Abstract: Historical, often experience-based forecasting approaches are failing to forecast accurately in changing conditions. Amplify’s layered forecasting system makes inroads into overcoming current challenges, and has been awarded by an EU innovation body, REACH, as a promising solution to a notoriously difficult problem.
The Forecasting Problem
If only we had a crystal ball: subsequent years of shocks have made experience-based forecasting difficult, causing us to err on the side of caution, while macroeconomic uncertainty requires us to be as efficient as possible, freeing up working capital. We must continuously adapt, stay ahead of market trends, and predict accurately what customers will need, how much, and when. Yet, many organisations struggle to keep up with the constantly evolving demands and conditions of the market. They are either investing too many resources, sometimes missing opportunities to grow over or under-forecasting demand, and having too much or too little inventory, decreasing efficiency and effectiveness, and ultimately hitting margins.
So how can we improve our forecasting ability in this climate to take advantage of opportunities now, while both growing our business and reducing the cost to serve? Even without the crystal ball, how do we maximise our chances of forecasting demand accurately and systematically across often hundreds or thousands of product and service lines? It is both crucial and extremely hard to get it right consistently!
Our forecasting system makes inroads into these challenges. It has been selected as one of 6 winner solutions out of 400 applicants by REACH, an EU innovation body fostering the use of AI in business, as a promising solution to a notoriously difficult business problem. It analyses historical data, then layers on both macroeconomic and relevant leading indicators (think price evolution of raw materials, download volumes of car-sharing apps, building permits issued, consumer search trends on new products or even common diseases) to identify patterns and predict future trends with the maximum possible accuracy and precision. This empowers our clients to be proactive rather than reactive when setting goals and strategies. It offers unparalleled analysis due to the implemented model competition that is picking the most suitable model for each variable from a set based on the data provided. In this way, the client companies could benefit from highly customasible and at the same time scalable forecasting based on the highest statistical and econometric standards.
Our Forecasting System
Imagine you are an industrial manufacturer offering millions of SKUs. Each SKU can have different specifics which need to be accounted for when deciding on the best forecasting approach. It is essential to initiate the forecasting process with exploratory data analysis to uncover which are the critical SKUs to forecast, i.e. which generate the largest share of the company’s revenue and are key to reducing the cost to serve.
What makes time-series forecasting a difficult nut to crack is the fact that there are complex seasonality and trend combinations attached to each business metric or KPI. Together with our clients, we work to understand the most relevant industry factors that affect business performance and then identify the right data sources to include in the process, building data sets where required. Our system offers seamless and dynamic integration from various external sources such as weather, mobility data, market trends, and macroeconomic data, unlike off-the-shelf tools that rarely integrate external factors in the analysis at any level of detail.
Next, we test dependence across the various products and locations. Products are often consumed jointly or are possible substitutes, which means that increased demand for one product may be associated with increased or decreased demand for another. Similarly, discovering common trends across locations allows us to ‘pool’ the data – which can be especially advantageous when historical data is limited or incomplete. Models where such cross-relationships between products or locations are explicitly taken into consideration result in improved efficiency, accuracy, and precision.
Based on the data collected, the most appropriate time-series model is then implemented in a data-driven way. The criteria based on which the model is selected are transparent, allowing more advanced users on your team to get a more detailed view of the selection process. Unlike other tools on the market, our system performs model selection across a wealth of both classical and cutting-edge statistical and machine-learning methods, guaranteeing the highest standard of performance and reliability. 10+ best-of-breed models (ranging from classic statistical models like regression and ENSEMBLE methods to open-source libraries like ARIMA and Facebook Prophet) compete in real-time by measuring model performance using statistical metrics. By incorporating these technological advancements, we empower our clients to leverage a wide range of models for their specific needs, rather than to be limited to a single option. For example, for each SKU in the product portfolio, the best-performing model is selected based on the product’s unique characteristics and forecasting needs, allowing for unlimited scalability and reduction of internal operational costs.
After developing the forecast, and evaluating its accuracy and usefulness, it is time to address the business problem properly and with the right audience. To facilitate that our system allows for user-friendly exports to report to a larger team on forecasting outputs as well as model performance in formats like PDF and PowerPoint slides. You are just one click away from generating a valuable report that enables data-driven decision-making. This feature saves a lot of operational time and effort that can be allocated to more impactful and strategic tasks, enabling team efficiency and fostering productivity.
Who could benefit from our Forecasting System?
The Forecasting System is designed to be user-friendly and accessible to a wide range of users with various levels of technical understanding and needs. It does not require extensive technical knowledge to operate, making it suitable for various internal departments, such as sales, consumer and marketing insights, performance analytics, and data analytics
By utilising the Forecasting System, internal teams can leverage the valuable insights and predictions to make informed decisions and gain a competitive edge.
- Sales and marketing teams: can predict future demand for products and services, and optimize their sales and marketing strategies accordingly
- Operations and logistics teams: can optimise production and logistics processes, to anticipate bottlenecks and plan for future capacity needs.
- Finance and accounting teams: can develop financial projections and budget plans, as well as to identify potential risks and opportunities.
- Supply chain teams: can optimise inventory levels, plan for future demand, and identify potential supply chain risks.
- Human resources teams: can anticipate future staffing needs, plan for employee turnover, and to identify potential skills gaps.
Amplify Analytix’s Forecasting in Action
Our European manufacturing customer was interested to know the impact of external macroeconomic factors on their business. The system not only has the functionality to perform forecasting using available internal data but also to incorporate the impact of external factors.
In step one, our team performed research to produce the right set of external macroeconomic factors relevant to the customer’s business. We considered as factors the price of commodities like gold, iron & other metals, unemployment rate, manufacturing PMI, customer confidence, price of crude oil etc. Connectors provisioned within the system enable easy access to external sources to pull this kind of data.
Utilising internal as well as select external factors, we were able to predict with an accuracy of 85% the lagged impact of the price of commodity on the customer’s sales. This helped our client consistently and timely to consider the changing market landscape in their business strategy. Moreover, we were able to quantify how much a unit change in each external factor would affect the sales and when allowing for accurate planning of future revenue streams.
Our forecasting system makes use of modelling innovations in data science combined with a deep understanding of business needs to frequently forecast with a great amount of flexibility. It is more versatile than currently available tools because it can seamlessly combine proprietary with external data. It can be used for inventory, production, sales, or financial forecasting and planning across a variety of industries where the environment is dynamic and companies need to plan across a variety of products, locations, and under different conditions. In conclusion, our forecasting solution offers the best of both worlds, providing unparalleled customisation on a variable level while maintaining exceptional scalability, making it the ideal choice for businesses seeking tailored precision and rapid expansion.