Intelligent automation for demand planning
Businesses need a unified real-time planning environment to optimize planning and democratize decision-making across lines of business and business activities, from strategic to operational levels. Machine learning (ML) can help increase the planning forecast accuracy. However, ML model outputs are not yet easy to use directly in business planning organizations. In this blog for manufacturing industry business decision-makers, we introduce how the business planning team can dynamically plan for customer demand. This includes discovering demand patterns, predicting demand combining human and machine intelligence, and enabling teams to meet business performance goals.
Investments in agile manufacturing, Industry 4.0, and IoT are driving transformation of manufacturing operations. But these investments require accurate demand planning and demand sensing. Customers are demanding shorter lead times and more transparency to order fulfillment to set, and achieve, accurate delivery dates. A limited view of demand guides the prioritization of production, leading to lost productivity, poor order fulfillment, and lost revenue. In this article, we discuss the traditional challenge, advantages of addressing these challenges with ML-driven forecasting, and finally touch on the Amazon Forecast powered Anaplan PlanIQTM platform which works on top of the core Anaplan offering. This is a purpose-built platform solution powered by Amazon Forecast which is now available to connect people, data, and plans.
Amazon has designed and manufactured smart products and distributed billions of products through its globally connected distribution network using innovative automation, AI, and robotics, with AWS at its core. Industry customers can take advantage of this in addressing their own needs. See the AWS for Industrial page for a broader view. This blog specifically highlights intelligent automation for demand planning backed by ML-based forecasting on AWS enabled by Amazon Forecast. Based on the same technology used at Amazon.com, Amazon Forecast uses machine learning to combine time series data with additional variables to build forecasts. Customers use Amazon Forecast to tackle forecasting problems that require increased accuracy in the face of real-world unknowns.
Traditional demand planning challenges for manufacturing industry customers
Being a Manufacturing business leader means that you will need to balance cost and customer service to accurately predict and shape true customer demand through forecasting processes. This can be challenging for a variety of reasons. First, the factors that drive customers to purchase aren’t typically clear. Secondly, there is limited visibility into the impact of price, promotion, product-mix, placement and process. It is challenging to align sales, channel partners and operations to drive the desired outcomes. As a manufacturing leader you know first-hand, the problem with calculating demand based on the sales forecast is that these estimates do not evolve from an accurate sales funnel picture. The alternative relies on an operational assessment of what will happen. This approach relies on historical production requirements, which do not factor in dynamic market changes. Finally, even when using ML and statistical algorithm-based forecasting tools, these are disconnected from the planning processes, and require significant effort and technical expertise like data scientists to run the forecasting process. The impact of these challenges is that products are produced in the wrong quantities. Inventory is shipped and sent to sub-optimal locations. Missed revenue occurs due to lack of inventory, or markdowns because of excess inventory. Equally and most importantly, forecasting error associated with planning ensues in poor customer experiences.
Forecasting for demand planning with ML
Working with manufacturing industry customers, we observe three key needs in building business-friendly incorporation of ML forecasting. The forecasting needs to drive more accuracy, be accessible, and be highly scalable. The accuracy needs include nearly continuous improvement of predictions by leveraging new and existing data. Accuracy needs to be improved through adoption of ground-breaking Deep Learning ML techniques such as those available in the Amazon Forecast ML engine. The ML setup needs to include more test forecast scenarios, functional features to find correlations, and ability to pinpoint new trends. Inferences need to be more accurate leveraging the historical data and forward-looking metrics. The best predictive model to use needs to be automatically selectable.
Business users need to create and run forecasts, this is the main pain point around accessibility. Manual preparation should be avoided in favor of automatic conversion of data to a Forecast engine-friendly format. Business users should find the ML-based forecasting configuration and analysis easy to adopt. Finally, the results must be easily understandable and usable in actionable plans. A nearly continuous learning loop is needed to grow, adjust, and improve the forecast to scale ML driven forecasting. Data migration into the solution/tool needs to be automated. Direct ingestion of internal operational and external data such as weather and macroeconomic data is needed on a nearly continuous basis as well.
Intelligent automation of demand planning
Intelligent automation does not require root-and-branch replacements of existing software given the broad solutions and services available to work and optimize the customer’s existing digital investments. One example of intelligent automation for demand planning is the out-of-the-box purpose-built platform from Anaplan. It can support nearly continuous forecasting, enables dynamic, collaborative, and intelligent planning. The customer starts with setting up PlanIQ with the Anaplan Demand Management model. The Forecast process powered by Amazon Forecast is then run. Deployment planning follows metric measurement. Resulting predictions help business decision makers get a deeper understanding of future demand drivers and the impact of their commercial decisions with a what-if lens. In the post-pandemic world, as new market patterns emerge, incorporation of external data becomes even more important.
Real-world benefits of demand planning done right
For discrete manufacturing, like apparel manufacturing, most products have short (3-4 month) life cycles. In a real-world example we observe enormous volumes (200,000+) of Stock Keeping Units (SKUs), 500 Million items/year, with thousands of retailers and typically no existing holistic view of the supply chain. Typical outcomes include reduction of Millions of dollars in inventory carrying costs and 10%+ decrease in excess inventory. For top-end home appliance manufacturers we observe a focus on improving forecast accuracy for SKU demand with their retailers and reducing the effort to create forecasts to less than a week. Using 10,000 SKU-location combinations and 3 years of historical data, the resulting forecast accuracy could be significantly improved and delivered in a day.
Demand planning as discussed in this blog article applies to process manufacturing as well. For a steel manufacturer, the key challenge was trying to predict when orders would come out of production. This was hard to do due to order configuration long lead times coupled with low visibility to available production capacity to take on new orders. The outcome driven was reduction by 20% in the amount of raw steel stock waiting to be configured into sellable products and reduced waste freeing up cash flow for other purposes.
We invite you to take advantage of intelligent automation for demand planning that incorporates the latest deep learning techniques powered by Amazon Forecast. Contact us to learn more about this as well as how AWS can help grow your Manufacturing or Industrial business.