Seasonality can have a huge impact on demand patterns for certain products. On the one hand, if not properly anticipated, it can expose a company to the risk of stockouts, which can lead to customer dissatisfaction. Similarly, without good visibility of the demand changes that seasonality can cause, companies risk accumulating excessive levels of stock, which will require tying up resources that would be better invested elsewhere, not to mention the huge financial impact of obsolete stock that will need to be discarded at the end of the season.
By adapting forecasts to take seasonality into account, you will be able to respond to these changes in demand in a timely manner. This will allow you to hold optimal stock levels before, during and after the seasonal peak, in order to keep inventory costs to a minimum while still guaranteeing a high level of service.
To achieve this there are a number of factors to be taken into account. For example, has seasonality caused changes in demand in the past? How can you forecast seasonal demand patterns for new items? Does seasonality impact on groups of products or just individual items? Is this impact only local, or is it national or even international?
As you can tell, there are many variables to consider, and it is for this reason that Artificial Intelligence (AI) is very useful technology when it comes to predicting seasonal demand. Throughout this article we will review the main elements that need to be taken into consideration in order to detect the best way to predict seasonal demand. Likewise, we will also delve into the potential of AI together with Machine Learning in this field.
Demand forecasting and the ordering process in an ideal world
Imagine you have 150 TVs in stock, with an average sales rate of 100 units per period, and a minimum order quantity (MOQ) of 20 units. The lead time is 7 days and since you don’t want to disappoint your customers with potential breakages, you have a safety stock of 50 units. When is the right time to order and how many should you order?
The theory of the ordering process
The most logical way to calculate your needs is as follows: on average you sell 100 units per season, which equals 25 per week. In 4 weeks you would be forced to “pull” your safety stock. So, to avoid this happening, you decide to place your order in 3 weeks’ time.
Since you will be expecting to sell another 25 units in week 4, your order doubles the MOQ. That is, you decide to order 40 new TVs in week 3.
The ordering process in real life
The situation described above is easy to predict. However, it is unrealistic. Demand patterns are often not as stable as in this example. What if demand is subject to a trend? Will you have enough stock if there is an unexpected peak in demand? How can you deal with changes caused by seasonality?
Understanding seasonal demand
During the warmer months of the year the demand for winter tyres is lower, while the demand for barbecues, for example, falls in winter. You don’t have to be a genius to realise this; it is well known that these items have a seasonal pattern. However, we do not know exactly when the season will start because it depends, in part, on the weather. For instance, on a sunny day in March, when it is only 17 degrees, there may still be plenty of people who want to have a barbecue because they are fed up with feeling the winter blues, and they believe that cooking and eating outdoors is far more fun. However, 3 months later, those 17 degrees will be perceived as cold.
As a result of this behaviour, you need to know both the seasonal influence on demand and the short-term demand profile. For example, if the temperature reaches 27 degrees Celsius at the beginning of May, there is likely to be a boom in sales. But if it drops to 17 degrees in August, this change is likely to have a less significant impact as it takes place at the end of the season. Moreover, with the end of the summer in sight, uncertainty about the weather is understandably greater.
When does the season really end? Let’s take barbecues again as an example. When should we remove the barbecue meat range from the shelves? Do you want to save costs by removing these products early, or do you risk keeping them longer (resulting in excessive stock levels) in order to avoid disappointing customers?
In order to ensure a high level of service and low inventory costs during the season, it is very important to introduce and withdraw products in a timely manner.
Best practices to ensure optimal stocking before, during and after the season
1. Improve the accuracy of your forecasts by including more external information in both short- and long-term forecasts and clean up historical data from events and promotions. You can then calculate the variance and, based on it, create your forecasted demand.
2. Develop a specific inventory strategy, such as increasing your initial stock to be able to cope with unexpected peak demand in the near future.
3. Review your Minimum Order Quantity (MOQ) and Economic Order Quantity (EOQ). Do you need to increase or decrease your order quantities to cover certain risks?
4. Change the level of your safety stock at certain times. In low season, your customers are likely to accept a lower level of service than in high season.
Improve your forecasting and seasonal demand detection through Artificial Intelligence
Improving demand forecasting is key to maintaining an optimal stock level and, for this, Artificial Intelligence can be of great help. In this respect, the advantages of methods that involve AI are that they are always more accurate. Traditional ways of forecasting demand often only consider one or a handful of factors, whereas methods that leverage AI can consider a much wider variety of factors that can influence demand, and are therefore better able to recognise seasonality.
To achieve this, there are a wide variety of models that can be applied to demand forecasting. Some of the most common ones are:
- Neural networks – computational models inspired by the workings of the human brain that are used in a wide variety of Machine Learning tasks;
- Tree-based models – Machine Learning models that use tree structures to make decisions based on features or attributes of the data; and
- Regression-based models – Machine Learning to predict numerical or continuous values from input data.
AI demand forecasting vs traditional forecasting
Improved forecasting accuracy
As mentioned above, through using AI, a wider variety of seasonal patterns/trends can be identified. And, again, this is due to the greater variety of data it can involve in the process. If the demand forecasting software normally considers variables such as historical demand, weather or events, AI could also weight – and interpret – data such as trending topics on social networks, web visit data, customer reviews on different platforms, macroeconomic data in real time, etc. All this aggregated information can contribute to more accurate forecasts and the detection of seasonal patterns.
Identification of complex correlations and patterns
AI can detect complex and non-linear relationships between variables. This makes it a suitable technology for forecasting in situations where traditional methods may struggle. Recognising seasonality that is particularly subtle could be an example of this.
On the other hand, AI is also very useful for predicting fluctuations in demand during special events, holidays or other unforeseen circumstances. By collecting data on similar events (such as previous promotions or holidays), it is possible to predict demand for upcoming events based on these historical records. In this way, forecasts can be made that predict the volume of demand, based on comparable promotions and equivalent seasonal periods.
AI models can adapt and learn from new data, making them suitable for handling changing demand patterns, seasonal changes and market dynamics.
AI can efficiently manage large amounts of data. This means it is well suited to growing businesses that need scalable technology to cope with the increasing complexity of their operations.
More efficient continuous improvement
AI models can ‘self-optimise’ and improve over time as they are exposed to more data and feedback, resulting in increasingly accurate forecasts.
6 tips for managing seasonal demand
With seasonal items you may perceive a high deviation from past demand. This results in a high level of safety stock. However, if you are able to account for these deviations, you will not need as much safety stock. It is important to correct your historical sales records to adapt them to seasonal demand patterns. This way you will have a realistic picture of your historical demand that will allow you to see if you can reduce your safety stock. When buying items for the off-season, use a dynamic stock for your seasonal pattern.
Distinguish between seasonal patterns of promotions/events
If you have run a promotion or event in the last 2 years, this may appear in your forecast as seasonal data. This is why it is important to separate seasonal patterns from the influence of promotions and events. If you do not clean this data and finally decide not to run that promotion or event, and the event is not repeated in the third year, your forecast will be wrong.
When you want to forecast the demand for barbecues for a specific city, but you only sell 5 units in that city, your statistical population is too small to make a reliable forecast. In this case, it is better to collect data from a wider area, which will give you more information on which to base your decisions.
SKU vs aggregation
Whenever you introduce a new item into your assortment, you will have insufficient information to make a reliable forecast. Here, you can use the seasonal pattern of a similar item or group of aggregated items. It is important to calculate the seasonal forecast for different hierarchical levels. If we use air-conditioning products as an example, we should divide them into heating and cooling products. We cannot group them together when it comes to demand forecasting as their high and low seasons are the opposite.
Forecasting peak demand
Although there are clear seasonal patterns, in some cases there may be variations from year to year. For example, Christmas is always celebrated on the same date in the calendar, but there are significant variations depending on the particular day of the week on which it falls. The same is true for Easter, which varies its date from year to year.
In both cases, we will see that peak demand can shift several weeks earlier or later from one year to the next. This poses a significant challenge in forecasting seasonal demand, as due to these shifts, one cannot rely on an exact repetition of purchase patterns from one year to the next. For this reason, several factors must be taken into consideration in order to be able to accurately identify when peak demand will occur.
Seasonal assortment management
We have talked many times about the importance of having an optimised assortment through ABC analysis, which gives us the highest possible profit margin and ultimately helps us to achieve our business objectives. This also means we need to decide whether there are some items that we can remove from our product portfolio during a certain period of the year because they are not in sufficient demand.
Conclusions: Employing the potential of AI in detecting seasonal demand patterns
Detecting seasonal demand patterns is not a new technique. But, if you continue to use Excel rather than specialised technology in your supply chain planning, you will be far less competitive by not adopting tools that leverage Artificial Intelligence to forecast seasonality.
The main advantage of AI applied to the detection of periodic trends is that it is able to take into account much more data than traditional tools and can therefore identify more complex and less obvious patterns. Similarly, these technologies are highly scalable, so they are ideal if your company is at a point of rapid expansion, as they will be able to keep up with your growth.
With all these elements on the table – greater accuracy in forecasting demand, detection of more subtle patterns, greater scalability – it seems quite evident that opting for a tool that incorporates Artificial Intelligence will improve many of your processes when it comes to detecting and forecasting seasonal demand.