Order Intake Optimization
A construction materials supply company receive jobsite orders against a catalog containing thousands of products. Most customers prefer to circumnavigate their ecommerce platform and directly email their designated customer service representative a set of site plans and product lists. The intake of these orders is time consuming and expensive.
A multi-step automation which leverages computer vision, a large language model, some prescriptive rules, and built-in human interactions along the way reduced processing time from hours per order to minutes.
Gas Pipeline Demand Forecasting
A team of traders working for a public utility company buys gas and coordinates the movement of that gas along interstate pipelines in order supply its retail customer base. This requires accurate and timely short-term forecast of customer demand. A confluence of factors contributes to demand on any given day, and the linear regression model they use for their forecast is at times innacurate and cumbersome to use.
We trained a machine learning model and automated it’s use. The result was better forecasts, requiring less human interaction, and have the potential to save the utility well over $500,000 annually through the timely buying and selling of natural gas in the market.
ERP System for Autologous Immunotherapy Treatment
A biotechnology company focused on the development and commercialization of therapeutics for cancer treatment. Its product, a cell-based immunotherapy cancer vaccine, is an autologous therapy which means that the drug is created using the patient’s own cells. Since these cells are live tissue, they have a very short shelf-life and must be processed within a very short time-window after they are harvested from the patient. The planning required to support the transportation and processing of the patient’s cells must be accurate, timely and precise. To accommodate this, we worked with the company to create a new breed of Enterprise Resource Planning (ERP) systems to manage this process.
At the core of this ERP system are two optimization engines. The first, is a “Capable-To-Promise” engine which allows a call center representative to offer available patient appointments based on available aphaeresis centers, transportation logistics and available manufacturing facilities. All this, while accounting for all other scheduled treatments that may affect the manufacturing and delivery of the new patient treatment.
Once a patient appointment has been selected, the second optimization engine schedules all the critical logistics and manufacturing steps that are required to successfully complete the treatment.
Production and Inventory Planning in aged Distilled Spirits
Large-scale production of aged distilled spirits such as bourbon requires the producer to maintain both a long and short-term view of their demand forecast and constantly make new decisions for the short-term liquidation of aged distillate as well as the short-term production of new distillate. These production decisions center around how much of each distilled spirit to produce and when to produce it in order to meet future forecasted demands given production capacity limitations and current inventory. The produced ingredients are then stored in barrels and are allowed to age for years on end. When the producer needs to satisfy demand for any one of its brands it must decide which of the ingredients (and at what ages) need to be liquidated (dumped from the barrels) and blended. The liquidation decisions are concerned with determining which stored spirit inventory to use to satisfy near-term product demand most effectively and efficiently. The production and liquidation decisions interact extensively with each other since, in the intermediate term, demand will be satisfied with some combination of current inventory liquidation and planned future production.
Using a linear programming model, we developed an automated liquidation management system.
Performance Incentive Calculator
A national quick-service restaurant chain pays quarterly bonuses to operations staff (store-level managers, regional and district managers) based on the quarterly performance of the store(s) under each manager’s supervision against 2 dozen performance metrics, focusing on financial performance, product quality, customer satisfaction, and treatment of employees under each manager’s charge. The business makes changes to these metrics on a 6-month cycle—adding new metrics, removing obsolete metrics, and changing the formulas for the calculation of existing metrics.
Additionally, for bonus-eligible employees who transfer between stores during a fiscal quarter, a complex set of business logic is used to determine which store’s performance should be used to determine their bonus.
The calculation of each performance metric, and the calculation of each bonus-eligible employee’s variable compensation based on those metrics was handled in SQL code as a batch process. The company had several issues that needed to be addressed to make their bonus system sustainable:
1. It was difficult for the business owners to see the impact of KPI changes without running these changes against a live dataset.
2. Changes to the Performance Metric calculations took several months to turn around and involved many different resources within IT.
Solution
We automated this calculation using an expert system with the following characteristics
1. Business owners are able to read and understand the logic used to determine employee bonuses
2. The change cycle time was reduced from several months to several weeks
3. Decisions are now decoupled from the data layer, allowing for easier change-impact analysis and the implementation of a requested “what if” tool.
4. Cloud hosting eliminates the need manage infrastructure.