Consumption Sensing – an Introduction

    
Consumption Sensing: Leveraging AI/ML to predict the shift in consumption patterns across the short, medium, and long-term horizon, leveraging internal and external data such as sales history, inventory, market share, shopper’s basket, macro-, and micro-events.

Never in recent history has there been such a drastic shift in consumption patterns as what we are experiencing right now. The Coronavirus Pandemic has fundamentally challenged retailers and consumer products companies to think differently about their supply chain and demand fulfillment. However, albeit the current situation is an extreme example, unforeseen changes in consumption patterns have always been and always will be a regular occurrence.

Many companies grapple with predicting consumption by applying standard forecasting, category trend analysis, or using adjusted shipment forecast as a surrogate for a consumption forecast. However, linking the consumption forecast outcome to the planning process is a real challenge as the analysis is often a one-off effort that is not systematically linked to the demand planning cadence.

This is where Consumption Sensing comes in – utilizing artificial intelligence and machine learning to forecast consumption demand across the time horizon, leveraging a variety of internal and external data sources.

Recently mentioned by a leading consumer products company, “In fact, Consumer Demand is the only demand that is worth forecasting. The next best option is forecasting customer orders and the worst is forecasting our shipments. The consumption signal is the only signal for which we have an actual to compare against in the form of POS and syndicated data.” 


Consumption Sensing can be applied over different timeframes to capture the impact of different types of market forces. On a short-term interval of 4-8 weeks, consumption sensing can leverage localized events such as weather data, i.e., extended cold spells or blizzards, to help identify local stores that will likely be running out of soups or other cold-weather foods. The impact of events like temporary government shutdowns, as another example, can be surfaced through the inclusion of socioeconomic data for areas where food stamps are heavily used.

In contrast, for medium-term intervals of 1-4 months, more extensive modifications in consumer preference can be caught through data reflecting more structural changes. For example, health advisories regarding E. coli outbreaks from lettuce can impact the preference for other lettuce types for as long as the source of the outbreak is unknown. Similarly, capturing data for wholesale substitutes in school menus as the result of calorie or ingredient concerns will help CPG firms rebalance inventory supply to match.

Finally, capturing demographic and socioeconomic data on a local and regional basis can infuse forecasting with insight on more pervasive trends over a longer interval of a year or more. Examples would be identifying regions of the country where the age demographic might prefer more organic foods, more pre-prepared foods, or more traditional ingredients for meal preparation. 

The latest coronavirus pandemic is an extreme and unfortunate example of a market force that is dramatically impacting retail and consumer products companies. Although Consumption Sensing is not designed to forecast the impact of sudden panic, it is precisely the kind of discipline needed to support faster reaction to a broader range of market forces impacting demand. Its AI and machine learning can significantly help companies manage through times of such volatility. One CPG firm, for example, has applied Consumption Sensing to their planning process specific to Amazon, leveraging market data plus the wealth of additional data Amazon provides, to achieve a 2x improvement in forecasting accuracy.  As a result, they are better prepared to address the pandemic-like bursts of orders.

Consumption Sensing is a new layer of precision in demand forecasting that is available today. Thanks to significant advances in the scale of data collection and modeling, Consumption Sensing makes it possible to capture the key factors driving near-term demand on a localized level across the appropriate store clusters and/or major retail players. And it will be critical to the success of retailers and consumer products companies in this dynamic and competitive environment.