Liquidity Forecasting With Machine Learning

For finance departments, liquidity planning is one of the main tasks essential for the company’s success. So far, huge Excel lists have been set up for this, and various scenarios have been calculated based on formulas. But companies want to know a lot more: In the automotive sector, predictive analytics plays a decisive role in the liquidity forecast—the keyword: data science.

Liquidity management pursues the goal of identifying financing requirements at an early stage and coordinating financial planning accordingly. All available information on current and future business transactions is collected for the liquidity forward. This also includes identifying various risk factors and taking into account possible scenarios accordingly. The automotive industry is a good example of this. This is because many business transactions are covered in this industry – from imports to sales, leasing, service, repairs, used cars, fleets, parking, and mobility services to export.

The automotive industry is increasing investments in Big Data & Analytics year after year, as a study by BearingPoint recently revealed. The economic benefits are obvious. In particular, in the area of ​​research and development and quality and warranty management, the respondents estimate potential savings of over 20 percent. Because predictive maintenance, i.e., predictive maintenance, promises decisive advantages for liquidity planning. However, it is no longer enough to rely on Excel spreadsheets alone to develop the full potential. The departments must be able to work together across the board. Innovative automotive companies rely on a community of data scientists and citizen data scientists who jointly implement machine learning projects. Together with the controlling departments,

Democratization Of Machine Learning: The Data Community

For the collaboration between data scientists and technical experts without data science know-how to succeed, the machine learning projects must be democratized. This includes building a data community: an interdisciplinary team of experts from IT architecture, business intelligence, IT, and data science, for example. They all have to come together on a common platform. It is necessary to implement a collaborative data science platform that enables joint project collaboration between business users, data scientists, and business analysts. In short, whether or not they know about machine learning, everyone should be able to use the tool.

From The Excel Spreadsheet To The Automated Machine Learning Model

The liquidity forecast is followed by a process that takes eight weeks and usually starts with the order for a car. The cash outflow is then predicted based on incoming invoices. The cash inflow is only forecast after delivery and sale. The difference between the expected outflow and inflow results in the weekly need for monetary funds, which ultimately determines whether and to what extent further funds must be obtained to secure liquidity. The previous forecasting model should be replaced by automated forecasting based on a machine learning model to calculate this demand better.

So far, the controlling department has usually compiled the forecasting model manually from Excel data and prepares it every month. However, this takes a lot of time and resources. In addition, the results are often imprecise, and scenarios are difficult to predict. With a data community and the right collaborative data science platform, forecast preparation times can be reduced from five to one day per week – by up to 80 percent. At the same time, the costs of the forecast can be reduced, and the accuracy increased. Innovative platforms can provide the liquidity forecast analyzes interactively.

Automated Liquidity Forecasting

For the machine learning project, data must first be collected from various data sources. This includes information from the system for order processing and the trading system and data from the manufacturers regarding production planning and special features of the production locations. This historical data is then integrated into the platform and processed automatically, with innovative tools to implement the forecasting model directly in the software.

The model then calculates the weekly forecast, for example, on Sunday evening. The model’s output is then made available to control via a dashboard directly in the platform – where the model can be validated again based on human experience and adapted if necessary. The model then goes into the backtest with real data and delivers the first results. The average errors per week are gradually being reduced using machine learning. This makes it possible to reduce the rate to up to 1 percent in the relevant forecasting period.

With the implementation, the forecasting model usually creates great added value six months after the start of the project: the automotive company can directly deduce from the forecasting model whether there is a need for action: If the need for liquidity is less than the liquidity buffer, nothing needs to be done – otherwise, action must be taken. Specifically, the added value of the model lies in the fact that the automation enables higher-frequency tracking of liquidity to be achieved. This is particularly important for risk management: Less working capital is required with the same liquidity risk. The forecast cash outflows deviate significantly less from the actual data due to the automated forecast. This enables new quality standards in the forecast accuracy of liquidity forecasts. In addition to the qualitative gains,

Interaction Between Man And Model

The financial sector can benefit enormously from the new analysis options. However, the numbers are essential for business success. That is why it is important for controlling companies to have the freedom to adapt certain data. This plays an important role, especially in unpredictable cases, which the model can hardly perceive: An example would be the corona-related bottleneck in electrical components, which the controller knows about from its human expertise – a fact that the model does not know. To counteract such effects, if necessary, care should be taken when choosing the collaborative data science platform that the department has the opportunity to intervene.

Also Read: “Machine Learning” Has Arrived In Accounting

More articles

Latest article