Accounting for Pandemic-Era Data in Demand Forecasting
Even without the uncertainties of a post-COVID world, businesses are struggling to make sense of changing consumer behavior and how it impacts demand. This struggle is especially relevant in consumer goods and retail industries because the sheer number of SKUs, huge inventory assortments and the operation-intensive nature of the business.
The key to profitability in CG and retail operations is accurate demand forecasting. Accurate demand forecasting can affect and be influenced by everything ranging from pricing to promotions to product assortments and even inventory. Understanding the demand planning impacts of seasonality, weather, events and other causal factors are essential to a retailer’s ability to forecast accurately. On the flipside, along with unused inventory, inaccurate forecasting can increase costs across the supply chain, labor and business operations.
Legacy forecasting tools have proven not to be helpful because they are based on traditional regression analysis methods and use comparable historic data, which has been highly disrupted by the pandemic.
But no precedent does not mean no data. AI is changing the game in demand forecasting because it has the ability to synthesize disparate forecasts and demand-driving factors into a more accurate view, helping to more quickly recognize patterns and trends.
Legacy forecasting limitations
Legacy approaches assume history repeats itself, anticipates stability and relies on predefined models that depend on historical demand. They fall short of fully accounting for weather changes, impact of new launches, economic downturns, etc. They also fail to recognize and qualify a trend as a potential opportunity. Moreover, most forecast methods require data of at least two years to establish a recurring pattern, which has been highly impacted since March of 2020.
Naturally, the forecasts coming out of such models lack accuracy in a rapidly changing environment; more so, in short- to mid-term forecasts where there is little scope of error. Without accurate forecasts the bottom line is greatly impacted.
One additional factor is that most forecast tools in the market are built to presume that actual demand occurs in stores, which produces a biased result given the proliferation of channels like home delivery and curbside pickup.
A common challenge for many CG and retail organizations is the proliferation of disparate forecasting solutions, all with narrow purviews of data, that have been built for specific use cases such as pricing, promotions or inventory. It is not uncommon for enterprises today to have five or six different forecasting platforms, all of which suffer from some degree of error given the siloed understanding that these individual tools have.
Machine learning-based forecasting: Relevance in rapidly changing environments
There are many factors that influence demand forecasting accuracy, both externally and internally. And often, a sizable number of causes and critical data points go unnoticed by traditional forecasting tools. However, with the integration of machine learning (ML) algorithms, forecasting tools have become more advanced to provide proactive and predictive insight into consumer behavior and preferences, and enable better decision making.
As an example: Winterwear sales could have been higher last year because of the early onset of winter. And, due to the delay in procurement, a certain percentage of customer demand remained unfulfilled.
Forecasting for demand this year should not just be based on last year’s sales. An ideal approach would consider other factors like:
- Weather department’s prediction of a longer fall and a less severe winter than last year.
- Unfulfilled demand of last year.
- People working from home and consequent change in style preference.
- Changing channel preferences.
- New product launches by competitors.
- The emergence of more competitors in specific categories.
Accounting for the number of variables and the magnitude of change can be overwhelming for analysts working with legacy tools. Here, ML-based predictive analytics can help ease decision-making by considering the following:
1. External factors: Weather trends and events, competition, fashion lifecycle, changing consumer behavior and preferences and more recently, the COVID-19 pandemic.
2. Internal factors: Unfulfilled demand and unsold inventory from last year, channel performance and volatility experienced over the past few months of the pandemic, available shelf space, new product launches, etc.
The next generation of demand forecasting
When done right, AI-powered demand forecasting can help companies both save and make money by being better tuned in to the needs of their customers. There are unified demand forecasting platforms accounting for all relevant factors resulting in a far more accurate enterprise-wide view.
Advanced demand forecasting tools help CG brands and retail organizations dynamically interpret, adjust and shape ever changing consumer behavior, whether it is helping a retailer figure out the right price point, determine the right promotional offer, or optimize the inventory investments in a more meaningful and accurate way.
There is nothing wrong with benefiting from historical data, but AI allows retailers to understand what is happening in real-time regarding consumer trends, journeys, and expectations – proving critical during times of uncertainty.
Todd Michaud is president and chief commercial officer of Hypersonix.