What makes for an accurate prediction? Good data and the right methods for modeling it. The supply chain industry has both. What’s often missing is a framework for developing and scaling supply chain forecasting solutions beyond the basic statistical models.
However, with the commoditization of AI technologies — machine learning, deep learning, and predictive analytics — access to the right modeling methods is no longer a constraint. So what’s feasible when it comes to supply chain forecasting? How far can you see and how fast can you go with adoption? Let’s take a close look beyond the buzzwords.
A technical framework for supply chain forecasting
Supply chain management (SCM) produces a wealth of big data you can use to support your decisions. The problem is that decision-critical data is stashed in the wrong places and out of view — in suppliers’ systems, on-premises databases, and cloud repositories.
For that reason, forecasting supply chain processes is hardly possible without first sorting out supply chain visibility.
Last year, only 9% of supply chain managers had visibility into upstream and downstream networks and emphasized data sharing with partners. That’s problematic, as you cannot deploy demand forecasting in supply chain management if your partners don’t share their data on inventory levels, sales volumes, and replenishment plans.
A solid 60% of SCM leaders plan to improve their levels of data sharing with their ecosystem partners in 2021 and onward:
Level of data sharing – current with plans to increase
Source: Capgemini — Fast forward: Rethinking supply chain resilience for a post-COVID-19 world
Data sharing and visibility are in place. What’s next? How do you go from being able to review and interpret data from your partners to exercising predictive demand planning and forecasting in supply chain management?
In short, you’ll need to learn how to collect the right type of data and then dispatch it securely for predictive analysis.
At Intellias, we’ve worked out the following technical framework for developing supply chain forecasting solutions:
- Determine and connect the required data sources for analysis
- Create a secure data management and governance framework
- Select the optimal set of demand forecasting methods in the supply chain
- Launch supply chain analytics pilots
- Scale towards adopting intelligent planning
Why the right data is crucial when implementing forecasting in supply chains
Demand forecasting is a staple of supply chain management. Everyone’s doing it in one way or another to create realistic production plans and inventory turnover projections. Most supply chain managers rely on a mix of historical transactional data and direct customer insights from sources such as surveys. Customer intelligence and other qualitative insights are particularly valuable when you’re planning to launch a new product line or expand in a new market.
But the problem with such “self-confessed” data is its accuracy: people are faulty forecasters.
The best example of this is New Year’s resolutions. Over 75% of people will abandon their resolutions in 30 days, while just 8% will manage to accomplish them. When demand planning in supply chains is based on self-reported data and managers’ judgments alone, the chances of operational mishaps are also high.
Take Shoes of Prey, for example — an Australian-born and globally scaled manufacturer of customizable footwear. Until 2018, the startup was rapidly acquiring customers who were enticed by the ability to design their own pair of shoes. The company had lean operations, didn’t burn too much capital, and had an incredibly high net promoter score. Their audience loved their products and the design experience. Before entering the mass market, Shoes of Prey did a bunch of market surveys to project the demand for customizable shoes and prepare their manufacturing facility to meet shorter lead times.
But several months into the mass market expansion, the team learned that what their consumers were consciously saying and what they subconsciously wanted were diametrically opposed, and the company ultimately failed.
While there were strong early signs that the sizing and short-run manufacturing markets might work for us, we weren’t able to clearly prove that these customers were willing to pay us enough at a large enough scale to cover our fixed costs.
The takeaway: Qualitative data and market knowledge can contribute to demand prediction. But it’s quantitative data that helps you get a more realistic picture of the actual state of affairs.
When customers say they hate the taste of a new beverage but your sales data suggests those awful-tasting drinks are selling like crazy (because of the publicity), you may want to hold off pulling them from the shelves.
What are good data sources for supply chain demand planning?
You should collect a combination of:
- ERP data: Demand data, manufacturing capacity inventory data, production volumes, stock data.
- Logistics data: Delivery mismatches, warehousing analytics, delivery routes, traffic data, weather conditions.
- Transactional and sales data: POS data, customer auction costs per channel, customer lifetime value, etc.
- Data from external sources: Market trends data, weather data, GIS data, competitor analysis, etc.
Supply chain big data sources: volume and velocity vs variety
Source: KPMG — Supply Chain Big Data Series Part 1
Establish a data governance and data management framework
The second step for applying data science to supply chain and demand planning is getting a good grip on data management practices. More data integrations means more responsibilities for securely storing and managing data.
A data governance system establishes authority and control over all corporate data assets and the conditions for using them. Such a system includes people, processes, and technologies for storing, managing, using, and protecting assets.
The components of a data governance framework include:
- Data architecture: A technical structure for storing and processing data as a core part of the overall software architecture
- Data modeling: Documented processes, methods, and approaches to data analysis, model design, development, testing, and deployment
- Data storage: Technical and operational facets for securely storing corporate data locally and/or in the cloud
- Data security: Means deployed for ensuring privacy, compliance, and appropriate access to all-purpose and sensitive data
- Data integration and interoperability: The process of acquiring, cleansing, transforming, federating, and replicating data
- Data quality: Standards and procedures for defining and ensuring data integrity and enhancing data quality
Data governance establishes standardized processes and policies for using big data within your company. Data management, in turn, enacts these in an operational setting. Both systems are crucial for implementing predictive analytics solutions.
Learn about the implementation of cloud technology for accurate routing and smart planning
Select the optimal demand forecasting supply chain methods
There’s no lack of statistical and machine learning models for supply chain optimization. So which should you settle on? The answer depends on your current levels of:
- Supply chain visibility
- Data management maturity
Ultimately, it’s the available data sources that will dictate your choice of supply chain forecasting models. For example, if you can only produce CRM transactional data, even the most sophisticated supply chain forecasting software won’t be able to tell you how delivery timelines or weather conditions impact customer demand.
On the other hand, if you have sufficient data but cannot formulate a precise (and feasible) business case for predictive analytics, you won’t get far either. For example, provenance supply chain data can be effectively used to analyze the sustainability of your operations and help estimate changes in carbon emissions if you opt for alternative sourcing and transportation strategies. At the same time, it’s not the best data asset for forecasting product demand.
What are the best methods for supply chain forecasting?
- Time-series forecasting using recurrent neural networks (RNNs)
- Vector autoregression (VAR) for multivariate time-series data
- Auto regressive integrated moving average (ARIMA) for univariate time-series data
- Multivariate time series forecasting using random forests
Sounds intimidating? The above are examples of machine learning and deep learning algorithms developed by data scientists for specific projects or operating as business intelligence solutions within proprietary demand management supply chain platforms.
In this post, we’ll look closer at building custom supply chain forecasting models.
Launch a supply chain analytics pilot
Forecasting solutions (predictive analytics) help you answer questions of the type What will happen next? For example: Can we effectively meet the ETA targets for all deliveries if this holiday season we make 20% more sales?
Don’t be afraid to go after granular problems. Even minor improvements can have a positive ripple effect across the entire supply chain. Per McKinsey, learning about the causal drivers of demand in retail supply chains, for example, can lead to a 10% to 20% improvement in forecasting accuracy, and subsequently a 5% reduction in inventory costs and a 2% to 3% increase in revenues.
Once you have a specific problem statement to explore, determine which data sources, analysis methods, and technologies you’ll need to get the answers you’re looking for.
At Intellias, our go-to tech solution for predictive analytics is AWS SageMaker — a cloud-based machine learning platform by Amazon. SageMaker provides an integrated development environment (IDE) that’s preconfigured to support the development of end-to-end machine learning models, assisting with everything from data management to model deployment to production. Another advantage is seamless access to AWS GPU/CPU resources for model training.
Bonus point: AWS also has a managed business intelligence service (Amazon Forecast) for obtaining time-series forecasts. In this case, you pay to use Amazon’s models instead of building your own.
The pinnacle: Adopting intelligent planning
Intelligent planning is the veneer of the technical credenza of supply chain digitization.
Intelligent planning stands for the application of advanced predictive solutions for data-driven decision-making in supply chain management
In other words, intelligent planning enables you to:
- Use real-time data to forecast demand by channel and market
- Analyze different sourcing strategies to determine the optimal supply scenarios
- Collect, visualize, and optimize the movement of different goods and assets
- Optimize logistics routes in real time
- Create supply chain digital twins for advanced modeling scenarios
- Track and manage inventory levels across markets, suppliers, and distributors
- Respond to emerging market trends by adjusting production capacity and stock levels
- Locate new opportunities for making your supply chain more agile and resilient
- Produce accurate provenance data from upstream and downstream partners
- Respond to emerging compliance regulations and government agendas (e.g. a potential Green New Deal)
Understandably, all of the above are ambitious targets, requiring both operational and technological maturity. Can you immediately leapfrog to intelligent planning? Not without completing the previous steps of excelling in data management, big data analysis, and supply chain visibility.
But should you aspire to achieve intelligent planning? Absolutely. Test and scale supply chain forecasting pilots. Explore additional data sources, then enhance your algorithms to predict a wider range of scenarios.
Pave your way towards implementing a supply chain forecasting solution with our deep industry expertise in big data and analytics. Contact Intellias experts to get transparent view into your supply chains future.
