零售和CPG的顶级时间序列预测挑战和解决方案

Discover the ways that SoftServeexperts address the biggest time seriesforecasting challenges that faceretail and consumer product goods(CPG) companies Demand forecasting is the foundation forsuccessful retail and consumer productgoods (CPG) enterprises. The impact ofdemand forecasting is felt at every stageof the supply chain—from operations tosales and marketing—and can make thedifference in efficient ordering, effectiveproduction planning, and deliveringthe items your customers require. Over the last few years, ourSoftServeretail teamhas seen a spike in interestfrom retailers and CPG companies on howto improve their demand forecastingprocess. Having successfully deliverednumerous forecasting solutions tailoredto a variety of companies, the teamidentified the top pitfalls. Here our experts offer their insight intoboth the top issues that hold retail andCPG companies back from accurateforecasts and how to solve thesechallenges. 1. Connecting the BusinessProblem to The TechnicalSolution THE CHALLENGE One of the biggest problems in time seriesforecasting is how many retailers and CPGcompanies misunderstand—or even missentirely—the connection between thebusiness value they hope to achieve andthe technical solution they want to employ. THE SOLUTION In such a situation, the solution is toreframe the problem. Companies can be soeager to run their historical sales datathrough state-of-the-art machine learning(ML) algorithms that they overlook otherkey opportunities that provide value. One example is how these elements helpedclothing retailers successfully launch newstores. The SoftServe team incorporatedstore demographics and location into theforecasting. This factor enabled theretailers to maintain the correct levels ofrelevant inventory, such as having morecold-weather items available for new storesin colder climates. SoftServe’s retail team has seen thisfirsthand when implementing solutions thatforecast more than just sales. In manysituations, our experts also providedaccurate inventory and labor needsforecasts. They did so by incorporatingrelevant variables such as complementaryproducts, store demographics, companyliquidity, and supply and logistical trends. Another value-add situation is how ourexperts have supported several big-boxretailers. For them, forecasts ofcomplementary products can lead to betterinventory maintenance. For example, say arecently released video game is sellingparticularly well. Predicting the likelyincreased sales of complementary productslike video game consoles and accessories—and aligning the extra inventory required—is extremely valuable. For many retail and CPG enterprises, usingforecasting to predict more than simplysales transforms their internal planning andbudgeting and ultimately leads to higherrevenue and growth. 2. Forecast Granularityvs. Accuracy THE SOLUTION THE CHALLENGE Unfortunately for such situations, it's a factthat the greater the granularity, the loweryou must set your forecast accuracythreshold. Failure to lower your forecastaccuracy threshold increases the chancesyou will end up overfitting on your results. Another major issue for retailers and CPGcompanies is the battle between forecastaccuracy versus granularity. While allenterprises don't need highly granularforecasts, some—such as small fulfillmentcenters known as "dark stores" locatedthroughout a city and within 15-30 minutesdelivery time of target clients—couldbenefit significantly. These dark storeswould see a significant impact withaccurate forecasting down to the smallestscale of an hourly and even quarter-hourlybasis. When a statistical analysis fits too closely—or matches exactly—with a particular dataset, it's known as overfitting. Overfittingcauses problems because it makes the dataanalysis too specific and potentially unableto reliably predict future events. Too many companies will overfit their short-term and store-specific forecasts becausethey are seeking very granular results whilestill trying to reach 95% or greater accuracy.While striving for the highest accuracylevels makes sense, in these situations itcan be detrimental in the long term.Ultimately, overfitting reduces businessvalue since the resulting forecasts struggleto correspond to additional data andtherefore fail at their entire purpose—providing predictions. It might seem counterintuitive to lower youraccuracy threshold, but adopting a modelbased solely on testing accuracy won’tgenerate the required results. To get thebest outcome for your specific needs, it’simportant to focus on simulating anddriving real experiments. This will allow youto calculate the business impact of differentforecasting approaches and find the resultsthat fit your business best. 3. Global Supply ChainDisruptions THE SOLUTION THE CHALLENGE There are two potential issues in anyforecast of supply chain changes. First arethe known-unknowns—such as hurricanes,earthquakes, cyber-attacks, or countryimport and export regulation changes—that c