Learn how to use operations research techniques and tools to forecast demand in retail more effectively. Discover the best methods, metrics, and criteria for demand forecasting and planning.

Analyzing historical sales data provides a crucial baseline for future projections. By studying past sales patterns and trends, retailers can identify recurring peaks and troughs tied to seasonal events, promotions, and other cyclical factors. Common statistical techniques used include moving averages, weighted averages, simple linear regression, and exponential smoothing to detect patterns and make informed forecasts. Historical sales data remains an indispensable starting point. But in today’s complex retail landscape, relying on just one forecasting approach is insufficient and potentially risky.

What are the best ways to forecast demand in retail?

Forecasting demand in retail is a crucial task for optimizing inventory, pricing, and marketing strategies. However, it is also a challenging one, as retail demand is influenced by many factors, such as seasonality, trends, promotions, competition, and consumer behavior. In this article, we will explore some of the best ways to forecast demand in retail using operations research techniques and tools.

1 Historical data

One of the simplest and most common ways to forecast demand in retail is to use historical data from past sales, orders, or shipments. Historical data can provide a baseline for estimating future demand based on patterns, cycles, or averages. However, historical data alone may not capture the effects of external factors, such as changes in consumer preferences, market conditions, or competitor actions. Therefore, historical data should be supplemented with other sources of information and adjusted for any known or expected deviations.

Add your perspective

Dr. Venky Venkata Reddy (డా. వెంకీ వేంకట రెడ్డి), PMP®

Full Professor | Design Thinking & Innovation | Artificial Intelligence | Operations Management | Entrepreneurship | Board Member
Report contribution
Analyzing historical sales data provides a crucial baseline for future projections. By studying past sales patterns and trends, retailers can identify recurring peaks and troughs tied to seasonal events, promotions, and other cyclical factors. Common statistical techniques used include moving averages, weighted averages, simple linear regression, and exponential smoothing to detect patterns and make informed forecasts. Historical sales data remains an indispensable starting point. But in today’s complex retail landscape, relying on just one forecasting approach is insufficient and potentially risky.

Like
Jay Sarmaz

Vice President of Data & AI at Case Status | Applied Data Science, Legal Tech, GenAI, Innovation | 6k+ 🏁
Report contribution
One most recent innovative approach involves data collection from public web sources on product availability, attributes, prices, and consumer sentiments across all retail stores in a given area. This allows aggregation of data across multiple sources and brands and allows a very accurate demand prediction.

Like
Alexandre Warembourg

Senior Data Scientist | Expert Prévision & Time Series | Google Certified Machine Learning Engineer
Report contribution
Although historical data is a very good estimator of futures sales, in the case of product forecasting there are a lot of external factors to take into account such as promotions, prices, national and local events that can't be accounted for by data. There are also some effects which are more difficult to see but which are present like cannibalism, trend break, halo effect ..etc. which can't just be shown directly in data. So to predict the demand, we need a robust machine learning or deep learning model to account for this variety of factors.

Like
Dr Subhendu Patnaik

MDI Gurgaon-Assistant Professor (Senior Grade) | CII-Delhi Expert Panel Member HR | Ex IMT Ghaziabad| Ex Chair Holder: Sir Cary Cooper Professor of Organizational Psychology | Thinkers50 member |
Report contribution
Climatic conditions, preferences based on cultural values, changes in demographics, real disposable income might add on to predications….

Like
Vasiliy Zamolotov

Manager (ex-SAS, ex-Accenture) | Operations Research | Data Science Lead
Report contribution
Some of the best approaches to forecasting demand based on historical data: the use of the Facebook prophet library, the use of random forest, ARIMA/SARIMA models, exponential smoothing, moving average. The choice of approach depends on the frequency of sales and the volume of data.

Like

Load more contributions

2 Qualitative methods

Qualitative methods rely on the opinions, judgments, or insights of experts, stakeholders, or customers to forecast demand in retail. Qualitative methods can be useful when there is limited or no historical data available, or when there are significant uncertainties or risks involved. Some examples of qualitative methods are surveys, interviews, focus groups, Delphi method, and scenario analysis. Qualitative methods can provide valuable insights into customer needs, preferences, and expectations, as well as potential opportunities or threats. However, qualitative methods can also be subjective, biased, or inconsistent, and may require a lot of time and resources to collect and analyze.

Add your perspective

3 Quantitative methods

Quantitative methods use mathematical models, algorithms, or techniques to forecast demand in retail based on numerical data and statistical analysis. Quantitative methods can be more objective, accurate, and reliable than qualitative methods, as they can handle large amounts of data and account for complex relationships and interactions. Some examples of quantitative methods are time series analysis, regression analysis, causal models, machine learning, and optimization. Quantitative methods can provide precise and consistent forecasts based on data-driven assumptions and criteria. However, quantitative methods can also be complex, costly, or difficult to implement, and may require specialized skills and tools.

Add your perspective

Alexandre Warembourg

Senior Data Scientist | Expert Prévision & Time Series | Google Certified Machine Learning Engineer
Report contribution
Quantitative methods for predicting demand, such as machine learning, are now part of the state of the art in demand forecasting. These models are essential for taking into account all the complexities of the problem, including the prices of different products, competitors, promotions, the socio-demographic characteristics of the store and therefore product sales, as well as estimating seasonality, trends, taking account of out-of-stock situations and other complex factors. These methods also make it possible to predict products with little or no history, and to learn from group effects, making forecasting more robust.

Like

4 Hybrid methods

Hybrid methods combine the strengths of qualitative and quantitative methods to forecast demand in retail. Hybrid methods can balance the trade-offs between accuracy and simplicity, data and intuition, and flexibility and stability. Some examples of hybrid methods are judgmental adjustment, combination forecasting, simulation, and fuzzy logic. Hybrid methods can provide more robust and realistic forecasts that incorporate both human and machine intelligence. However, hybrid methods can also be challenging to design, execute, and evaluate, and may require a high level of coordination and communication.

Add your perspective

5 Forecast accuracy

Forecast accuracy is the degree of closeness between the actual and predicted demand in retail. Forecast accuracy is important for measuring the performance and effectiveness of the forecasting methods and tools. Forecast accuracy can be assessed using various metrics, such as mean absolute error (MAE), mean absolute percentage error (MAPE), mean squared error (MSE), or root mean squared error (RMSE). Forecast accuracy can be improved by using appropriate methods and tools, validating and updating the data and assumptions, and monitoring and correcting the errors.

Add your perspective

Alexandre Warembourg

Senior Data Scientist | Expert Prévision & Time Series | Google Certified Machine Learning Engineer
Report contribution
Metrics such as mae or mape are not sufficient to guarantee that our forecast is robust. We need to use weighted metrics such as WMAE or WMAPE as a first step to take into account disparities between time series levels. Secondly, it's interesting to relate these metrics to a baseline, such as a business expert forecast or an average, which will give meaning to the figure. Finally, these KPIs should be complimented with business metrics such as overstock, missed sales or a reliable forecast number.

Like

6 Forecast evaluation

Forecast evaluation is the process of comparing and contrasting the different methods and tools used to forecast demand in retail. Forecast evaluation is essential for selecting the best method and tool for a specific retail context, objective, and constraint. Forecast evaluation can be done using various criteria, such as accuracy, reliability, simplicity, cost, timeliness, or usability. Forecast evaluation can be done by using quantitative measures, such as error metrics, or qualitative measures, such as feedback, ratings, or reviews.

Add your perspective

7 Here’s what else to consider

This is a space to share examples, stories, or insights that don’t fit into any of the previous sections. What else would you like to add?

Add your perspective

What are the best ways to forecast demand in retail?

1

2

3

4

5

6

7

1 Historical data

2 Qualitative methods

3 Quantitative methods

4 Hybrid methods

5 Forecast accuracy

6 Forecast evaluation

7 Here’s what else to consider

Operations Research

Rate this article

Thanks for your feedback

More articles on Operations Research

More relevant reading

What are the best ways to forecast demand in retail?

1

2

3

4

5

6

7

1 Historical data

2 Qualitative methods

3 Quantitative methods

4 Hybrid methods

5 Forecast accuracy

6 Forecast evaluation

7 Here’s what else to consider

Operations Research

Rate this article

Thanks for your feedback

Explore Other Skills