Skip to main content

Demand forecasting has never been more critical or difficult, so how can businesses adapt in the face of uncertainty?

data woman
(Image credit: Future)

Much has been written about demand forecasting and planning; extensive processes have been designed and many implemented in some fashion, often aided by consultants, and backed by systems. Perhaps you have also had advice on ‘best practice’ and significant amounts of time are now given over to ‘forecasting’ as part of a monthly routine.  

That time of the month comes around again; a new ‘forecast’ must be generated, and you are sat in another meeting to agree it. Called the S&OP (Sales & Operations Planning), or, if you are up to date on terminology, the IBP (Integrated Business Planning) meeting; you must decide on the numbers or give your perspective on what they should be. There could be at least two scenarios that typically play out.  

It could be that a team has conducted statistical analysis and are presenting an unconstrained demand forecast that has been built on understanding trend, seasonality, cyclical patterns, and causal factors. The focus of discussions is on the inputs to the analysis; what data has been captured and used, what assumptions are relevant and are they valid. Once aligned on the assumptions that have shaped the analysis and you now look to agree on any actions needed to shape demand toward supply capacity and commercial objectives. Note, in this scenario, the demand forecast is not expected to match the financial budget, sales target, market share objective or any other ‘plan’. 

Or, alternatively, some statistical analysis has been done. This forms the statistical ‘baseline’ and, because it does not match a desired number, is generally either ignored or overridden by significant adjustments. Most, if not all, of the discussion about the statistical analysis, is on the output. The monthly ‘forecasting’ meeting is characterised by the pursuit of an aligned output; the ‘one number’ forecast all functions can agree on. Lengthy justifications for the different functional inputs; Sales seeking to either lower the plan (to then overachieve) or boost the plan to ensure supply, Marketing to hit objectives and justify spend, Finance to ensure adherence to budgeted revenue and expenditure. A general conflation of demand forecast and demand plan; the terms forecast, plan, budget are used interchangeably and essentially represent the number the business would like to achieve. A passing reference, at best, to the accuracy of forecasts; perhaps top line numbers by brand or area look OK and it’s an operational issue that stock outs occur or inventories build up a compromise ‘forecast’ being confirmed, largely based on judgement. 

Where would you position your business on a spectrum defined by these scenarios? 

Demand forecasting is not easy; forecasts are bound to be wrong. It can be considered a thankless task but, done well, demand forecasting can drive revenue growth whilst reducing costs and capital. If your business is closer to the second scenario, and most are, then there are some important steps to take in making forecasting a strength rather than a problem. And it is more about people and culture than it is about statistical wizardry. 

Demand forecasts, demand plans; what’s the difference? 

This may sound basic, and indeed pedantic, but terminology is important. For clarity in this discussion, we will work with the following definitions: forecast - an estimate of future demand based on analysis of data and derived by application of statistical formula or algorithms. Plan – a structured set of activities expected to create a defined demand. The component elements of the plan may be estimated by statistical analysis and/or the application of domain knowledge. 

This distinction is critical; to drive improved accuracy and derive maximum benefit from a forecasting process, businesses need to be comfortable with, and value, estimates of demand that do not match plans, budgets or targets. The fundamental concept is that by creating an unconstrained forecast of demand, the business can apply a data-driven approach, using the ever-increasing amount of data available to gain dynamic insights, remove, or greatly reduce, bias in the generation of forecasts and plans and respond pro-actively to emerging demand, with coherent cross-functional actions. 

In short, if the demand forecast is not what you want to play out, then what are you going to do about it? Change the forecast to a number you are happy with or trigger actions that seek to shape demand toward your desired outcome? 

There are many proponents of ‘one number’ forecasting and S&OP processes are very often geared to producing consensus on a number that all functions are aligned on. Therein lies the problem; how to achieve compromise with conflicting objectives and strong incentives that inherently bias inputs? Mindset is at the core of the issue. If the demand forecasting and planning process is a means to confirming a desired output, be that a functional plan or target, a reconciliation with a budget, then inconvenient data and analytics risk being ignored, and errors perpetuated. 

‘The forecast is wrong’   

Forecast error is a fact of life. There are some factors that compound it: 

Bias – driving demand forecasts to hit a desired number 

Consensus – achieving compromise by fudging a number  

Disaggregation – working at a higher level in a product, regional or organisational hierarchy and then cascading this estimate to lower levels, wiping out any significant granular variations 

False patterns and relationships – mistakenly identifying links and factors, allowing judgements of what could relate but does not. 

We need to qualify the performance expected of a demand forecasting and planning process. There are two key dimensions: supply chain tolerance; what is the degree of error that the activities across the supply chain can manage? We may wish for minimal error, but we need to be clear on when variance to forecast and plan starts to cause real disruption, drive up costs and draw in working capital. Forecastability- what are the demand characteristics of the products we are seeking to forecast? Are they high volume with stable sales patterns? Are they apparently erratic with spikes that are difficult to explain? 

A good starting point to understand forecastability is to determine how variable demand is for a product and compare this to sales volume. This will allow products to be segmented based on their demand patterns.  

 A pragmatic segmentation of products is likely to include:  

  •  High volume/Low variability:   
  •  Strong patterns and clear causal factors should make for accurate forecasting  
  •  Can form the baseload for efficient supply chain operations  

High volume/High variability: 

Likely to be event driven, so understanding event triggers and capturing data and information becomes the focus of forecasting efforts 

Supply chain design will need to support flexibility 

Low volume/High variability: 

Forecasting effort will need to be gauged based on the profitability of items in this category. Again, understanding what prompts demand will help identify if further data capture will support forecasting 

Likely to be managed by either buffer inventory or designing the supply chain to respond in a timely manner 

Low volume/Low variability 

With steady demand patterns, an automated approach should release resources for focus elsewhere 

Managing production and inventory of these items will need to balance quantity and frequency of replenishment. These items could be part of a ‘rhythm wheel’ approach where they play a fill in role to support asset utilisation. 

Understanding the relative weight of the demand segments in your business should feed into thinking about supply chain design. Higher predictability supports lean operations, high variability requires responsiveness and the ability to adapt supply chain flows to actual demand. Postponement, the delaying of final product assembly or configuration until requirements are crystallising, is a well-documented supply chain design that aims to service volatile demand effectively. 

With a picture of demand characteristics and an understanding of how forecast error impacts the efficiency and effectiveness of supply chain operations, suitably tailored performance measures can be set for forecasting. The focus for a programme to improve forecast accuracy can also be defined by understanding variability and assessing the data that needs to be captured and analysed to inform the core forecasting equation, namely: trend, seasonality, cyclical patterns, causal factors, randomness and unexplained variance. 

Trend, seasonality, and cyclical patterns are the mainstay of standard statistical approaches to forecasting and rely on robust sales history. Applying regression techniques to determine causal relationships with factors such as price, marketing, sales promotions, and competitor activities is an area of significant opportunity as more data is captured and shared. The technology and skillsets to combine these techniques requires investment and should be gauged on the specific circumstance and needs of the business.  

Changing mindsets 

We are increasingly awash with data. As artificial intelligence (AI) is applied to process and sift vast datasets in more sophisticated ways, businesses can start to access the analytical tools needed to take demand forecasting capabilities to new levels. Many businesses, however, show a marked tendency to believe that investing in technology alone will solve a problem or radically improve performance. Supply chain improvement initiatives, in particular, are often predicated on new systems. In the scenarios we visited earlier, adding new analytical software to a situation where demand forecasts are vetted against personal interest will simply negate return on investment or even create a negative outcome.  

Whilst technology can play an important role, the approach and mindset of the people that prepare and confirm the demand forecasts and plans for the business is critical to successful, sustaining improvement in forecast accuracy. The two scenarios painted at the beginning are fundamentally about mindset. To take a data-driven approach to demand forecasting, businesses need to: 

  • understand the statistical techniques increasingly available, both the opportunities and the limitations, with senior leaders critical to the positioning of analytics in planning processes  
  • engage functional teams with a focus on how they contribute to and shape the inputs to forecasting  
  • measure inputs to the process on the basis of the how they add to forecast accuracy  
  • clearly define forecast performance expectations that reflect product and market dynamics as well as supply chain operating capabilities  
  • work collaboratively to share insights and information based on domain knowledge 

In the first scenario described earlier, the essential premise is that the business invests in analysis and everyone contributes to making this analysis as comprehensive and rigorous as possible. If the output is a demand forecast that creates gaps to plans, budgets and capacities, then planning collaboratively, the business looks to close these gaps with a set of aligned actions. Faith is placed in analytics, judgements are subjected to scrutiny and ongoing learning is emphasised. 

Changing behaviours takes time. Incentives and key performance indicators should be developed that foster new ways of working and reduce the potential for bias and conflicting objectives. The potential offered by a data driven approach to forecasting is significant; speeding up the translation of actual demand signals into coherent demand plans (and hence supply) with unconstrained forecasts can change the whole dynamic of the process. Demand forecasting in this context is no longer a monthly chore, but a core competence that can bring competitive advantage. 

Now more than ever, in a digital age, there is an opportunity to transform demand forecasting with abundant data and technology that brings analytics within the grasp of many businesses. Combining these technical advances with a culture and way of working that is data-driven, businesses can reap the benefits that improved forecast accuracy brings, when even marginal gains have a big impact and can drive: revenue growth through better service, with the ability to both anticipate and shape demand, cost reduction across the supply chain by translating demand signals at speed, inventory and working capital efficiency as safety stocks are targeted and asset utilisation with expediting mitigated. 

Demand forecasting has never been more critical or difficult, so how can businesses adapt in the face of uncertainty? 

Calum Lewis, Founder and Principle Consultant, OP2MA (opens in new tab)

Calum Lewis is the founder of OP2MA, an innovative consultancy that focuses on transforming supply chains for sustainable growth. Calum has extensive experience in leading businesses and delivering exceptional operational and financial performance. With the LEGO Group, he embedded best practice supply chain management to drive five-fold sales growth to £300m.