Seasonal Indices, Anyone?

Seasonality is a fact of demand for many products. The demand changes with the time, however, displays the same general pattern of ramping up/down time after time. Such repeating demand patterns generally occur due to seasonality. It does not have to be annual, just cyclic in nature & repeated at regular intervals. To tell if demand is seasonal, simply look at the demand curve over a long enough period of time and if you see the same pattern repeated over time, it most likely shows seasonal pattern. How much history do you need to make a firm determination? Well – that is a question that depends on the length and frequency of the season. Usually, you should have data that shows more than three complete cycles of seasonal demand to be sure that seasonality exists. Based on the frequency and the length of the season, you may do fine with historic demand data for as less as 4-5 months or 3-4 years or longer to be sure that there is a seasonal pattern in the data. For example, if the pattern was monthly and showed a spiked demand in the first week of the month followed by a low-stable demand for the rest of the month, then the historical data for just a few months will be sufficient to establish the pattern. Now, consider a demand pattern that shows cyclic activity once a year, say in early summer that lasts for 45-60 days: then you will need 3-4 years worth of history to establish the pattern. 

Forecasting techniques normally de-seasonalize the historical demand data before projecting the demand into the future. The projected demand is then re-seasonalized to produce the final projected demand. Why go through all this trouble? Because, the de-seasonalized historical demand generally represents a time-series that is much more stable and lends itself easily to statistical projection. Chances are, that this process will enhance your forecast accuracy substantially.

So, how do you de-seasonalize data? That is not particularly hard either. The forecasting solutions find a base-line demand that is stable and then create a seasonal index on top of the base-line demand that recreates the seasonal cycle over the length of the season. The index is created using the historical seasonal demand over the base demand and can be reviewed frequently to make ensure the indices reflect the demand patterns correctly. The seasonal indices can be created by week or by month depending on the total length of the demand cycle and the build-up and decline patterns. In some cases, the indices may be created even by the day to reflect a high ramp-up of demand. 

Read more about the demand planning process here or click here to download the article on the process. This article was also published in Logistics Insight Asia.

 

Want to know more about supply chain processes? How they work and what they afford? Check out my book on Enterprise Supply Chain Management at Amazon. You will find every supply chain function described in simple language that makes sense, as well as see its relationship to other functions.

© Vivek Sehgal, 2009, All Rights Reserved.

 

Is it MAD or just a Bias?

Well, it depends on what is it that you wish to measure. Of course, we are talking about the forecasting accuracy measurements.

MAD stands for mean absolute deviation. This is an average of the absolute measure of error between the forecast & actual value without regard to whether it was overestimated or underestimated. Since this is an absolute measure, its value diminishes for any relativistic comparisons when the demand varies greatly among the products  or seasons. An absolute error of 50 over a  base demand of 500 may be just fine while the same error on a base demand of 100 probably needs some action!

Alternate measures like MAPD (mean absolute percent deviation) or MAPE (mean absolute percent error) present the error as a percent and therefore relate the size of the error to the base demand. This is an average the absolute error between the forecast & actual value computed as a percent.

MSE (mean squared error) is another measure used for measuring forecast accuracy. This measures the average of the squared individual errors. It measures the dispersion of forecast error, however, since it uses the squared values, it accentuates the larger errors more than the smaller ones.

The bias tracks the trending of the forecast and may signal problems with the forecasting algorithms or settings if the trend is significant. Therefore, this allows for correcting potential problems that may otherwise become substantial. 

 

Want to know more about supply chain processes? How they work and what they afford? Check out my book on Enterprise Supply Chain Management at Amazon. You will find every supply chain function described in simple language that makes sense, as well as see its relationship to other functions.

© Vivek Sehgal, 2009, All Rights Reserved.

 

Demand Planning: Essential to Your Success (Part 2)

If you had to pick a single process that has the largest impact on the company’s plans and operations, what would it be? Better pick demand planning since it is the starting point for a lot of processes that collectively make retailers hum.image_thumb12

[Click here to print or download this article]

In part 1 of this series, we presented various business processes that can benefit from a single source of projected demand. These processes included planning and execution functions in supply chain and merchandising functions spread over a long time horizon. The exhibit below provides a quick summary of the processes covered in the first part of this series.   imageIn this part, we present how retailers can proceed to create a single source of forecasted demand to drive their processes and align their functional plans with their operations.

Creating a Single Version of Demand

So far, we have seen that projected demand drives a number of diverse processes in a company. Given this versatility of the demand planning process and potential use of demand forecasts driving many other processes, one would think that companies will have a single forecast to ensure alignment among all the downstream processes. This however, remains a myth. In reality, companies routinely use many different demand forecasts to drive their processes for planning and execution. It is not uncommon to have different historical data as well as techniques used to generate forecasts that drive different processes. For example, the long term demand projections generated for merchandise planning are routinely an affair of a budgetary projection that reflects more of the firm’s financial growth targets rather than any statistically indicated growth trends. Same is true for aggregate demand projections used for network planning.This disconnect is only partly due to the lack of awareness. The other major reason is lack of proper tools for demand planning. Most corporations just do not have the right tools to maintain a single source of demand forecasts to address all the above processes. It is common to have a statistical demand forecasting application used for the execution processes, but then have a simpler, often subjective tool for addressing the needs of the mid and long-range planning processes that also need forecasted demand. This leads to inconsistent demand projections being used for different processes leading to plans that are misaligned with the operations. Lack of alignment between plans and operations causes misguided capital investments, infeasible plans, and unmet targets for revenue, profitability & budgets. They also lead to under or over capacity in the network, inventories, and the resources.To avoid this misalignment, firms must get out of the siloed mentality and ensure that functional plans that share critical inputs like projected demand are based on using a single source of truth for such data. This is not very hard to achieve if corporations are aware of the different functional requirements and implement a single solution for creating demand forecasts driving all their planning and operational needs. This can be achieved by establishing clear process goals and having a tool capable of manipulating the demand forecasts in many different ways to address the unique but closely related requirements of these separate business functions.

Establish Series of Forecasts Required

Establish what types of demand forecasts are required. Not all retailers need to create plans covering all the business processes, nor do they need to create them with the same objectives. For example, if your logistic operations are largely 3PL based, the changes in the network flow capacities can be accommodated with relatively short lead-times and a large capital outlay planning may not be required. In another example, if your distribution is largely based on cross-docking operations, the future plans should largely plan for increased number of shipping and receiving operations to accommodate growing demand rather than conventional warehouse storage. Therefore, the first step towards creating and using a single demand projection is to establish what processes are critical to a company’s continued operations and depend on forecasted demand.This means that well-defined requirements exist establishing the frequency of the forecast, level at which it will be generated, units for the forecast data, horizon definition, length of history to be consumed, data cleansing & enhancement pre-processes, and the consuming process for the forecast. This will help in evaluating the right solution and validating the feasibility of producing such forecasts from the single source of demand data.

Establish Functional Meta-data Standards

Next, understand how a single source of demand data will be modeled to cater to differing needs of individual functions. An important aspect for creating functional plans using the same demand data requires well thought out meta-data and master data models. Do the different business units and regions use common master data? Do they organize data using identical hierarchies and meta-data definitions? Make sure that all the functions use common definitions and understanding of the following meta-data structures and these structures fully address their needs.· Item master data and attributes· Item groups and hierarchies for aggregating and dis-aggregating demand data· Hierarchies for locations and organizational units· Time & horizon definitions· Retail and cost for items, discounting structures, and consistent definition for units of measure like cases, boxes, and pallets

Identify a Single Source of Demand

Identify a source of demand history that can be used for projecting demand for all functional areas. There are various options to think of: firms can use the actual point-of-sale (POS) data from the sales at stores or individual customer order data from other channels, outbound shipments from the warehouses, or receipts at the store. While the latter two may provide easier to implement processes to obtain demand history, the former usually is the best source of collecting demand data. Select a source based on the granularity of demand required and establish technology solutions to support the data being generated. If POS data is selected, remember that it needs to be collected from physically distributed stores across time-zones in relatively near-time fashion to support good-quality demand management processes.

Finally, Get a Tool that Works

Finally, get a tool that provides the flexibility to use a single source of demand history and create many different forecast series as required by the consuming processes. Of course, the solution must have the basic functional capabilities required for demand forecasting such as the ability to consume history & create forecasts: such capabilities are not part of this discussion. Check for the following capabilities to address the requirements of various processes that need demand forecasting data. What makes these solutions versatile is their ability to manipulate data, slice and dice it, and roll it up and down to create different views of the same data. Specifically, the following features create such capabilities that enable or prevent the solution from catering to all the processes mentioned above.

Dimensions and Attributes:

A good demand planning solution must allow modeling and working with the basic dimensions of demand. This also provides a flexible framework to manipulate data along these dimensions & create views that are most relevant for the process under consideration. Ask if your solution can model the three basic dimensions of product, location, and time in order to qualify the demand data and model attributes for the members of these dimensions that can then be used for quick analysis of demand by slicing and dicing the data. For example, product attributes like their sales velocity, style, targeted customer segment, or season provide good criterion for grouping and reviewing the demand at aggregated levels relevant to different processes. Having the ability to model such attributes and manipulate data using these attributes is an integral part of the solutions that would cater to different functional processes.

Hierarchies & Roll-ups:

A good solution must be able to define hierarchies for each of the main dimensions of product, location, and time mentioned above. For example, the hierarchy along the product dimension allows the users to create product categories and to aggregate demand along the levels of this hierarchy to look at demand by category, product group, department, and so on. The solution must also allow multiple hierarchical representation of the same underlying entity: this means that products can have a merchandising hierarchy that groups them together for use in merchandising processes, but they can also have an inventory group hierarchy to quickly manage the inventory levels. These groups are generally created by using the attributes and their values for the entity. For example, the inventory groups may be created by using an attribute that models the sales velocity of the products, while the merchandising categories may be a result of attributes like style, season, and the target customer segment. Having multiple hierarchies allows the users to aggregate demand data along different paths and analyze it for specific process needs.

Horizon Modeling:

The solution must allow flexible horizon modeling. This helps the users to construct a funnel-shaped horizon with finely defined time buckets for early time-periods and coarse time-buckets for future time periods. This makes the solution more responsive, faster to run, and allows for modeling longer time horizons as would be typically required for long-range planning processes. It also reduces forecasting errors by aggregating demand for the farthest time periods. For example, the immediate time periods can be defined as days, followed by weeks, followed by months and quarters. A planning horizon of a year defined with weeks will create 52 time periods, while a funnel shaped time horizon modeling first three months as 16 weeks, followed by 9 monthly periods and 2 quarters would actually allow a meaningful forecasting horizon extending up to 18 months into the future, yet having only 27 time periods for the forecast. The latter approach to modeling will be much more efficient for computing while both models will provide equal functional utility as long as supply lead-times for the products of the firm are less than 16 weeks. The shape of the funnel depends on the lead-time characteristics of the firm’s products and their procurement practices, but as long as the solution provides a flexible way of modeling, it can be implemented usefully.image

UOM & Conversions:

Finally, the demand planning solution must be able to model demand time-series in any relevant unit of measure. As mentioned above, some processes like replenishment need the projected demand data in individual product units, while others like merchandise planning needs the same data in dollars. The solution must allow for modeling multiple units of measure and provide the ability to convert from one unit to another. It should also be able to represent multiple time-series for historical and projected data since not all units of measure can be converted from one to the other. For example, when products with different physical units (one measured in meters, other in lbs.) are grouped together under a common merchandising group, they must be converted to a common unit such as dollars to make any sense: this can be achieved either by having multiple time series in dollars & other measures or by modeling retail price per unit and converting the sales in units to sales in dollars. In another example, if products are rolled up using their handling characteristics (conveyable and non-conveyable), their demand may be presented in cartons & pallets, or by weight or volume. If the solution allows for such modeling flexibility and easy conversion from one unit to another, it allows itself to be leveraged in different functional contexts required by different processes.

Summary

Demand forecasting caters to many organizational processes that are spread across the time horizon and functional boundaries. To ensure that long-term organizational plans are aligned with the short-term operational objectives and the processes across functional boundaries support each other, it is imperative that companies implement demand planning solutions that will allow them to create a single demand forecast to drive these processes. Such a forecast must use a single source of historical demand and forecasting techniques that use similar assumptions. This cross-functional alignment in plans and operations will establish process synergy, reduce plan conflicts and volatility, and create operational stability that otherwise remains elusive.In part 1 of this 2-part series, we presented the business processes that require forecasted demand to create plans that support everything from supply chain network capacity planning to every-day replenishment operations. These processes span across time and functional boundaries. We also presented how their requirements for projected demand differ by horizon, data granularity, and units.In this part, we conclude this series by presenting how companies can break the functional silos to create a single source of demand forecasts to support their plans for different processes and ensure functional alignment as well as operational stability as a result. This requires careful planning and the right tools as discussed in this concluding part 2 of our series. © 2009 Vivek Sehgal, All Rights Reserved

Demand Planning: Essential to Your Success (Part 1)

If you had to pick a single process that has the largest impact on the company’s plans and operations, what would it be? Better pick demand planning since it is the starting point for a lot of processes that collectively make retailers hum.

image Demand planning consists of processes that allow a retailer to forecast demand into the future and manage it. It has the key dimensions of product, location, and time to clearly identify the projected demand. Once the projected demand is available, it is used to drive all types of functions for a retailer. Depending on what function is the projected demand expected to drive, it may be expressed in different units, such as dollars or boxes or lbs., different levels such as product categories or individual SKUs, different organizational units such as merchandising departments or regional markets. Owing to these many uses of projected demand, the process is sometimes segregated into departmental boundaries that create their own demand projections. This practice leads to poor coordination among the organizational units, produces inconsistent results, and leads to wasted opportunities in optimizing operational costs and efficiencies.

Processes Impacted by Demand Planning

There are many processes that use projected demand forecasts. They span from long-range planning processes like network design that have a horizon of a few years to execution-level processes like replenishment with immediate impact on operations. Using the time horizon as the basic context, these processes can be divided as follows.

Click to see complete image

Long Range Planning Processes:

These processes use demand projections to plan for the supply chain flows within the network. They need aggregated demand forecasts for a longer horizon and may not have any immediate impact on the firm’s operations. Examples of such processes are supply chain network design and network capacity planning. These processes help answer questions like: How much volume of product will be flowing through the supply chain network in the projected years? Where does this flow occur along the existing network routes? Are new routes required? How is the current network poised to handle this projected flow of product volumes? Do the warehouses have enough storage capability and operational resources to support the projected volume of flow of merchandise? Is there adequate transportation capacity available along all main network arteries? They are designed to evaluate the network flow capacities that will be required to support the firm’s projected demand volumes. Any changes in the network have a long lead-time for implementation and require substantial capital investments: whether it consists of increasing the capacity by opening new facilities or through automation; or reducing the capacity by closing existing facilities or changing the facility locations for more optimal flows. Given these long lead-times and the need for large capital layouts, these evaluations are generally done years in advance. However, they only require aggregated projected demand in terms of number of number of cases, pallets, volume or weight of the flows expected for the projected demand. Demand projections for these processes can be directly created at these aggregated levels since they tend to be relatively more accurate over the longer horizon desired for these processes.

Click to see complete image

Mid-range Planning Processes:

These processes use the demand projections for revenue and budget planning. They need projected demand at a more granular level than the processes above. Examples of such processes are merchandise financial planning and product portfolio planning.

These processes generally work with projected demand for product categories in dollar value of the merchandise at monthly and sometimes weekly levels. The objective of the merchandising processes is to develop the merchandise plans and create targets for revenues and profitability, and planned budgets for promotions, clearance, marketing, and procurement of the merchandise. The portfolio planning processes generally evaluate the projected profitability using the projected demand and arrive at optimal assortments (product mix) for the projected plan horizon. Demand projections for these processes are generally created at lower levels and rolled up for use in these processes.

Click to see complete image

Short-term Execution Processes:

These processes use the demand projections for supporting immediate operations of the firm so that customer orders can be fulfilled and stores are adequately stocked with the right merchandise at the right time. There are many processes that take advantage of the short term demand projections for this purpose, such as inventory planning, purchasing, receiving, storage, store fulfillment, and inbound & outbound shipping for the warehouse. Most of these operations need very granular demand forecasts at item and facility level, often in daily or weekly buckets. Other processes that also benefit from this level of projected demand are price optimization, promotions, clearance, and seasonal product life-cycle events. Demand projections for these processes are created at lowest grain of product and location often in expanding time buckets along the horizon, for example, the forecasts may be produced on a daily basis for next 2-3 weeks, weekly basis for the next 2-3 months, and monthly basis thereafter. Supply lead-times generally affect the length and size of the time horizon of demand forecasts created for these processes.

Click to see complete image

Summary

Demand forecasting caters to many organizational processes that are spread across the time horizon and functional boundaries. To ensure that long-term organizational plans are aligned with the short-term operational objectives and the processes across functional boundaries support each other, it is imperative that companies implement demand planning solutions that will allow them to create a single demand forecast to drive these processes. Such a forecast must use a single source of historical demand and forecasting techniques that use similar assumptions. This cross-functional alignment in plans and operations will establish process synergy, reduce plan conflicts and volatility, and create operational stability that otherwise remains elusive.

In this part 1 of this 2-part series, we presented the business processes that require forecasted demand to create plans that support everything from supply chain network capacity planning to every-day replenishment operations. These processes span across time and functional boundaries. We also presented how their requirements for projected demand differ by horizon, data granularity, and units.

We will conclude this series in part 2, where we will present how companies can break the functional silos to create a single source of demand forecasts to support their plans for different processes and ensure functional alignment as well as operational stability as a result. This requires careful planning and the right tools and we will discuss both these requirements in detail in our series concluding part 2.

© 2009 Vivek Sehgal, All Rights Reserved