Exclusive portal designed for our Customers and Partners

Log in

Request for Access

countries flags

Blog post series: Plan your predictive analytics and machine learning journey with CCH Tagetik - Part 2

Jun. 26 2020 by Prof. Dr. Karsten Oehler, Solution Architect - CCH Tagetik DACH / Marco Van der Kooij, Managing Director - ForSight Consulting

Performance Management Budgeting & Planning

We’re often asked “how can machine learning and predictive analytics support Finance?” and the answer is that there are different areas where AI and ML could make the difference In the first blog of the series we have talked about data quality improvement and data collection (link to the blog), in this blog we will focus on a second area: Profit and Loss forecasting.

To predict sales and costs accurately is essential to have a good enterprise planning system. Sales forecasting is usually the first step in any strategic planning, business planning, budgeting and forecasting process, if you want to ensure all subsequent steps are based on realistic sales predictions. Forecasting is also necessary for internal aspects of the P&L. But to gain better understanding of future costs is usually a time-consuming task: resource planning involves assessment of consumption and forecasting of factor prices.

Even in an era of machine learning, forecasts are often performed by somewhat simplistic extrapolations or manual data entry. Companies which rely on this method of forecasting often face problems such as:

  • Forecast inaccuracy: usually measured by indicators like mean average percentage error (MAPE). The fallout is planning problems like inventory shortage and capacity shortage. Even small problems in volume forecasting accuracy can lead to significant profit and cash variance.
  • Manual forecasting: The major effort of manual forecasting because experts in specific areas are required to provide future assessments. The results of these must then be accurately consolidated. Manual data tends to be biased and therefore needs adjustment to be realistic.
  • Lack of transparency: Known or unknown drivers which aren’t accounted for when making predictions limit forecasting ability. Specifically, in cost forecasting, resource needs are usually extrapolated without knowing whether consumption is really necessary, due to unknown drivers. These so-called “black holes“ in overhead costing are more or less unmanageable due to limited transparency. Trials to enhance transparency usually require a huge effort in planning analytics. Each cost type has to be analyzed, identifying potential drivers.
  • Limited support: There is usually limited support for assessing sales increases or cost management activities and unrealistic simulations are often the result.

Machine learning makes it fresh

Time series forecasting is a well investigated topic. However, with machine learning, time series analysis is attracting fresh interest. Things have moved on: first, modern tools simplify usage; second, by capturing more data, it is possible to consider a broader range of drivers (revenue and cost forecasting can be improved by using external drivers like economic climate); and third, method restrictions like linearity of driver relations can be overcome.

A broad range of companies can use time series-based forecasting although some preconditions should be fulfilled:

  • A certain amount of historical data must be available
  • Volatility shouldn’t be too high unless it is related to meaningful seasonality
  • Companies with a significant percentage of overhead costs and a history of operational data, which contains potential drivers, can combine them with already available driver structures, like bills of materials, work breakdown structures, routings, overhead, and analysis, etc

Using time series forecasting offers real benefits in the light of an integrated process. For an over-all assessment of the profit situation, sales and operations planning is needed. As a starting point, sales volumes (predicted) must be linked to necessary resource consumption using bills of materials, work routing or any other available structure. Possible constraints must be considered, but indirect or hard to measure effects must also be assessed. Costs resulting from complexity for example: How does the number of product variations affect purchasing costs, planning costs etc? Information about driver dependencies can be used as a starting point to assess costs. Machine learning methods can identify impact and possible time lags. Together with the price, accurate costs can be forecast.

Predictive forecasting in financial planning

To use predictive forecasting in planning processes, it should be an integrated part of the planning process and easy to set up. Little technical and statistical knowledge should be necessary. Methods like ARIMA (auto regression integrated moving average), vector auto regression (statistical), random forest or neural nets can be used to predict the future on different product or customer-hierarchical levels.
To provide sufficient flexibility, forecast methods should be selected directly from planning forms and also be applied in batch processing. The system should check whether a method is appropriate and suggest the right input driver and parameters. Measures of significance and band-widths of confidence should be provided to express the forecast quality. The results of the forecast can be directly used in the preceding planning steps.

Multiple positives from automated forecast planning

The outcomes of an integrated and automated forecast are multiple:

  • Accuracy in sales forecasting produces a better foundation for planning; increased accuracy of cost forecasting, particularly for overheads, creates realistic planning assumptions.
  • The cost of manual forecasting can be reduced by automation.
  • The learning effect should not be underestimated. Better understanding of sales drivers, such as discounts and marketing expenses (including time lags), can be used for simulation, activity planning, variance analysis, etc. At the same time, increased transparency of costs and drivers makes for proactive cost management. Methods like activity based costing can be supported to manage overhead costs.
  • Revealing of confidence corridors helps assessment of the inherent risk of the business, since forecast numbers are uncertain, even with the best forecasting method. This is a good foundation for a risk analysis (which we’ll describe in an upcoming blog) and helps to calculate value at risk resulting from forecast errors. 

By supporting not only the finance aspects of planning but also focusing on sales and operation planning, CCH Tagetik provides the right foundation for an integrated forecast and planning process. Enhanced by predictive forecasting functions, the overall process can be streamlined and automated. Using the drivers detected by predictions increases transparency, provides an accurate profit estimation, and enables meaningful what-if-scenarios.

To discover more download the white paper "Machine Learning for Controllers. Use cases for Forecasting, Planning, and Simulation" here.

Share this post!