- Lean Manufacturing
- Lean Service
- Case Studies
- About Us
- Contact Us
Knowing What to Make and When to Make it – The Role of Forecasting in Lean Enterprise
Knowing the future has been a human fascination since the beginning of time. In business, information about the future is essential for decision making. However, as we all know humans cannot see in to the future. The best we can do is make an estimate of the most likely outcomes in the future based on a set of assumptions. This is called a forecast and there is a virtual mountain of literature written on the subject of business forecasting. I am therefore going to try to provide a brief overview of our experience in developing a forecast.
Why do you Need a Forecast?
In an ideal world everything would be available “Just in Time” In this world the customer would place an order and all the materials labour and machinery would be available instantly to produce this order. In reality this is not possible. Supply of materials can take days, weeks or even months, hiring and training people can take weeks and months and purchasing machinery and factory facilities can take months or even years. Therefore we have to commit to obtaining materials, labour and assets well in advance of needing them. To do this, we need to make some estimate about how much of these resources we need and this is where a forecast comes in.
Where to Start
The best predictor of what your business will sell in the future is usually what it has sold in the past. So the first step in creating a forecast is usually to obtain the sales history for your business. For high level planning of your operation, monthly sales history is ideal. In our experience it is sufficient to forecast for 12 months at monthly intervals. This will give you an indication of seasonality through the year (eg is summer busier than winter?) and other trends.
Monthly data is sufficient, because the decisions that we are going to make with this data are about assets, people and capacity. These are medium term decisions and not week-to-week or day to day decisions. In addition the shorter the time interval (or time bucket) that we choose, the less accurate the forecast will be. This is because of the averaging affect we get as we aggregate the data from days to weeks to months (also know by statisticians as the Central Limit Theorem)
The same applies to the level of product mix data we examine. For example, if we were running an automotive manufacturing business it would be easiest to forecast the total number of vehicles (units) that we expect to produce in the plant. Predicting the model mix of those units between sedan, wagon, hatchback, SUV and engine type is more difficult, but probably still achievable. Once we start trying to split the units up by model, options, paint colour, trim colour etc. the data becomes very complex indeed. As the data becomes more complex, history becomes a poorer and poorer predictor of the future. Therefore we need to look at data at the highest level that is useful to us – usually at volume per value stream in lean terms.
Sales History is Fine, but we Don’t Expect to Sell the Same as Last Year?
Of course we are not going to make exactly the same products in the coming year as we made last year.
In this case we need to identify trends in our business and the changes or exceptions that we know will occur in the coming year that will lead to the sales of units being different. For example:
- What has your historical annual growth rate been? Is there any reason it will be different next year. If not, apply the historical growth rate to last year’s data.
- What seasonal trends did we see last year (e.g. soft drink sales are much higher in summer). Was last year a normal season or should we consider more seasons of history to gain an average. There are lots of fancy statistical and software tools to identify trends (eg. Regression analysis). Use them with extreme caution. I prefer to apply a “common sense test” by talking to a wide range of stakeholders to validate my assumptions about trends.
- Were there any demand spikes (e.g. promotions, one off factors), that occurred last year that will not occur this year. Adjust them out.
- Likewise are we expecting any one-off demand spikes this year that did not apply next year. How much are we expected to sell during these events? Factor this in.
- Are there products that we sold last year that will not be sold last year? If so will they be phased out completely or replaced by new products? Will the sales of the new products be the same, more or less than the old product? If the sales are changing, why are they changing, what is this assumption based on?
As you can see, if you try and do this at the lowest product level, it is a massive and complex task, however if you aggregate the demand by value stream it becomes a lot more manageable.
As a rule we recommend change by exception. In other words, if there is no reason to say that sales this year will be different to last year then the forecast should be for a repeat of last year.
At this stage we can build a simple table of sales by product and month for 12 months.
What is if it is a New Factory?
What if there is no sales history to go on and the product is new? Generating an accurate forecast is going to be more difficult in these circumstances. Again stick to what you know. If you have done your strategic marketing properly, you will know the size of the target market, and how much market share you can reasonably achieve. You will have identified target customers and know how much business you are likely to win and your chance of success. You can also estimate the timing of those wins and the likely ramp up of sales (this is actually something you can usually negotiate with your new customers).
Project Based Businesses
For businesses involved in large-scale projects, such as infrastructure or major capital equipment, sales history can be less relevant. Fortunately the nature and long lead time of these means that good information is usually provided by your customer on future demand well in advance (in some case months or years in advance). This data then needs to be aggregated for all projects and then turned in to a monthly forecast in the same way that sales history is in order to (as far as possible) level the monthly demand on your factory. Avoid the temptation to launch the whole project in to production at once (unless your deadline requires it) or leave starting the project until the last minute. If you allow a single project to consume 100% of your production capacity, then you make it very hard for your other customers and run the risk of having a sales “gap” at the end of the project. Far better to start early and work on the project over a longer period of time allowing capacity to take on other work as it is arrives. This does have cash flow implications, but if possible, try to negotiate terms with your customer and suppliers that will allow you achieve this.
Try Sensitivity Analysis
It is a good idea to do some contingency planning. Ask some hypothetical questions.
What is the highest level of sales that you realistically might expect to achieve? Under what circumstances would you be likely to achieve those sales (check your assumptions). What would the sales forecast look under these assumptions?
What is the realistic worst-case scenario for sales. Under what circumstances would you be likely to achieve this? What would the sales forecast look like in this scenario?
Making Sense of the Forecast
After having driven your sales and marketing team crazy pushing them for accurate data and challenging their every assumption, you will now hopefully have a clear an concise picture of your expected future demand over the coming year. So what do next?
From the forecast you can calculate the key requirements for your manufacturing process including your required production rate (or takt time) and your expected demand on key suppliers. From comparing the required rate (takt time) with the rate your plant can achieve (cycle time) you can determine the production hours you need to run. In other words if Takt time exceeds cycle time, you can probably afford to reduce your production hours and thereby reduce the takt time, whereas if cycle time exceeds takt time you may need to increase production hours (more people) or increase capacity (more assets) or both to ensure that you can meet your expected customer demand.
Beyond the Forecast – Sales and Operations Planning
Your initial forecast may provide a good start in planning your operations and supply chain. However things change over time. Therefore forecasts need to be updated. More frequent updates mean, over time, a more accurate forecast as you are able to check and adjust your forecast against real outcomes and new information.
A monthly Sales and Operations Planning Process (S&OP) is a highly effective way to keep your forecast up to date and make sure that your business is able to meet its future customer requirements. Again there is a huge literature on S&OP, but it essentially involves three main steps:
- A Demand Review: Reviewing up updating the monthly rolling forecast. This means updating the month just gone with actual data, adding a new month at the end of the forecast period and making any adjustments based on new information. The outcome of this review will be an updated forecast and a list of assumptions behind the forecast.
- A Supply Review: This involves evaluating the impact of changes to the forecast on the supply chain including manufacturing capacity and suppliers. The output of this review will usually be some recommendations on labour, materials and assets that are needed to ensure that the forecast demand can be met.
- An S&OP meeting: This is a senior leadership meeting that reviews the previous months performance and signs off the forecast and approves any investments or other changes necessary to meet the forecast demand.
By regularly reviewing and updating demand and supply in this way the business is better equipped to deliver to customers effectively and at a competitive cost. Surprises will still occur, but by at least having this discussion about expected demand and supply the surprises are likely to be more manageable and have less impact than if no planning had been done.