Six Sigma – iSixSigma Forums Old Forums General A FORECASTING MODEL

Viewing 10 posts - 1 through 10 (of 10 total)
  • Author
  • #32191


    Hello All,
    I just need some help to build a simple forecasting model in Excel spreadsheet and if anyone has an idea, please can you share it?
    I dont know too much about it, so I want something really simple and easy to understand but it should explained in details. Thanks!



    your question is not clear to be understood!. what you want to predic?  did you mean that you want to make a model to do the forecasting ?



    You are correct! I need to help to build a Model to forecast…Please assist. Thanks!



    Hello Corey,
    There are many types of forecasting models that exist as designated software… ie Forecast Pro….  Note: software is as good as the data being called upon…so if outliars are not scrubbed…junk in = junk out….
    Are you forcasting future sales on existing product lines?  or  Are you forecasting future sales on products that will be rolled out onto the marketplace?  This is a big difference in the dataset that would need to be compiled to derive best possible forecast…follow?


    Six Sigma Saviour

    Hi, I’m interested in your comment about removing outliers.  Just wanted to add a caveat to that thought. You can only ever remove an outlier in forecasting if:
    *It was a long time ago*It can be explained.
    You can never just remove points just to make your model work better!
    I remember forecasting working so much better in university- I guess the lecturers must have taken the time to find time series that you could actually model nicely.  Since ive been trying to apply forecasting to my projects, it has been of very limited use. Anyone else been getting better success?



    I agree with the 2 pts:   A data point (outliar) must be expalined… In term of forecasting, one must be able to expalin all the ins and outs of a warehouse transaction….that could register in a dataset being called upon to use in a firecasting model….
    With that said, does the data collection filter types of outliars which need to be defined upfront in an operational definition(or automated into program logic):
    Example: If a recently introduced product say January 03, had 2 transaction of sale movement(demand) to external customers…. one would consider this as valid demand for future forecasting
    BUT what if 1 of the 2 customers made a return in full amount originally ordered…  this datapoint of a sale(demand) was actually washed due to the customer return… so if you do not filter this sale(demand), then you may drive the warehouse to order on production for demand that is not going to surface…


    Michael Webb

    If you are trying to predict sales (future orders), it is important to remember that what drives these is probably outside the four walls of your plant or office.
    Consider how you forecast shipments in the plant: you have hard data on the capacity of various work centers to complete their work. You can then compute the workloads across the facility to forecast shipments, right?
    What if you tried to do that shipment forecast based strictly on statistical extrapolations from previous shipment activity and nothing else (no workcenter capacity, open orders or WIP, bills of materials, schedules, or other operational data) ? How accurate and reliable would your shipment forecast be? Not very.
    The same applies to sales. To create a forecast model that has a hope of being more accurate, you must base it on the activities which are driving the output. You need measurements of the marketing and sales activities which can indicate their capacity to produce the output (orders) you want.  
    Unfortunately, traditional sales forecasting is usually based on guesses and gut feelings, not structured measurements, so it hasn’t worked well. It is a battle that has been raging within companies for a long time.  
    What is needed to truly increase the accuracy of the sales forecast is valid process definitions and appropriate measurements over time which reveal the capacity of those “work centers” to produce the desired output. 
    By the way, someone is likely to say this won’t work because as soon as you get everything all set up something in the market will change, so the process yields will decrease (or increase). But this is like saying that you shouldn’t use the MRP system in the plant because sometimes you get a batch of bad raw material that reduces machine yields. You are much better off with the information – even if it is bad – so you have time to take corrective action.
    Sales organizations sometimes focus more on generating the adrenaline needed for superhuman efforts to create results than they are on scientifically analyzing what is really happening in their market place. But, that is starting to change.
    And, as is always the case in the beginning of something new, a little bit of change can go a long way. So it might not necessarily require at total re-engineering of your sales force. Just see if you can get some process data on sales activity: how many quotes or proposals are being generated for the product? Are they being proposed to new or repeat customers? Can you somehow measure some attributes of those opportunities, such as how well qualified they are? I believe these are the things which will lead to incremental improvement in forecast accuracy.  
    If you know of anyone doing these kinds of things already, please let me know! I would love to talk with them.
    Michael J WebbSales Performance Consultants, Inc.


    Michael Webb

    I have learned about a new resource that might help you with forecasting:
    Sales Forecasting, A New Approach, by Thomas F. Wallace. ISBN 0-9674884-1-9
    This book, published in 2002 has been selected by APICS – The Educational Society for Resource Management as a reference text for their certification exams. It offers powerful approaches to forecasting:
    – Emphasise team work, not formulas
    – Forecast less, not more
    – Focus on Process Improvement, Not Forecast Accuracy
    It certainly will turn your head to some productive and powerful approaches that have a very credible  track record.
    Hope that helps somewhat.
    Michael J. WebbSales Performance Consultants, Inc.



    Corey, anonymous,
    There is a way to create dedicated diagnosis filters for your forecasts. Those work well on noisy (= unpredictable) input data. Diagnosis and prognosis are very closely related items, aren’t they?
    The diagnosis is effective, easy to understand, requires some dilligence to set up and is easy to run afterwards, especially in Excel. It’s not fuzzy-logic, by the way. Let’s discuss in more detail what you really need.
    Please contact me at:

    [email protected]



    While in GE I was doing this type of approach for forecausting of sales volumes. We were using ratio of won/lost quotes as output and different levels of segmentations – Country, Market, Business size etc,. Another dimension used was –   New Customer/new Product- Existing Customer/New Product -Existing Cust/Existing Products growth. Then the last one was split into Share Shift and  Business Growth.
    Won/lost quote info gives you insights into market elasticity.  Taking it by segments makes the final model more accurate – and the more segmetation you use the more accurate it becomes. For  a different subject – price variation studies – in certain GLM models we were able to explain up to 70% of total price variations using various segments. But as downside of  higher segmentation – you have to monitor a lot of parameters to update your model and it it might become unmanageable.
    And there is another but – you are absolutely right by saying you can’t just take your in-house activity and extrapolate them into your sales. There are lots of factors outside your office  – and outside your control either.

Viewing 10 posts - 1 through 10 (of 10 total)

The forum ‘General’ is closed to new topics and replies.