Der Magic Software Germany Blog

Montag, 18. Februar 2013

Forecasts ohne Integration - Ein sinnloses Unterfangen

Am 08. Februar haben wir gemeinsam mit unserem Partner H+W CONSULT ein Webinar zum Thema Forecasts und warum in diesem Bereich das Thema Integration so wichtig ist gehalten.

Unsere Redner, Herr Prof Dr Michael Capone, hat für uns nun noch einmal alle wichtigen Informationen zusammengefasst:


The goal of this short article is not to provide instructions for forecasting, but to explain why the most common forecasting methods are so inaccurate and to shed light on a forecasting method that can produce very accurate forecasts and is, therefore, growing in popularity.
Forecasting is the process of making statements about events in the future. Forecasts are not to be confused with goals or plans. These are distinctly different, albeit related. A goal is a wish, for example, “We strive to sell 3000 widgets this quarter.” A plan is a set of deliberate actions for achieving a goal, for example, “In order to sell 3000 widgets this quarter, we´ll offer a 20% discount and increase our sales activities.” A forecast is a prediction, for example,“ based on our sales for the first month, we´re predicting quarterly sales of 2.800 units.”

We are confronted with forecasts on a daily basis, because financial markets depend on them. Shareholder expectations and stock prices are linked to forecasts. A positive forecast can drive investor confidence and increase share price. A negative forecast causes stockholders to sell.
Aside from these very public forecasts, there are many other important statements we make about the future that are equally important in that they can affect financial performance. These forward looking statements can be grouped by ERP module:

  • Finance needs to forecast revenue, which is not the same as sales, as well as receivables and payables, which together make cash flow. 
  • Operations need to accurately predict machine utilization and material requirements.
  • HR seeks seek accuracy in scheduling staff, hiring new employees, and training.
  • Efficient supply chain management relies on accurate predictions regarding future inventory levels and delivery times.
  • Sales, service and support (CRM) depend on accurate forward looking statements pertaining to order and case volume. 

There are 2 forecasting methods widely used today:

  • Qualitative forecasting techniques are based on the opinion and judgment of consumers and experts. Examples of qualitative forecasting methods are informed opinion and judgment, the Delphi method, and life-cycle analogy. 
  • Quantitative forecasting models are used to make statements about the future based on historical data. Examples of quantitative forecasting methods are last period demand, simple and weighted moving averages, exponential smoothing, and seasonal indexes. 

The life-cycle analogy method is probably the most useful of the qualitative methods. It assumes that the company´s forecasts will reflect the market state. A company in a mature market will forecast single-digit expansion of sales, staff, and output whereas a company in a declining market will forecast decreases in revenue, orders, productivity. Of course, this method ignores many of the other factors that drive demand and is, therefore, not very accurate, but it does serve as a benchmark.

The quantitative approach is most commonly used today and it is largely responsible for many inaccurate forecasts. The first problem with forecasts based on historical data is selecting the proper base. The table at right shows historical data for 2 years. These figures can represent orders, personnel, units of production, inventory level, or revenue. When we calculate the change month-over-month for January, we conclude that the numbers increased 177% 2012 over 2011 and we forecast that the number will also increase 177% in January 2013. When we compare month-to-month historical data to forecast January 2013, we forecast a monthly growth rate of 6%. There are several other approaches for interpreting this historical data, and the forecasts they generate will almost always differ significantly.
Another reason the accuracy of these methods is poor is they assume the variables also remain constant, which they do not. A traveller who attempts a cross-country road trip and records an average speed of 140 km/h for the first 5 hours, cannot forecast this speed for the remainder of his trip. His future average speed will depend on his fuel supply, financial means, weather, traffic, number of drivers, physical and mental state and car condition. If the data in the table at the right represents production output, the forecast for March 2013 is only remotely related to output levels in March 2012. Production forecasts for March 2013 are more closely related to current variables like sales activities in January and February 2013. Similarly, if the data in the table at right represents historical staffing levels, the staffing requirement for June 2013 is a function of order volume for May 2013 and has little to do with staffing levels in 2011 and 2012.
An accurate forecast depends on numerous variables. These variables can be internal and external. In the example with the traveler, the internal variables are fuel supply, financial means, number of drivers, physical and mental state, and car condition. The external variables are weather, road conditions and traffic.

For a manufacturer, the forecast variables depend on the type of forecast.

  • The order volume for next month is a function of the current sales data like RFPs, quote volume, and sales cycles, as well as external factors like competitor`s activities, unemployment and interest rates. In the table at right, the sales forecast for January 2013 is not a function of the actual sales for Januar 2012, but rather a function of recent sales activities. 
  • The revenue forecast next month is a function of order volume this month, current sales data,  production output and capacity, staff levels, productivity, delivery times, and payment terms. 
  • A forecast of machine utilization next quarter depends on current order volume and sales data like quote volume and sales cycles, material delivery times, staffing.
  • A staffing forecast for next month depends on order volume, production forecasts, and external factors like unemployment.
  • A materials forecast for next quarter depends on production and staffing data and external factors like
  • weather.

Clearly, an accurate forecast relies on internal and external variables. The impact this data has is expressed as a regression formula or algorithm. If you use SFA or CRM today, you´ve seen a simple regression expressed as a relationship as various stages with probabilities: The sales forecast for the next quarter is calculated as 10% of current opportunities in the qualification phase, 20% of deals in the quote phase, plus 75% of deals in the negotiation phase. Such forecasts are based on a combination of historical relationships and current variables, but not historical data. It may seem like we´re splitting hairs here, but there is a difference.

These statements are distinctly different and accurate:

  • “Monthly sales for the past 12 months were 10 units and therefore we forecast selling 10 units nex month” 
  • “Our historical success rate is 60%, our sales cycle is 27 days, we currently have 10 open quotes for 20 units, therefore, we forecast 12 units next month.” 

Forecasting depends on dozens of internal and external variables. The internal data resides in the ERP system. Ideally, the data flows from the various ERP nodes into a Master Data Management system, where it is made available to the various departments for analysis. This is seldom the case, though. ERP is rarely a contiguous system. It is usually a network of disparate, disconnected, and dispersed systems. Gaining access to these multiple systems and the variables that affect forecasts is very difficult. And even when access is provided, collecting and then calculating forecasts in a timely manner is complex and tedious. Therefore, getting real-time or near-time data to produce accurate forecasts requires an integration with data sources.

The implementation of a real-time regression forecast is, therefore, an integration project that is implemented in steps: 

Step 1. Identify the variables that impact results.
Step 2. Identify the data sources.  
Step 3. Quantify the impact the variables have on results. To do this, you need a statistically significant set of historical variables. It will become apparent during this phase that internal data is often more difficult to access than external data. 
Step 4.  Integrate. Establishing a synchronization between CRM and ERP systems to exchange sales and manufacturing data is significantly more complex than connecting to web-services to monitor interest rates, stock prices and weather.
Step 5. Maintain. Even though most SFA tools like SFDC have a simple regression, the relationship is usually expressed as a static probability and not a moving average that changes every week to reflect 
actual performance. The probability field in SFDC is a numeric field and not a formula. The historical 
relationship between variables and performance will change frequently, because historical data grows 
daily. This phase is critical to improving the accuracy of your forecasts. There are many SFA users 
that have never updated the default probabilities for sales stages to reflect their company`s actual performance.

Companies that use regression forecasting supported by some level of real-time or near-time integration are forecasting with 90%+ accuracy. This result is not achieved in days or weeks, but a 90% forecasting accuracy can be achieved in less than a year. The advantages for the company include a more efficient allocation of resources, better cost control, reliable planning, greater agility, and the achievement of JIT management.