Here is a short video on demand forecasting using history. Otherwise known as Time Series Forecasting. We take a look at the demand history for a few products and see what the history can tell us about the future.
There is a fair bit of ideology around forecasting in the manufacturing business. On the one hand:
You don’t need the forecast, just become demand-driven and improve your response to the customer
or, on the other side:
The future is AI and machine learning. Algorithms will tell us what we need to know
Or how about this one:
History is not an accurate guide to the future.
This last one certainly has some truth to it. However, I ask have two responses:
- Define “accurate”. Accuracy is not binary, it is obvious that the history will neither be 100% accurate or 0% accurate. Demand history won’t tell us everything about the future, but it will tell us some things.
- Forecasting is not just about history. There is so much more to forecasting than the history. But history is a good place to start.
Why start with history? Three main reasons: First, the data is readily available. All we need is one table, a demand history, with three columns: Item, period and qty.
Second, the methods are available and proven. It doesn’t require proprietary mathematics or licencing some fancy technology. (Although a surprising number of forecasting software programmes just apply exponential smoothing and make out like it is an exclusive technology.)
Third, you can measure, learn and improve. And you don’t have to wait for a year to collect data about your forecast performance because we can run a model on the history and calculate what the forecast would have been for those periods and then compare with the actual.
So, this video is designed to run through the demand history of some examples and see what we can learn from the history. Before we go to machine learning and predictive analytics, there is much to do with relatively simple techniques to create a baseline forecast. Once we have a baseline, we can measure accuracy and then use this accuracy performance to evaluate the alternatives.
We show four alternative methods for forecasting with an example data set in an Excel forecasting tool:
- Moving Average: Very simple and often a good starting-point for a baseline. You will be surprised how many times a moving average beats a manual forecast based on judgment and hundreds of man-hours every month. This is perhaps the equivalent of a phenomenon in investing where a random selection of stocks beating the average fund manager performance!
- Simple Exponential Smoothing: Similar to a moving average in that it is capturing the level of demand. But often better as it puts more weighting to the more recent periods which have more relevance to the future than the oldest periods.
- Double Exponential Smoothing: Uses SES above, but also looks at the trend and can project the trend into the future.
- Triple Exponential Smoothing: Uses DES above, but also looks at repeating patterns, most commonly seasonality.
This shows the three things that history can tell us about the future: The level of demand, the trend and repeating patterns.
The Excel tool that is used in the video is part of our Demand Forecasting training. This training is joint effort with Nicolas Vandeput, a data scientist and specialist in demand forecasting and inventory optimisation. We will be going though all the theory behind these forecasting methods, plus more importantly, how to develop a demand planning process.
This process will be necessary to select and optimise the automated forecasting methods, choose which items have the time put in for manual judgment and then measure to ensure that finite resources are spent in the areas where there is a maximum result.
My job is to provide all participants with actionable learning and Excel tools that can be used to implement and deploy it straight away in the business. We combine the theory, best practices and hands-on advice to make forecasting part of a successful demand planning process.
Here is some more information about the Demand Forecasting Training. The June 2022 class is open for registration now (as of 9 May 2022) and we limit participation to only 12 people. This ensures that we have the right amount of attention to each person and their own situation. The answer to most questions about forecasting is: “It depends!” and we need to take the context and your situation into account in providing the right answer.
There will be approximately 1100 people reading this post every week so the class will fill up soon. If anyone has questions, the best way to get them answered quickly is to click this link and book a short meeting with me. If you are serious about registering in the training, let me know in the meeting comments and I may hold a space for you until the meeting.
For a deeper dive into the theory and best practices for forecasting, here are two webinars that I recently did with Nicolas Vandeput: