amount, granularity, variety, and velocity of data that organizations are able to capture. Capturing this data, however, is only the first piece to the puzzle. Current planning systems are not able to effectively capitalize on big data, leading many organizations to use ad-hoc methods such as manual interventions and overrides.
3. Supply chain planning today relies heavily on forecasts. However, in practice, large forecast errors are very common which results in large errors in decisions recommended by planning software. Forecast errors are driven by noisy data as well as human biases in forecasting, necessitating tools that capture the information in forecasts while correcting for biases. We have also observed lack of consensus among the forecasts generated by different supply chain stakeholders.
4. Supply chain planning software today relies on stylized mathematical models which make strong mathematical assumptions due to historical computational constraints. The last two decades, however, have brought orders of magnitude improvement in computing power from the high performance computing industry and through the cloud. Not only is it now possible to run detailed optimization problems that do not rely on a lot of assumptions, but the boost in computational power also unlocks the ability to take a purely data driven approach: a formulation where input data is used to directly drive the end decision (common in machine learning).

These observations led us to create our unique offering which is composed of three systems.

1. End-To-End Data Storage: When we started AD3 we immediately realized that current data solutions weren’t enough. We saw data scattered throughout the organization leading to poor visibility and making any data manipulation incredibly painful. We created an end-to-end data storage system which works in conjunction with existing database systems in order to pool data across teams, locations, and products. With connectors to modern data visualization software like Tableau, users can now easily visualize end-to-end data from the entire supply chain.
2. Digital Twin (Parallel Universe): Supply chain planners need a way to quantify the impact of both past and future scenarios in order to evaluate risk mitigation strategies, adapt their supply chain structure, and make better decisions. AD3 is the first to build a tool, the supply chain digital twin, that gives planners these capabilities. The digital twin uses the granular data from AD3’s data storage solution in order to create a parallel universe where the underlying world – the decisions, supply, inventory, etc. – can all be changed and the resulting performance can be accurately measured.
3. Decisions via Optimal Machine Learning (OML): We take a purely data driven approach where the input data is used to directly drive the end decision. Our decision framework has been carefully designed in order to produce optimal solutions (unlike most machine learning formulations which do not guarantee optimal results). Unlike current decision frameworks which use a very limited set of data, our optimal machine learning system can use all data that the user decides may be relevant to the final decision. Our approach also reduces the reliance on forecast accuracy for supply chain planning and adjusts for human biases in supply chain decisions. This is enabled by our digital twin which allows the performance of decisions from the OML system to be accurately measured.