Demand planning in many businesses is a thankless task. Demand planning is about predicting the future and consequently there is no such thing as a ‘perfect’ demand plan. Demand planners are rarely rewarded when they get it roughly right, and are the scourge of the business when they get it very wrong!
What’s key to remember is that demand planning is not an isolated task. It’s not about a planner being locked in an office developing increasingly complex forecast algorithms and churning through years of historical transactions to come up with ‘THE’ number. A demand plan should be a cross-functional consensual view on what is likely to occur with agreed assumptions.
Many businesses struggle with demand planning, so here’s our top 10 tips on getting it right.
Don’t demand plan in isolation
Demand planning is a function of a business’s wider S&OP or IBP process. The demand plan must be a cross-functional collaboration. It requires input from sales, marketing, finance, supply chain and manufacturing; there needs to be a consensus view on what everyone thinks the future holds. If you don’t already operate an S&OP or IBP platform, then you should consider the implementation of such a platform a priority.
Lies, damn lies and statistics
Many businesses have a false belief that there’s a perfect algorithm that will brilliantly forecast their sales. We’re here to tell you there isn’t. There are a multitude of statistical forecasting methods available to the demand planner, from simple linear regression through to Holts exponential smoothing; but what every algorithm has in common is it will never be ‘perfect’.
Use statistics only as a guide
Statistical forecasting should only be used to give the baseline which can then be manually modified to account for market promotions, market intelligence, customer forecasts, manufacturing constraints etc. Using statistics as a baseline avoids the need for zero based forecasting each month, which nobody wants to do.
Staying on the statistical forecast theme, complex algorithms have their place, but very often using simple growth factors applied to the same period last year, or the previous x months of sales can be just as effective. Furthermore, the wider business can better understand and more likely engage with simpler methods of forecasting the baseline.
Okay, so this may sound like a contradiction to the above, but you need to remember that one method of forecasting may not be right for your entire product portfolio. Using a straight line trend from recent sales for a ‘bread and butter’ product may work, but it won’t work for a seasonal product or a product at the start or end of a product life-cycle. Remember to adapt the methods you use for products with different demand profiles.
Your customers aren’t perfect either!
There’s been lots of talk in the last decade on supply chain collaboration and customers sharing their forecasts with their suppliers. This is indeed happening more and more, but the reality is that your customers are also never going to be in the position to supply you with the perfect forecast. Use their forecasts as a guide only, don’t depend on them unless you have a contractual agreement on delivering (and being paid!) to their forecast without exception.
Don’t forecast your own mistakes
When you’re forecasting using historical transactional data, remember that the demand pattern of previous sales may reflect errors you made in previous forecasts. A customer may have wanted a 1000 units on a Monday, but you hadn’t predicted that level of demand so you could only supply 500 units on the Monday, and the rest the following week. Consequently, you need to ensure that you don’t forecast this order split as it isn’t reflective of the customer’s requirements.
Launder your data
Historical data will always be full of exceptions. There may be project based sales, or sales may have been constrained due to manufacturing issues, holidays, supplier delays etc. In essence, and as previously said, your historical sales transactions may not always be a reflection of what your customers wanted and when they actually wanted it. Wherever possible try to flag and remove these exceptions. If it’s too difficult to flag all the exceptions a good rule of thumb for excluding outliers is highlighting anything more than 1.5 times the interquartile range above the 3rd quartile.
There isn’t one number…
Any good demand plan will consider upsides and downsides. Demand planning shouldn’t be about arriving at a single number; it should be about considering the range of possibilities. A business with a good demand planning function will consider what’s the worst case scenario and what’s the best case scenario, and how each business function will manage those scenarios.
Measure and manage
Measure and monitor the deviation between your actual sales and your forecast. We recommend using one single simple measure such as mean absolute deviation (MAD), or mean absolute percentage error (MAPE). Monitor these measures every month and just like any other business KPI, try to identify root cause issues and aim to continuously improve forecast accuracy.