Products that are sold consistently make demand forecasting a seamless process. On the other hand, products that don’t sell consistently or sell on a seasonal basis have erratic demand patterns which are nearly unpredictable. 

How to accurately forecast demand for products that have erratic demand behaviour? 

This is a question that has eluded Demand Planners for ages now. Erratic time series with no consistent patterns or trends make demand forecasting and deriving insights a nightmare. In this blog, let’s delve into how AI powered demand planning techniques help unravel these time series with ease.

‍What are the consequences of poor demand forecasting accuracy?

Poor demand forecasting accuracy is an obvious result when you struggle to deal with erratic demand. This in turn affects your supply planning efficiency and leads to inventory imbalances such as understocking and overstocking.

‍

Understocking leads to missed revenue opportunities and poor customer experience. Whereas, overstocking leads to working capital being tied up and additional storage costs.

‍

‍How do you deal with erratic demand behaviour? 

In this blog we have discussed 3 common ways to deal with erratic demand patterns.

‍

‍Anomaly Capping

One of our CPG clients was experiencing highly inconsistent/erratic demand for specific products. The demand for these SKUs had regular occurrences during a certain duration with high quantity variations. Essentially there were no specific patterns or periodicity for the highs, lows and the zero periods in the demand. Therefore, their predictability remained low. 

We adopted an anomaly capping approach to forecast demand accurately in such a case. Anomaly capping is a technique used to handle extreme values or outliers in a dataset. It is important to ensure that anomalies do not overly influence the forecasting process, leading to inaccurate predictions. Here’s how the process of anomaly capping works:

• Computing the upper and the lower limit of the demand data
• Identifying any anomalies that lies outside the defined range
• Replacing the anomalies with a capped value so that we can prevent any extremely low/high values
• Forecasting future demand with the cleaned dataset

Anomaly capping technique helped the client to get more stable and accurate demand forecasting model, even with an erratic time series data that had sudden spikes or dips due to unpredictable factors.

‍Impact delivered

‍

‍Availability Based adjustment 

One of our Q Commerce clients had been encountering instances of out of stock often for certain products that have erratic demand behaviour. They were forecasting demand to align with the set availability level for these products. However, availability levels of these products were highly fluctuating due to several factors including customer demand, supplier lead times, fill rate etc.

 We forecasted demand based on availability based adjustments. Availability based adjustments involve incorporating information about product availability, supply chain disruptions, or other factors that affect the actual demand for a product. These adjustments helped to improving the demand forecasting accuracy by accounting for situations where demand was constrained by factors other than consumer behaviour. 

‍

The adjusted demand represents what the real demand would have been if not for the disruptions in the supply chain. This gives a more accurate picture of the demand and ensures that the client is optimally stocked to meet this demand without running out of stock.

Remember that availability based adjustments might be more complex in the real world scenario owing to several factors like the type of disruptions, fluctuating impact levels and availability of historical data. The key here is to get a clearer and accurate picture of demand by considering the impact of availability.

‍Impact Delivered

‍

‍Event Based Predictions

One of our CPG clients had been experiencing intermittent demand for certain products. Here, the historical demand data shows very little variation in demand quantity but a high variation in the interval between two demands. Though specific forecasting methods tackle intermittent demands, the forecast error margin was considerably higher in this case. As a result, they faced inventory pile ups during the periods where there was very low or no demand. 

Certain SKUs had very high demand during particular days of the month (Block periods) and had very low/no demand during the days in between (Non Block Periods). The sparsity in the data makes accurate demand forecasting for the block periods a challenge. So, we created events during the block period which will help us gauge the impact of events and predict demand accurately for these durations. In other words, this is a form of planned demand forecasting as we are aware of the periodic spikes in demand. This way, the client was able to procure the right amount of inventory to cater to these block periods without having excess inventory at the end of these periods.

Additionally, if there were any promotions run during these block periods, we also make sure to factor in their impact while forecasting.

‍Impact Delivered

‍

Further, erratic demand demand patterns can be unraveled with actions on the inventory planning side of things including dynamic safety stock recommendations and real time supplier evaluation. 

Demand forecasting for products with erratic demand behaviour could be a daunting task but not impossible. A few demand planning softwares use technologies like AI and ML to make it possible to analyse sparse/erratic data, derive valuable insights and forecast future demand accurately. Embracing tech and combining it with your business insights for enhanced results is the first step in your quest for unmatched demand forecasting accuracy. Remember, this is an ongoing journey, and every step forward is a step towards better operational efficiency.

‍

‍