Forecasting true demand has always been a challenge, and it’s especially difficult for food companies operating in fresh and perishable categories. At the same time, accurate forecasting is essential for upstream supply-chain planning and execution, and reducing food and financial waste.
In times of market inconsistency like the COVID-19 pandemic, most forecasters have been left grasping at the truth — trying to balance typical demand patterns with the impact of panic buying, category shortages and logistical complexities.
Even in “normal” market conditions, without alignment and visibility around a single, consumption-driven forecast, upstream suppliers (retailers, wholesalers, and producers) are left to generate their own, siloed forecasts.
When forecasts are created independently, biases, errors and safety stocks are compounded, because each forecast reflects only the order history, patterns, price fluctuations, and real or perceived availability of inputs of its immediate downstream partner in the chain.
This bullwhip effect, in which a plus- or minus-5% change in actual consumer demand impacts upstream suppliers by as much as 40% in either direction, means that maintaining a consistently accurate inventory, even in the best of times, is challenging. Today, during the pandemic, it’s impossible.
The Promise Is Here
The use of technology to connect partners across the supply chain isn’t new, but over the last decade, it hasn’t been as effective as hoped.
However, we’ve reached a new tipping point in our ability to use technology to efficiently incorporate numerous demand signals. Big-data analysis platforms, artificial intelligence (AI), machine learning, and agile forecasting models all play an important role in generating significantly more reliable forecasts. Finally, tools that leverage these technologies are being proven out, and we’re able to look across the supply chain and use all of its data in a way that wasn’t possible before.
AI-driven demand-forecasting platforms don’t rely on historical sales modeling like many traditional demand platforms. Instead, they use dynamic demand influencers, algorithms, real-time data and cloud technology to accurately predict behaviors and trends.
For example, audience demographic and lifestyle trends, shopping patterns, weather events, trade tariffs and retailer marketing initiatives all have significant impacts on buyer behavior and demand accuracy. But this data typically resides in multiple, disconnected repositories.
Using algorithms and predictive analytics, AI and machine-learning-powered demand-forecasting platforms analyze massive volumes of data regardless of where it lives. This technology can automatically identify the most relevant factors impacting consumer demand, anticipate changes in demand and buyer behavior ahead of time, reduce chargebacks, and deliver unbiased probabilistic predictions about future demand.
Unlocking the potential of all this data not only enables more accurate demand forecasts, but also supports a holistic approach to inventory management. As confidence in forecast accuracy grows, initial inventory balances can be held at lower levels — while still meeting customer requirements.
Also, because AI and machine learning can process inputs so quickly, insight is delivered while it’s still meaningful. Action can be taken to meet the changing demand or availability of fresh or perishable items and partner or consumer needs. For example, what if during the pandemic, fresh produce and meat suppliers would have been able to more quickly repackage and shift delivery from restaurant wholesale distribution to consumer grocery outlets?
The United Fresh Produce Association estimated that the overall produce industry will lose at least 40% of sales during the COVID-19 outbreak. And while not all of this can be shifted to consumer grocery or retail, brands that can transition more quickly than their competitors are able to protect more revenue and jobs.
AI-driven forecasting platforms apply multiple algorithms to automatically determine the most relevant metrics for each product or SKU in the system. Time spent gathering, updating, integrating, and reconciling competing data from multiple spreadsheets is eliminated, allowing forecasting professionals to spend their time optimizing demand and coordinating with other departments such as marketing to find new revenue streams. The value of collaboration across departments and external partners, operating from a single source of truth, cannot be underestimated.
Results from the use of AI-driven models are compelling. Nounós Creamery recently employed such a tool to automate demand forecasting and decrease time spent on forecasting from two hours each week to just 10 minutes. Leaders at Nounós were frustrated by a lack of real insight, and were wasting valuable hours and money by manually comparing data from their accounting software and inventory-management system. This process only provided a vague sense of how much yogurt the company should manufacture, and which flavors to prioritize.
Nounós was able to pull in data from all relevant sources and rely on the algorithms to make forecast recommendations. The projections are so accurate that the company reduced overproduction by 40% almost immediately, saving significant revenue in product loss. It has also allowed Nounós to anticipate actual expected demand for each flavor, driving greater efficiency in the manufacturing process.
This new approach to demand forecasting is growing in adoption for financial reasons, but also because it’s highly scalable. It works just as well for specialty food brands as it does for large multinational brands. Consumer-driven demand forecasting is here to stay. Using the latest analytics and technology presents a tremendous opportunity to tame the bullwhip effect, and meet demand at the exact right point.
Are Traasdahl is co-founder and CEO of Crisp.
Timely, incisive articles delivered directly to your inbox.