Demand Forecasting: Buy or Build?

Demand planning is essential in many businesses. From retail to manufacturing, a failure to do proper demand planning may lead to overstock, supply chain disruption or lost sales, all of which can be costly. An accurate demand forecast lies at the basis of demand planning, and many businesses cannot function without one.

Buy Versus Build

Demand forecasting models can be acquired as out-of-the-box solutions, often as part of a demand planning tool. These models are usually easy to integrate and provide reasonable-quality forecasts.

As an alternative, a business may develop a custom demand forecasting model. While this requires an investment, it can also lead to more accurate forecasts and other benefits.

The best option for your business depends on many factors. In general, building a custom demand forecasting model is only feasible for larger businesses where demand planning is crucial for their everyday functioning.

Below, we will explore some of the pros and cons of custom demand forecasting models.

Adapting Demand Forecasting to Your Business and Data

Any demand forecasting solution needs data to base its forecasts on. The bare minimum would be data on historical sales and assortment. This can be combined with a multitude of other data sets, such as promotion details, to improve the quality of the forecasts.

For an out-of-the-box demand forecasting solution, the input data must be transformed into the format the solution expects. If your data is more detailed than the solution requires, this may lead to a loss of information and, hence, a loss of potential forecasting power. A custom demand forecasting model may be tailored to your data in such a way that it can leverage all details to improve the forecast quality. For an in-depth exploration of deploying custom demand forecasting at scale, check out our webinar on the topic.

Sales and Demand

An example area where detailed data may give a custom forecasting model an advantage is the conversion of historical sales to historical demand. In order to forecast future demand, a demand forecasting model needs to learn from historical demand. In many cases, however, the true historical demand is not known; instead, historical sales are used to estimate this demand. Often, sales and demand may be the same, but sometimes they are not. For example, if an article is out of stock, customers may opt for an alternative, which means that the true demand is higher than the sales.

It is relatively simple to create a crude estimate of the demand by extrapolating the sales in out-of-stock situations. Any good out-of-the-box forecasting solution should be able to do so. However, if your business has search data from an online store, shopping lists from an app, or customer counts from a store, such data sources may be used in a custom demand forecasting model to greatly enhance the accuracy of the extrapolation, resulting in better forecasts.

Advanced Features

An out-of-the-box demand forecasting solution may limit you to the existing feature set, but a custom model can be extended at will. For example, the forecasting interval can be easily increased to hours instead of days for promotions, or promotion cannibalization and cross-selling can be added to increase the forecast accuracy of articles influenced by promotions of other articles.

Automation, Automation, Automation

Automation is key for demand forecasting. Many out-of-the-box solutions require manual actions to add new articles to the assortment or adjust forecasts for holidays. Such actions are error-prone and labor-intensive, especially for businesses with thousands of articles in their assortment. A custom-built forecasting solution can automate these tasks, resulting in a lower administrative burden and better-quality forecasts.

Cost and Benefit

Cost is, of course, a very important factor. Out-of-the-box demand forecasting solutions typically have a low upfront cost, but a higher license or service fee. Custom-built demand forecasting models require an upfront investment, as well as in-house data science expertise. Nevertheless, they can be less expensive than out-of-the-box solutions in the long run.

But even if a custom-built solution is more expensive in the long run, the benefits may tip the balance in its favor. A 5% better demand forecast, like the one Albert Heijn achieved partnering with Xebia, can significantly improve efficiency in your business.

Conclusion

Whether it is better to buy an out-of-the-box demand forecasting solution or build a custom one strongly depends on the business requirements. For larger businesses, especially those dealing with perishable goods, a custom demand forecasting model may well be worth the investment.

Are you interested in developing or improving your demand forecasting? Check what we offer and send us a message!