Five ways to help you improve your demand forecasts

Edut Birger
Marketing Associate

Demand patterns throughout 2020 have fluctuated dramatically for many businesses, displacing demand for everything from restaurant dining to transportation and accommodation. Due to this, forecasts that primarily rely on historical sales or transactional data will not give you the full picture as to when and where you’ll experience demand spikes and dips in the future.

Consumers are changing where and how they shop, what they shop for, and how much they buy at faster rates than ever before. It can be extremely difficult for retailers to identify which consumer trends are temporary and which are here to stay. Building resilience into your forecasting models by making them more dynamic will ensure anomalous periods are accounted for in the future, so you can be prepared.  

What happens if your forecasts aren’t accurate?

According to The Institute of Business Forecasting, the average one-year forecast for businesses specializing in consumer products can be off by up to 20%. If a company generates $1 billion in revenue, this discrepancy translates to as much as $225 million in losses annually on costs like overstocks.

When demand unexpectedly differs from previous sales patterns, businesses struggle to identify the cause of that change, and miss opportunities to improve forecasts to account for these demand causal factors in the future. This can not only affect opportunities to make sales, but also might lead to out-of-stocks or overstocks of inventory, which cost retailers more than $1.1 trillion globally each year.

Without accurate forecasting models, you will be caught off guard by demand causal factors that can influence everything from your supply chain management to your workforce planning and optimization. This translates to losses in revenue and increases in operational costs. The good news is the right external data will  improve your models, reducing your error rate and driving better business outcomes.

Five steps you can take to improve your demand forecasting models

Hone your data modeling expertise

Before you can start improving forecasts and models, you need to ensure that your own internal data is clean and accurate. This includes having large repositories of data on your own product sales and performance, so you have an accurate idea of what’s going on internally for your business.

You can use a demand analysis to help you review product performance across geographies, brands, SKUs, pricing tiers, and customer demographics. All of these considerations will help you paint a picture of your own market penetration and channel exposure. From there, you can develop your understanding of what to expect during a “typical” year for your target market.

It’s important for you to not only understand demand, but also the actual consumption patterns of your target market. Consumers don’t always consume at the same rate that they buy, so you may need sophisticated modeling techniques to account for this.

Identify demand anomalies

Analyze your own data to pinpoint exactly when and where your business experienced previously unexplainable demand shifts that you weren’t able to plan for. Here, you should remove the seasonal, monthly, or weekly trends your business experiences to truly identify your baseline demand. This is a great starting point to help you understand what kind of external factors might have influenced these spikes or dips.

One notable demand disruption has been the change in foot traffic patterns for consumers. With normal operations limited during the pandemic, warehouse and grocery stores are seeing a 30% increase in foot traffic and convenience stores are noticing a 3% increase. Consumers have also reduced the number of shopping trips they make by 15%. However, grocery e-commerce has risen to 31%, and that trend isn’t likely to go away any time soon. Changes in foot traffic can be significant drivers for demand and can cause otherwise unexplainable peaks or spikes.

While you need to optimize your business operations for the near term, you should also be aware of long term consumer behavior trends to make sure your models are resilient.

Correlate anomalous demand with demand causal factors

For example, understanding how shelter-in-place orders affect sales for things like household supplies versus electronics will help you understand where to boost your supply chain, or identify opportunities to run marketing promotions. That way, when future shelter-in-place orders are enacted or lifted, you’ll know exactly how that affects your consumer’s buying behavior, and where to make adjustments to reflect these demand shifts.

You can use a tool like PredictHQ’s Beam, which allows you to map your own demand data on top of real-world events to find correlations between demand and things like severe weather, academic events, and live TV broadcasts. Once you establish correlations, you’ll know what kinds of impactful events to look out for in the future.

Leverage external data sources

Many retailers are finding themselves in a data deficit due to the unpredictability of 2020’s economic trends. Consumption and channel disruptions have displaced demand for many businesses, especially those in the CPG, food and beverage, and travel industries.

Events and other demand causal factors are considered to be uncontrollable drivers of demand. Internal data only tells you part of the story, but external information sources can help connect the dots on what demand drivers are most impactful for your business. The more information you can feed your forecasting models, the smarter they will get. Higher forecasting accuracy will lead to better business predictions.

Utilize machine learning techniques

Artificial intelligence and machine learning models learn from whatever knowledge you provide to them. Your algorithms are only as accurate as your data sources, so supplying them with more relevant information helps train your models to identify future demand patterns.

Leading organizations use advanced analytic models with different inputs (point-of-sale data, primary consumer research, social listening, online search trends, and more) to understand how demand evolves at a granular level.

You can leverage techniques like cross grid validation and ensemble forecasting to improve your models and reduce forecasting errors.  

These are just a few examples of steps you can take to build more accurate and resilient demand forecasting models. Depending on your business, certain demand causal factors might be more impactful than others. To better understand how external factors like events affect your demand, contact us below.