Our partner Celestica recently published the following article, ‘What If You Could Take The Guesswork Out Of Forecast Planning?’. The author, Osgood Vogler, Director, Analytics, Celestica Supply Chain Managed Services, describes an insight-based demand management process: So, how do you take the guesswork out of forecast planning? Let’s find out.
Demand planning has a big impact on business performance. Planning error can put revenue at risk by driving component shortages. Persistent planning biases can tie up cash by driving excess inventory. Furthermore, the act of planning and dealing with planning error is time consuming and drives costly overhead. In fact, it is common for supply chain management executives to cite "planning errors" as the greatest obstacle they face to achieving their goals and objectives.
The factors which impact demand management and forecasting are nearly endless. Uncertainty in end markets, shifts in the competitive landscape, constant time-to-market pressure, economic volatility, geopolitical and environmental issues all play a role in component shortages, excess stock and lost revenue.
Given this volatility, it is not surprising that organizations are struggling to make effective demand predictions. To avoid the financial risks associated with planning errors, supply chain leaders and manufacturers should consider building an “insight-based” demand planning process, which brings together analytical tools and data with key human inputs across various functions. This “next generation” demand management approach will allow supply chain operations to evolve and scale with the ever growing volatility and uncertainty of today’s markets. The insight-based demand management process contains several key principles.
One size does not fit all
One solution is never going to address every challenge an SCM executive will face, so it is important to determine the best approach for your supply chain through segmentation. One planning approach may work well for one group of parts but not for another. Segmenting parts in a supply chain is incredibly useful to help guide the development of a cohesive demand management strategy.
There are three questions that are central to the demand segmentation:
- Why is planning necessary?
- How important is the part to your business?
- How predictable is the demand?
Several considerations will likely go into answering each of these questions. For example, to answer the first question about whether planning is necessary, SCM executives need to determine if supply is constrained and how quickly customers expect their order to be fulfilled. If planning is absolutely necessary because supply of a particular part is constrained, an organization needs to determine how critical that part is to the supply chain, what profit margin is realized from the sale of the part and whether the demand is predictable across related parts and products. This exercise is important because it will help supply chain leaders understand exactly where planning is necessary and how to drive exceptional performance in their supply chain operations.
Measure where it matters
Defining what actually needs to be predicted to effectively manage a supply chain is a requirement for accurate and efficient demand planning. While prediction accuracy is often measured at the lowest level of granularity, such as by item, customer or region, these factors may not actually matter as much as prediction accuracy at a higher level. For example, the overall demand accuracy by part type at a regional distribution center may be more important to supply chain performance than item-customer-region level accuracy.
To accurately judge one approach versus another, the primary metric for evaluation purposes may need to be established at a different level. For example, if a planning process needs to determine “how many widgets do we need?,” the answer might be “we know we need 1,000 pieces.” However, if the demand planning process needs to determine “how many widgets of each color do we need?,” the answer might be “we are not really sure, say 600 black and 600 blue.” In this scenario, a forecast bias was created and it led to an order of 200 additional widgets.
To eliminate these low-confidence guesses and move toward a more informed demand forecasting process, the inputs used to generate a plan should be carefully selected. Some common examples of guesswork in the demand forecasting process can include systems forcing planners to input forecasts at granularity that is lower than what can reasonably be estimated and sales teams tasked with translating customer intelligence directly into a demand plan.
Guesswork should never be hard-wired into the demand management process. The best results are most often achieved through human knowledge of the market and customers behavior coupled with analytics such as data on observed patterns, market trends and dynamics.
Find the right blend
Effective demand management requires a blend of two perspectives. The first is the customer’s perspective “looking outside in” at an organization's products and the second is the supply chain’s perspective “looking inside out” to the supply base. Understanding how the customer’s needs, wants and behaviors translate into demand is just as important as understanding what is known and/or knowable at different points in the marketing, sales and supply chain cycles. Human wisdom combined with analytical insights need to be operationalized and integrated into a cohesive process.
For example, what your sales and marketing teams really know about customers in end markets at various points during the sales cycles needs to be captured and leveraged effectively. Furthermore, shared parts and bill of materials (BOM) commonality may present opportunities to generate more accurate and meaningful aggregate forecasts for the supply base. For instance, if two parts have BOMs that are 80 percent in common, it may be more effective to forecast the common parts separately from the unique parts.
Always keep a running score
Of course, implementing an insight-based demand management process structured around an understanding of key insights from human wisdom and analytical data is not a “set it and forget it” decision. Segmentation questions and criteria evolve with the business. Modeling and collaboration are ongoing activities. What is now a “guess” may become a “known” and what is currently “known” may become “unknown”. Scoreboards keep us honest and drive constant evolution. Insight-based demand management never stops.
Looking for more great information from Celestica? Check out these other blogs in our series:
[Video] Celestica’s Top Priorities for Improving Forecast Accuracy
Forecast Accuracy: Keep Your Demand Management Process Honest
Control Tower Success: Six Critical Steps to Ensure Your Project Thrives
- Demand planning frequently asked questions
But, more importantly, I think, is to determine what the forecast is going to be used to plan.
As the writer correctly intimates in the article, aggregate forecasts are likely to be more accurate that forecasts at lower levels (e.g.: "We know we need 1,000 pieces" versus “we are not really sure, say 600 black and 600 blue").
Forecasting, therefore, is pretty good for planning CAPACITY, but pretty bad for planning STOCKING LEVELS. Nevertheless, the inherent frailty in forecasting and the its resulting damage can be mitigated against using forecasts at the highest possible levels for planning STOCK, and then changing planning and execution around two other factors: (1) agility and (2) positioning.
Companies always respond with greater accuracy to the market when they substitute RESERVE CAPACITY (both in production and logistics) for RESERVE STOCKS (inventories). This is because, as I said, forecasts are great for planning CAPACITY but pretty bad at planning STOCK LEVELS for individual products, styles, colors, sizes, and sales locations.
This also means that keeping STOCKS at the highest levels as long as possible (where forecasts are more likely to hold greater accuracy) and employing RESERVE LOGISTICS CAPACITY to replenish take-away locations more frequently, makes a supply chain inherently more accurate and more responsive in its execution. This has many advantages that I will not take time to list here.
I guess what I am saying is this: I don't think one can take enough of the guesswork out of forecasts to make the truly reliable. We all, I think, agree that forecasts are going to be wrong (in the final analysis). However, one CAN take much of the guesswork out of EXECUTION by changing how one uses forecasts for planning--and changing more of what you plan to those things that will make a difference.
Thanks for listening to my opinion.
Horizon 1: The design and structure of the supply chain network itself. Having flexibility designed into the supply chain via shorter LTs, alternate sourcing options, logistics options and other strategic vendor relationships can pay great dividends. Strategic buffers of both inventory and capacity can help the system cope with variation in demand and/or forecast errors (but also come with a cost).
Horizon 2: The planning window. In any supply chain where there are supply constraints and/or lead time constraints, commitments are made to drive materials ahead of actual firm demand from customers. The sequence when loading up a plan should be "ready, aim, fire" not "ready, fire, aim".
Horizon 3: React well during the execution window. Robust systems that enable E2E visibility and real-time what-if analysis definitely help supply chains mitigate the impacts of planning error and drive execution performance in spite of planning errors. The more accurate the plan, the less mitigation efforts are required but because plans are not perfect, mitigation is critical.
On Richard's point about statistics, statistical forecasts are indeed derived from assumptions. For a given set of data and a given set of parameter settings, statistical forecast outputs themselves are the result of calculations. Practically speaking, there is uncertainty in the choice of parameters fed into a statistical algorithm.
Beyond just parameters, there is even a boundary of uncertainty around the choice of algorithm used to generate a statistical forecast. In our experience running extensive simulation driven optimization of statistical algorithms, the choice of algorithm is often a critical choice. Some algorithms in given situations may produce significantly biased results. Armed with the ability to simulate the performance of algorithm libraries against a data set, data scientists can make more informed decisions about statistics and thereby take "guesswork" out of even the statistics - or more precisely, the choice and configuration of statistical algorithms.
Thanks again for the comments.
The explanation of the word guess is pretty confusing, but giving a forecast isn't necessary a guess because statistical methods are being used.
Leave a Reply