Best Practices for Great Forecasts

For many suppliers, accurate demand forecasts can be hard to come by. Forecasts not only take a while to develop using traditional methodologies, but human error and bias can muddy the water as well. Inaccurate forecasts can lead to inefficiency and waste throughout the supply chain, coming at a great cost to their business.

Statistical forecasting is the best-in-class methodology for generating accurate forecasts. It removes the errors and biases that trouble traditional forecasting methods, providing suppliers with a reliable forecast at a fraction of the time. Statistical demand forecasting software, like Crisp, takes this methodology to a new level by incorporating new data automatically and calculating forecast changes in minutes.

There are several important factors to consider when trying to generate an accurate statistical forecast. How much data do you need to start out with? How are outliers handled? Below we’ll describe what you need to get started, and what you’ll want to do, on your journey to develop  great forecasts with Crisp. 

Collect your data

You know the saying “hindsight is 20/20?” Statistical forecasting puts this idea into practice by using your historical sales data to assess the impacts of important demand drivers like seasonality, holiday effects, and trends. It then uses this information to extrapolate what is likely to happen in the future. So how much historical data does it take to sufficiently understand important demand drivers?

Quality forecasts can be created with sales data from just the past two to three years. This is generally considered an adequate number of cycles to estimate the effects of regularly recurring events like holidays, seasons, and even promotional periods.

Newer brands with shorter sales histories can leverage statistical forecasting for more accurate and faster forecasts too. Crisp can assess trends, as well as integrate user feedback, to generate a forecast even when there isn’t as much historical data available. You can also build relationships between products to increase the amount of data available to seed the forecast. This is useful when you need to forecast for a brand new product line as well!

Once you’ve compiled your data from the last few years, be sure to spotcheck it for consistency across orders. This will help to keep your data organized once it is uploaded to Crisp.

Give context to data anomalies

Anomalous events like stockouts, pipefills, or one-off promotions can cause your sales to dip or spike dramatically. Those spikes and dips in sales create outliers in your data, which introduce noise to your forecast and obscure your product’s true demand signal. Ultimately, this noise can lead to decreased forecast accuracy. 

You can inform Crisp’s algorithm about unusual changes in past demand to dampen the impacts of noisy data and improve forecast accuracy. Start by visualizing your product’s sales history through the product’s details page. On the plot, look for eye-catching spikes and dips in your data: they typically represent anomalous events that you may want to explain to Crisp. 

Crisp will also try to proactively identify anomalies in your data. You can review the data points that Crisp has flagged as outliers by clicking the historical outliers button in the product’s details page.

After you identify the anomalous data, click on the data point to get started with entering an event. Select an event type and describe what the event was, then allow Crisp to estimate the impact of the event to save time. Saving your event will cause Crisp to integrate the latest information. From there, it will calculate a new and improved forecast for you within minutes.

Train Crisp about common events

You can train the Crisp algorithm to track the impact of a specific type of event by tagging your data with promo IDs. These IDs can be used to tag a variety of events that appear within your historical data, like sales driven by off invoice allowances, percentage discounts, and ad campaigns. The ID tells Crisp to log what it learns about events with the same promo ID across product lines. As you tag your data with events that have the same ID, Crisp’s understanding of the event’s impact will grow. Crisp can then use this information to better predict the impact of future events of the same type.

Start by entering a few events to tag your historical data, particularly focusing on data from promotional periods. As you enter these events, include promo IDs to identify events that are similar. Then, allow Crisp to estimate the impact of each event by selecting the “I don’t know” impact option.

After tagging some historical data, enter events in the future to inform Crisp about upcoming promotions. Tag the future event with the relevant promo ID and allow Crisp to estimate the impact. By taking the extra steps to train Crisp about your promotions, you will find that Crisp’s ability to estimate the impact of future events will become even more accurate over time! 

Create a consensus forecast

Statistical forecasts create a reliable baseline, but they will not be able to perfectly predict the future. Your team’s own insights remain an integral part of the forecast. .

Your colleagues and team members may know about future changes that will drive up demand for your products beyond what the statistical forecast could have predicted, like partnering with a new retail chain. A collaborative forecast that combines the statistical baseline with your team’s insights is therefore ideal.

To that end, account administrators can invite their team members to their Crisp account. Once those team members are given editor permissions, they too can enter events into Crisp to inform it of upcoming events that will shift demand. This will ensure that Crisp always has the latest information from the experts!

Unlock the opportunity

Accurate forecasts can unlock tremendous opportunities for your business. Taking the time to develop a reliable forecast is worth it, and with Crisp, it’s easy too!