Poor forecasting or demand planning can damage your business
There has never been greater pressure on businesses to forecast quickly and accurately, as companies seek to reduce work in progress (WIP), optimise value streams and drive profit through internal efficiencies such as reducing their working capital.
Most organisations need to reduce inventory or stock held against projected customer demand yet maintain the agreed customer Service levels. Accurate and quick Forecasting is therefore an obviously low cost way to achieve this.
For many businesses however inaccurate and late forecasting is severely damaging their profitability and threatening their future survival. Please find below MXI's top tips for ensuring your business does not fall victim to poor forecasting methodology.
1. Excel for forecasting - The potential pitfalls
Many businesses still use Excel and a time series sheet of data that needs forecasting. The trend line feature in Excel is the least effective of over 20 techniques for forecasting because of the value it applies to older time series data, ignoring the more valuable trending in recent data. Not ideal for managing the recession in consumer spending!
Better to spend a little and use a proper exponential smoother. Organisations looking to make progress on Lean Programmes to improve customer services, MRP results, supply chain issues, reduce working capital (cash) should considering using something other than Excel.
2. Why and when to use the Exponential Method
Exponential smoothing recognises the value trend, extracts the surrounding noise in data and can apply seasonality, trend and special events. Statistical results show the exponential method to be one of the most reliable. It is good where limited time series data exists.
3. When can you trust your data?
The answer is it takes time, trial and error and once your solution is implemented take the time to learn your data and get to know the results. Although be prepared to challenge the results. Consider separating the data into 80 / 20 paretto analysis. Splitting some data into product/family (aggregating) levels and some at the SKU level can also be effective.
4. Don't Automate too much.
Data needs checking all the time to improve its relevancy, even processes that are comfortably bedded need validation. Best practice and Industry leading performance can only be enhanced through regular monitoring and intervention.
5. Be aware of over and under estimation and their impact on recession forecasting.
Forecasting Software algorithms (Box Jenkins, Exponential Smoothing, ARIMA models) will tend to overstate an upswing and the downswing of a recessionary market. Dr Sam O. Sugiyama ( a shining light in forecasting academia) re-iterated his belief, at the recent forecasting conference in Boston, that many forecasting methods over-project downturns and upturns similarly. The benefit here of automated forecasting software is that the Planner has more time to fine tune a forecast. Human experience and intervention heightens the accuracy.
6. Special Events and Automatic Forecasting.
It's possible to capture previous events and assert those patterns into a forecast. However human intervention and local expertise will always play a part here. One of the products that MXI recommends,Forecast Pro, is particular strong in this area for a desktop product.
7. Top -Down Planning generally better.
Resist the urge to plan from the ground up. Most organisations will make a leap forward in planning and forecasting performance when planning at the aggregate level. Item / SKU level planning which is then aggregated to product or family tends to be more flawed owing to the noise in the data at SKU / Item level. Gaps in the time series, data integrity issues and interruptions will disturb the forecast more than the reverse approach.
8. How much data do I need for good forecasting?
In the ideal world on monthly data you need around 48 - 60 months of data. Seasonality and trends can be seen in the earlier months with forecasting accuracy being assessed against the more recent data. So the first 36 months of data is used to fit the model and establish base lines and the most recent 12 months to calibrate accuracy. In this example meeting the lean manufacturing targets or leaner supply chain targets, or reducing the working capital in your organisation will not be improved by using more than 60 months of data.
We hope that you find this information useful, should you be interested in our preferred product Forecast Pro and how it can help your business, please do not hesitate to contact us.