Author: Dr Keith McNeil

Introduction

Demand planning has become one of the “hot” buttons of supply chain management.  On most Saturdays, advertisements appear in the main capital city newspapers for Demand Planners, something that was far less common as recently as five years ago.  An internet search will demonstrate an absolute plethora of software packages for forecasting and demand planning.

However, a large number of companies still lack a demand planning process.  The most common reason given is that “we know forecasts will be wrong” or “ours is an industry in which it is impossible to forecast”.  These statements roll off tongues as flashes of the bleeding obvious and immediately are used to justify moving on to do something else and neglecting a critical supply chain activity.  Notwithstanding these criticisms of forecasts, most businesses in the end are driven by a series of planned sales numbers.  Some, unfortunately, are driven by several series of numbers, e.g. production’s view of sales demand as opposed to marketing’s view.  Surely, the sensible action is to have a process to develop one agreed sales forecast involving all key stakeholders.

This article describes the importance of forecasting, the importance of structure in a demand planning process and some relevant issues in selecting forecasting software technologies.

Why Forecast?

Forecasts of sales demand should be the key building block from which to plan a supply chain.  One agreed forecast for a business drives financial planning, sales and marketing, production planning, distribution planning and procurement.

Given that we all understand that forecasts will be wrong, why is a forecasting process so important?  There are two main reasons.

The first reason is the need to develop one agreed set of numbers as a basis for the business plan.  For some years, supply chain professionals have been aware of the bullwhip effect.  This is where even minimal variability in customer demand is interpreted by upstream participants in the supply chain with increasing volatility, with the volatility increasing as the participant becomes more remote from the end-customer.  Simulating this effect is the foundation of the well-known “beer game” that is played at management retreats.  The reason for the effect can be attributed to individual supply chain participants second-guessing what is happening with ordering patterns, to batching of orders by some participants and to miss-reading of promotions.  If the various participants strive to get closer to customers through collaborative forecasting, then this bullwhip effect is greatly reduced with consequent minimisation of inventory accumulation and stock-outs.  Similarly, within organisations, working with one agreed demand plan through a sales and operations planning process has the same effect.

The second reason for having a forecasting process is that routine monitoring of the variance between forecast and actual sales should drive safety stock policies for the business.  This in turn assists in setting customer service levels at the desired levels, reducing required investment in inventory, minimising obsolete stock, production planning and scheduling and overall improved profitability.

The impact of implementing a forecasting and inventory management system at Longs Drug Stores in the U.S. ($3.7 billion sales; 400 retail outlets) was assessed by Lee and Whang (2001).  Over three years from 1997 to 2000, store inventory value fell by 38% ($90m), inventory at distribution centres fell by 65%, sales increased by 20% and product availability increased to 99%.  This is an excellent and well-documented case showing the benefits of forecasting and inventory planning.

The Golden Rule is that “Increased demand forecast accuracy translates into increasing perfect order fulfilment”.

Principles of Forecasting

1. Forecasts are more accurate for shorter time horizons.  This means that one way for a business to increase relevance and effectiveness of forecasting is by reducing lead times as much as possible through elimination of non-value adding process time, reducing response time of the process itself and reducing set-up times and economies of scale to drive for more flexible processes.
2. Forecasts are more accurate for larger groups of items.  For example, a brewer can produce more accurate forecasts for a brand of beer than for the individual package types of that beer.  Further, the brewer can develop a more accurate forecast at a state level than it can for each demand location or each distribution centre in that state.  Hence, organisations should consider options for producing forecasts at an aggregate level but have greater flexibility to deploy at a SKU level closer to the point or time of actual order placement.  An example is to use the principle of postponement.  Forecasts can be made to achieve manufacture of base units of a product with customisation of the final product made once orders are received.
3. Independent demand items should be forecast.  Dependent demand items should be calculated.  For example, demand for motor cars should be forecast.  Demand for tyres should be calculated by multiplying the forecast for cars by five.
4. Forecasts should be managed according to value of products.  Products should be analysed by a Pareto distribution as shown below.  The A items that provide greatest value should be forecast more regularly than C items, if forecasting resources are constrained.  Further, if a statistical forecasting tool is used, the forecasts of C items may be accepted directly from the tool, but it is essential to have management overview of the statistical forecasts of A items (see below).

Demand Planning Process

A Demand Planning Process describes the activities that are required to develop one agreed organisational plan to drive sales and operations planning and financial planning.  The first stage of the process is for forecasts to be developed through statistical analysis and judgements.  The second stage is to have management and stakeholder overview of the forecasts to ensure ownership but also to override with strategic requirements, business policy and business knowledge.  In large organisations, demand planning often involves overview or constraining at a central level, unconstrained forecasts that have been prepared at a local level.  This may be to ensure fit with the overall marketing strategy of the business when supply is constrained.

The forecasting process should be based on historical demand data, information from the market and should also account for impacts of strikes and promotions.  Ideally history of sales should be demand history and not actual sales history.  The reason is that sales history may reflect what the customer was “forced” to accept at the time due to shortcomings in customer service or business performance rather than what they had preferred at the time.

The management and stakeholder review is critical for obtaining ownership by sales management and is where ideally customer overview should also be obtained (collaborative forecasting).  Overall, demand planning should be seen as a process using expert people supported by forecasting tools where appropriate, not the other way around.

Forecasting

There is a range of critical questions to be resolved by management in developing a demand planning process:

• Which SKU’s are to be forecasted? How many are there?
•  How far into the future?  Minimum period must cover purchasing and manufacturing lead times?
• What is the time period for stating forecast quantity? Months, weeks, or weeks for next 13 weeks and by quarters thereafter?
• How frequently should the forecast be made?
• How frequently should the forecast be reviewed and revised?
• What would constitute an acceptable level of forecast error?

Statistical forecasting is based on a range of operations research techniques including Time Series and Bayesian analyses.  A time series is composed of four components.  These are:

• Secular trend which is the smooth or long-term growth or decline of a series;
• Seasonal variations which are periodic variations that recur regularly within a period of a year;
• Cyclical trends which are periodic in nature over a period of several years and cannot be counted on to be repeated with predictable regularity; and
• Irregular movements which are unpredictable movements in brief periods of time.

There are many forecasting software tools available for a great range of prices. As always, you largely get what you pay for.  In my opinion, forecasting software, when implanted with inventory planning technology, provides the best bang for the dollar among the many supply chain software applications on the market.  Implementation risks are generally quite low.

Below are some criteria that can be used to determine if a software technology will be useful and what features might be beneficial for a business.

1. Organisations that have a very large number of SKU’s will invariably find that a software tool will be of great assistance.  Before selecting a tool, develop a set of historical data for some fast-moving SKU’s and a set for some SKU’s with irregular demand.  Insist on potential vendors demonstrating capability with data that you understand prior to any purchase commitment.  Buying software technology is not equivalent to buying a car and at least being confident that the product will deliver basic functionality.
2. Determine if the tool can be integrated with an inventory planning system, ideally provided by the same vendor.  These systems provide a capability to translate forecast variance (tracked through the forecasting tool) into safety stock requirements according to desired customer service levels.  Remember, this is one of the two prime reasons for forecasting in the first place.
3. The more expensive tools provide a range of algorithms by which forecasts can be done and automatically select the best to use.  This is very important when managing large numbers of SKU’s.
4. Users require the ability to review historical data and filter out the impacts of strikes or promotions.  The more expensive systems generally allow the user to examine demand history graphically and move individual items of data with a mouse manipulation.  Users may also wish to weight the most recent history.  Another useful feature is the ability to nominate “history” for one product as the basis for a newly introduced product.
5. An important issue for a business is that the tool is not seen to be a “black box” without buy-in to outcomes from the business.  The more expensive tools can provide features to distribute forecasts to managers for overview and modification and provide tools to allow forecasts to be graphically manipulated by the use of a mouse.

Measuring the Quality of Forecasting

The introduction of a demand planning process should be accompanied with the right measures.  The way managers are measured can deliver undesired outcomes:

• A sales manager who is measured on ‘beating budget’ will tend to under forecast
• A distribution manager who is measured on ‘no stock-outs’ is driven to over forecast
• A production manager who is measured on lowest unit cost is driven to over forecast
• A procurement person who is measured on ‘no shortages of materials’ will tend to over forecast.

The key measure is forecast accuracy.

Conclusions

This article has emphasised that the two critical reasons why organisations must have a formal demand planning process are to work with one agreed set of forecast demand numbers and to track forecast accuracy to drive inventory planning.  There are a substantial number of software systems from which to choose and some suggestions for selection have been provided.

Forecasting and demand planning are essential steps towards perfect order fulfilment.

Reference

Lee, H.L. and S. Whang, “Demand Chain Excellence: A Tale of Two Retailers”, Supply Chain Management Review, March/April 2001.