SCM Awareness: Introduction to Forecasting
You might think the first piece of demand we should cover is Customer Orders. Well for most businesses’ forecasts are more important. This is because the lead-time back through the supply chain is so long that you have to plan production based on a forecast of demand rather than actual customer orders. There are exceptions of course but generally forecast is the key in most companies.
So, what is a forecast? A Forecast is an estimation of future demand. There are many ways to statistically calculate the forecast. Generally speaking, you don’t need to know that. There are many forecasting tools out there that do the maths for you. Even excel will do the maths if you know how to set it up. Instead we will look at the principles underpinning the forecast and some guidelines to be aware of.
What makes up your forecast? Well numbers yes. But where do these numbers come from? In my experience they come from a wide range of places. These include
- Historical/Statistical analysis
- Are there seasonal factors
- Current Customers sales projections
- New Customers sales projections
- Past forecast accuracy/bias
- Known issues/trends outside the past experiences
- Upcoming changes in Marketing & Promotions
- What is the competition doing?
- What is the economic outlook?
- What new products are we launching?
- What products are we discontinuing?
- Are we changing price?
- What have management committed to investors
All these elements will impact on the accuracy of your forecast. The more information you have the better but likewise you can get lost in the information and have a sort of data overload. You can even be in a situation where you have conflicting sets of data. Where one piece of information tells you one thing, and another tells you something else.
One example I saw relating to this was to do with a product launch. New numbers had come in for the product launch from the customer and they were way above what had been expected. There was now a panic trying to get everything in place to meet the new numbers. But when I went through the numbers, I immediately felt they were not correct. They were way above past launches. Several times greater in fact. I phoned the salesman who had passed on the new numbers from the customer. I asked a few simple questions. How many stores was this product going into? How many stores had a couple of similar products gone into? Was there any promotion happening? Had the customer explained where the new numbers came from? Based on the answers to those questions this new spreadsheet of numbers just did not feel right. I got him to go back and ask for confirmation. And within an hour we heard back that the numbers were wrong. No explanation but we guessed that they had given us numbers for another product. The corrected forecast was in line with what we had expected. So, this is an example of where you can’t simply rely on numbers. The forecast owner in an organisation must understand the business and be able to interpret what the numbers mean.
So, what are some of the things you need to know about forecasts?
Well First of all Your forecast is wrong! Do not take it personally. Every forecast is wrong. Well I am probably exaggerating for effect. I am sure there are some businesses where the forecast is accurate. But those are very limited circumstances, and it is probably a case where demand out strips supply deliberately so the company knows everything they make will sell and they know they will only make a limited number. But this is rare. Human nature means our ability to forecast has always been exaggerated because we want the certainty that gives us. For most cases there is enough uncertainty to make your forecast wrong. You can reduce the uncertainty but no matter how much you spend you cannot ever be 100% certain of the future especially the further out you go.
Studies have shown that we can only be accurate out around 150 days in the future. I worked for a company once that supplied a major tech company. We were supplying to them based on perhaps the most detailed forecast I have ever seen. They had dozens of people just working on the forecast and they invested millions in the latest forecasting software. And still they got the numbers wrong. Some products sold better than expected. Some sold worse than expected. A celebrity using their product would boost sales. Another celebrity using a competitor’s products would hurt sales.
Several things can impact on the accuracy of the forecast.
- Winning or losing business is an obvious one.
- Promotions in the marketplace (price or advertising campaigns)
- Seasonal factors (e.g. a product sells more in the winter than in the summer).
- New products launching that may take some business from existing products.
- New products launching that may complement existing products and increase their sales.
- Order pattern from customers.
- Life cycle stage of the product
- What competitors are doing (e.g. promotions, new products, delivery issues)
- Customer trends and fashions
- Media coverage
- Innovation and new technology.
- And of course, the Experience of the forecaster. Often this is the key one. Certainly, this is where most of the blame will go if the forecast being wrong causes problems. But an experienced forecaster with the information they need, and a good understanding of the business is probably the key element to creating an accurate forecast.
Your Forecast is Biased. This is linked to the idea that your forecast is wrong. Bias is where there is a trend one way or another in accuracy. A forecast from a region may tend to be too high or too low. For example, you may have a salesperson who is naturally optimistic they may give you a forecast that is too high.
In the past I dealt with a salesperson in one country who was extremely detailed. He would include every potential customer in his forecast. I learned quickly to understand his body language and tone in conversation and know which forecasts to include or exclude. Had I just taken his numbers without a follow up discussion the company would have overproduced quite a bit of stock. You should be able to detect the historical basis by analysing past forecast vs actual. That is often included in your forecast software. You must feed that back to the region in question, but you must also factor that bias into the calculations until you are confident that it has been resolved.
Your forecast will have a trend and a pattern. It might trend upwards or downwards over time but there will be some sort of a trend to it especially over the longer term. It could be linear, or exponential (rapidly increasing), cyclical (in a wave form) or a bell curve over the entire lifetime of the product but there will be some trend to the forecast. You must know where you are in that trend.
You can often do this by looking at historical data and plotting this into the future. When you look at the data what do you see? Is it trending up or down? Does the customer tend to order every second Friday? Do orders for one product fall when another goes up. All these are patterns in the data that influence the trend. Of course, the best thing is to also speak to the customer and understand what drives the pattern.
Products will generally have a life cycle and you need to factor that into your forecast for the product. The traditional one is a slow ramp up at initial launch, followed by rapid rise once it becomes popular and then tapering off as it reaches maturity and eventually enters decline.
However, some products can have long stages of maturity and some short. For example, a semiconductor going into a new phone may be unique to that phone. It will see a sudden spike in demand up to a high level and then after a few months it will see a rapid decline as a new phone replaces that one. These are sometimes referred to as Vertical demand parts.
The opposite can also be true. Demand is regular and predictable and has been like that for many years and there is no plan to discontinue the part. In truth you are still dealing with a product with a lifecycle but the maturity phase at the peak is so long it appears more or less flat.
However, in reality the vast majority of products will follow a more typical lifecycle.
Products could have cycles within the lifecycle. While the overall trend over many years may follow the lifecycle there could be other cycles that also impact on that. These are often macroeconomic events outside the control of the company. For example, the there is an economic cycle often seen as boom and bust or recession and this will generally have an impact on sales. Economic shocks will also impact on the economic cycle. The problem is it is impossible to predict these economic shocks. Anyone who tells you they predicted events like 9-11 or the collapse of the banking sector in 2008 or Covid-19 is either a multi-millionaire from the profits or exaggerating their abilities. In your forecast you can’t predict those sorts of shocks, but you can be aware of economic trends and factor those into your forecast calculations.
There could also be Seasonality factors: This is where the forecast moves up or down depending on the season (or month etc). This is similar to cycles but is considered separately because it is a predictable change for a recurring event. For example, sales of Christmas Trees will spike at Christmas and fall to zero for the rest of the year. You need to know seasonality as if you look at an annual forecast that says you need 1200 trees and decide to evenly distribute that throughout the year you will have 1100 trees from January to November that you cannot sell and then be short 1100 trees for the December demand. So your production forecast should be to harvest some trees in late November and some in Early December and definitely none in January to mid November.
There could be Noise in your forecast data: This is Random data points (peaks or spikes in demand that have no real bearing on the forecast). For example if you shipped 200 samples to support launch in a new region then that would add “noise” to your data and should be disregarded for anything other than predicting sample requirement in another launch. These are often referred to as outliers and a collection of outliers makes up the noise.
Not all Noise is bad. Sometimes that noise is useful for comparison with similar events. Carrying on the previous example if you know you need 200 samples per region per launch then you will know to include 200 samples in the forecast for a launch in a new region.
Another good example of this is promotions. For example where you have a regular promotion twice a year in a chain of stores and they cause sales to double then you disregard that promotion volume for most of the year but use that data for future promotions.
Your forecast will have a Horizon: This is the length of time the forecast will look out. You cannot look out to infinity. It needs to be at least longer than the lead-time of the product and components and materials that go into it. However, in reality to properly plan the business it needs to be much longer than this. Master production schedule forecasts in the factory will generally be at least 12 months. The S&OP will be at least 24 months. The long-range business plan will have up to 5 years in it or even more. This is to allow long term planning as well as short term planning. That allows the business to look ahead and see if extra equipment or even factories are required all of which can take years to construct. Financing from banks etc will require plans that show the business is sound over several years. Of course, the further out the forecast the less accurate it will be particularly at a finished goods level. But this is still necessary for the long term running of the business.
Your forecast will have intervals. This is the time interval or “bucket” the forecast is presented. E.g. we will sell 10 a week or we will sell 50 a month. Intervals are normally in weeks, months, and quarters. Rarely will forecasts be in days as that level of detail will be too erratic and “noisy” especially once you go beyond a few weeks.
Therefore, forecast will generally be in Weeks for maybe 8-12 weeks. Months for the rest of the 12 month period. And quarters for the following year or two. Beyond that for the 5-year plan it may be in Yearly buckets.
Forecasts for close dates are more accurate than for far off dates. Generally, there will be patterns in demand (e.g. that customer always takes a pallet a week). However, the further out you go the harder it is to predict. In many ways it is like the weather. We have a good idea what the weather will be like for the next week to 10 days but beyond that it becomes harder to predict. In general, we can say that the weather in Summer should be better than the weather in winter, but you can’t be sure (especially if you live somewhere like Ireland). The same applies in business. The probability of something unexpected changing that for next week and you not being aware of it is slim. Even products impacted by the weather can be predicted in the short term.
For example, if the weather is forecast to be sunny and hot you might sell a lot of Barbeques. But if it is raining and cold then sales of barbeques will drop. You can use that sort of logic to predict seasonality (e.g. you will sell more Barbeques in summer than winter). BUT if it is too hot and too dry then the government might ban Barbeques to reduce the risk of forest fires. The further out into the future the more likely it is that something will change that will impact on the forecast.
Forecasts for groups or families of similar products are more likely to be accurate. An individual item may have quite volatile demand but within a group of products that volatility will tend to balance out as a spike in one may correspond with a drop in another.
For example a company that sells Polo-Shirts may find that the demand for an individual polo-shirt is quite volatile and difficult to predict but across their entire range of polo-shirts they can be more confident in the forecast. This way they can order the fabric, buttons, sewing capacity etc and have all this in place for the overall family of polo shirts without needing to know the demand for an individual size, colour etc. Therefore if they can find a way to shorten the time from getting information from shops to the factory while at the same time delaying production until the last possible moment they can react faster.
A company who did this particularly well is Zara. They operated a very lean supply chain with little inventory in the chain. But their store staff all carried handheld computers into which they would enter details of customer requests. This information was instantly sent to their central HQ and aggregated with information from all other stores. Therefore, they could see that demand for Small Red Polo Shirts with blue buttons was higher than the medium size. They could then adjust the production plan accordingly. This sort of flexibility is sometimes referred to as late differentiation where the final product is only committed to at the last possible opportunity.
Forecast should include an estimate of the error in it. As we stated earlier your forecast will be wrong. You should put an estimate of the error rate in the published forecast. There are statistical methods that will help you calculate the error or if you are using dedicated forecasting systems that system will probably calculate the error for you. Common ones are Mean Absolute Percentage Error (MAPE) and Mean Absolute Deviation (MAD). It is important to track and publish the error so that if it grows too large you can take action to rectify it and fix the forecasting process.
This needs to be tracked, communicated and investigated. Publishing the error rate also provides an additional benefit in that unfortunately some senior managers do not understand that forecasts are inaccurate. If you don’t publish an expected error level, then you will probably have to explain why it was not 100% accurate.
Forecasts need to be reviewed regularly. This is a natural extension of the concept that far off dates are less accurate, but it is important enough that it needs to be called out. A caveat to that is you need to review and amend your forecast as often as needed in your industry. For example, if you are in fast moving consumer goods you probably need to review your forecast weekly. If you are making specialised heavy equipment, then monthly or less often may be acceptable.