How to Calculate and Improve Forecast Accuracy in Retail

Inventory misstocking cost retailers 1.77 trillion in 2023. Demand forecasting is the classic solution for avoiding huge losses due to overstocks or shortages. In this article, we will consider methods and calculation options and find out which forecast error formula is the best for your business. Also, see below the tips on improving forecast accuracy in the face of modern market challenges.

Key Takeaways

Forecast accuracy underpins financial health and keeps inventory lean.

• Avoids costly overstocks & lost sales. 

• Tracks variance to improve models. 

• Machine learning refines predictions. 

• Links planning with execution. 

• Drives trust across supply chain.

Researchers at McKinsey found that 77% of businesses invest in technology to increase supply chain transparency and improve forecasting accuracy. This demonstrates that understanding processes has become a priority, and the emergence of AI- and machine-learning-based predictive tools in the market further supports this trend.

Consequently, interest in measuring forecast accuracy has grown thanks to these tools' superiority over manual calculations and expert predictions. But are they indeed 100% error-free? To answer this, we need to take a closer look, explain the process of calculating accuracy, and determine the acceptable forecast bias. 

But first, let's start with the basics.

In the context of demand forecast accuracy, it is the correspondence between the forecasted and actual demand. Note that actual demand is the total number of orders or purchase requests, regardless of how much product was in stock.

If the accuracy is low, you risk purchasing more inventory than you need or making customers unhappy due to product shortages.

Should the forecasting accuracy then always be close to 100%? Is that realistic?

Forecasting should be seen not as an end but as a means. Forecasting is only necessary if it helps you achieve your goals. For example, if your goal is to achieve maximum product availability, focus your demand forecasting accuracy on that and try to reach the appropriate accuracy.

Calculating and determining forecast bias is complex and not particularly exciting. So the question is: Is it essential? Let's examine the benefits that accurate forecasting brings to retailers.

Forecasting demand and subsequent planning allows you to create the right cycle: maintain the amount of stock in line with planned future sales (considering promotions, seasonality, and other factors). Ordering only as much as you need when you need it – isn't that what all retailers want? In this case, the goal of forecasting is to avoid shortages and stockpiling. This is even more important when it comes to perishable products.

A satisfied customer equals a loyal customer. They will keep coming back and that is something accurate forecasting helps to ensure since it allows you to effectively guarantee that the products customers need will always be available.

In retail, demand is never stable. In addition to the usual factors for retailers, such as Black Friday, Christmas, and other holidays, seasonality, and weather changes, there is another factor – the influence of social media. If a product goes viral, sales can increase thousands of times out of nowhere. Although it is quite difficult, it still needs to be predicted. Achieving a high level of forecast accuracy is important because it is always more profitable to be prepared for increased demand than to rush to the rescue when the shelves are empty and there's no time for the supplier to deliver a new batch.

Related:  Replenishment Planning: Examples, Importance & Best Practices

Assortment rotation is a common practice for retailers, but how do you know how customers will react to new goods? With accurate demand forecasts based on performance indicators of similar products, you can find out how much of the new product you need to buy and what the projected sales will be.

Your success directly depends on the efforts of your partners: the speed and accuracy of your suppliers. Good planning is the first step to improvement. Try sharing your annual demand forecast with your suppliers for them to adjust production and deliveries accordingly.

Optimal inventory levels ensure fewer goods are stored in the warehouse without excess stock, while increasing product turnover and reducing storage costs. Moreover, with precise demand predictions, you can avoid overpaying for emergency deliveries due to shortages and eliminate the need to transfer goods between outlets.

Having the right product on your store shelves at the right time reduces lost sales, and leads to an overall increase in customer satisfaction and revenue.

What exactly is the forecast accuracy formula used for calculations? Let's see the main metrics, formulas, and examples, and consider each forecast error in detail.

This metric allows you to determine the deviation of the forecasted value from the actual one. It is calculated as a percentage. This is the easiest way to determine the size of the error, which can be used to compare different time periods or products. At the same time, it is sensitive to actual small values and can lead to distortions in which actual demand is = 0. The formula is as follows:

For example, if you planned to sell 100 packs of coffee and sold 80, the forecast error would be 25%. This means that you underestimated demand by a quarter.

This method helps to identify large errors in the difference between actual values and forecasts. This is especially valuable if you want to avoid forecast errors when forecasting demand during peak seasons or when working with expensive products. Be careful, however, because this calculation method underestimates systematic small errors and is not suitable if you are concerned about both large and small errors.

Calculation formula:

yi – the actual value of sales or demand for the i-th product or period.

ŷi – the predicted value of sales or demand for the same product or period.

n – is the number of observations.

This is the average of the absolute difference between forecasted demand and actual sales. MAE is most often used when it is necessary to determine the total amount of errors, whether they are over, or whether the forecast underestimates sales. It is effective, for example, in evaluating the forecasting of daily baguette sales.

Forecast accuracy formula:

Imagine that you have been looking at baguette sales forecasts for three days and found that you should have sold 110, 140, and 125 units but actually sold 100, 150, and 130. The actual errors for each day were 10, 10, and 5, respectively. That is, the sum of errors is 25. We divide this by the number of days of research and get 8.33. This is the average deviation from the forecast.

Mean absolute percentage error MAPE is a measure of the average absolute difference between forecasted demand and actual sales. It is needed to understand the size of errors relative to actual sales. In other words, mean absolute percentage error allows you to clearly understand how accurate the forecast was as a percentage. The calculation is used to compare forecasts for different products or time periods.

Formula for the mean absolute percentage error:

n – number of observations.

Ai – actual value of i at the moment of time

Fi – predicted value at time i

Let's return to our case with baguettes. To calculate the mean absolute percentage error, you first need to find the percentage errors for each day: 10%, 6.67%, and 3.85%, respectively. Next, you need to sum these percentages and divide them by the number of days. The result is an absolute percentage error of 6.84%.

This metric is used to measure the tendency of forecasts to constantly overestimate or underestimate the actual demand. This is necessary to identify systematic problems in forecasting and make timely adjustments. However, the calculation does not give an idea of the size of the errors: it only indicates their direction.

Forecast accuracy formula:

Let’s consider the example of predicted and actual baguette sales again. In this case, we will calculate the difference between the forecasts and the actual figures: 10, -10, and -5, respectively. We look for the sum of these values, and it is -5. We divide the result by 3 days of research and get -1.67. This means that the calculation model underestimates the demand by 1.67 for each observation.

So, we have seen how to calculate accuracy, but now another question arises: what is a good result? What should be the target accuracy for your business model?

Assessing the quality of a forecast depends on the predictability of demand, which is influenced by market trends, macroeconomic factors, seasonality, availability of sufficient space in the warehouse, as well as special events of the retailer (for example, sales).

Nevertheless, the generally accepted indicator for retailers stands at 70–80%. At the same time, an accuracy of 60% is considered good for goods that sell slowly (for example, home appliances or luxury goods) as it is more difficult to predict demand.

It’s important to consider the following principles:

• Measure an achievable indicator for yourself: for example, profitability. Based on it, model situations where a good forecast will affect the result and ignore everything that will not be affected by the forecast. For example, if you sell seasonal goods, ignore changes in demand during the off-season.
• Optimize forecasts at the same time as planning. Put your safety stocks, planning cycles, and task deadlines in order. Analyze in detail the products, batch sizes, and promotions that do not make your business efficient, even despite an accurate forecast.
• If you manage a chain of stores, it's beneficial to create separate forecasts for each outlet, as the inputs will vary by location.

Now we have come to the tips on how forecasting can be improved in supply chain management and retail.

The more accurate and up-to-date your inventory data is, the more effective your demand planning will be. So first set up data collection from trusted sources. Nowadays, this process can be automated: integrate a single platform that will be synchronized with all the applications and tools you use for planning, sales, and customer data. It will collect information in real time.

To effectively manage your assortment, understanding where forecast accuracy is critical versus where some margin of error is acceptable is essential. This is where ABC(D) analysis comes into play, categorizing products based on their importance. For example, Fast-Moving Consumer Goods (FMCG) like milk require high forecast accuracy. If customers can't find their preferred milk, they may leave the store, leading to a loss in sales not just of the milk but also related products like cereals.

Conversely, products such as 10-year-old whiskey outside the holiday season can tolerate forecasting errors. They have a long shelf life, and off-season demand for expensive goods is inherently unpredictable.

When considering XYZ analysis, which emphasizes demand variability, it's important to note that for Z-category goods—characterized by high demand variability—achieving accurate forecasts is often unattainable.

Advanced demand planning systems, like LEAFIO AI, can both be set automatically and adjusted individually for each company. The software constantly checks the accuracy of automatically generated and adjusted forecasts. If the accuracy goes beyond the set limits, the system sends an alert to the user to correct the problem.

There are seasonal and non-seasonal external factors. The former are predictable (holidays, precipitation, temperature, etc.), while the latter are irregular and unpredictable. Use historical data as well as general market information to determine their impact on your sales.

There are many methods for forecasting external factors: trend analysis, moving average, exponential smoothing, regression, etc. You should mix and match them depending on your needs. For instance, if you have seasonal factors, exponential smoothing with seasonal adjustments is the right choice.

Demand forecasting software is your assistant that gives you ready-to-use data for considered planning and strategic decision making.

For example, LEAFIO's AI-Powered Demand Planning Software is an advanced platform that calculates the demand for each type of product as accurately as possible based on a complete history of sales, promotions, seasonality, and other factors. Using it, you can:

• Make informed strategic decisions.
• Respond to changes in demand quickly and in a timely manner.
• Increase customer loyalty.
• Boost profits and reduce costs.

LEAFIO AI Demand Planning Software is used by supermarkets, grocery stores, liquor stores, e-commerce, toy stores, and many other retailer industries.

Imagine several possible scenarios based on specific assumptions. This will help you find opportunities and identify risks. Test these scenarios using the what-if analysis. This will help you adjust key variables and model possible outcomes. The combination of these methods will help you make informed decisions and make more accurate forecasts.

For example, if you're forecasting sales for a year, create an optimistic scenario (demand growth due to economic stabilization), a pessimistic scenario (crisis intensifies, demand declines), and a baseline scenario (nothing changes). Check how price changes in each case will affect profits and prepare for all possible outcomes.

This tip brings us back to the first point – the importance of accurate data. Each department has its own information. The sales department will provide data on the level of demand: historical and current. Marketers will share information about market trends and upcoming promotional activities, finance staff will share information about the budget, and so on. Combine this information, and you will have a comprehensive picture to create a forecast.

Ensure that you can react to changes quickly: monitor data and results and make changes to forecasts. If the deviation between actual sales and forecasts is significant, it's time to review your data and improve your forecasting process. You need to track results in real time to be able to make the most accurate decisions. By the way, this task can also be entrusted to AI-based demand forecasting software.