In forecasting there is an old saying that forecasts are never accurate, and that what’s important isn’t an exact number, but a range. Advancing digitalization is making this approach more and more effective, as we use forecasts to identify trends, and use machine learning and artificial intelligence (AI) to react quickly to short term volatility. 

A forecast range has always been valuable as it allows us to know with a reasonable degree of certainty how much we need to produce, and within a tolerable margin of error. But it’s not without its limitations – it is after all limited in the sense that you are likely to over or under produce to some extent.

Far from machine learning and AI making statistical modelling obsolete, they are making it more valuable

But now, by focusing on the direction of trends and range of error, we can overlay digitization and analytics to quickly adjust to reality where needed. Collaboration and increased transparency has allowed organizations to act faster, reducing cycle time and moving decision making into periods where the forecast is more accurate. Far from machine learning and AI making statistical modelling obsolete, they are making it more valuable. We can now use time series forecasts as our base and use new data analytics to identify short term spikes in demand, and react accordingly.

Time Series Forecasting Isn’t Going Anywhere

The forecast is an integral part of planning. It is the compass that guides the supply, production and financial plans in the right direction. It is a critical input to the Sales and Operations plan. Forecasting is used for many purposes in supply chain: manufacturing capacity, number and type of machines, employees and skills, long lead-time components, inbound and outbound transport capacity, warehousing space, and material handling equipment as examples.

Even highly flexible businesses that claim to not have a forecast, and only produce to demand, must make some estimates for the future in order to produce economically, hire the right number of employees, and establish sufficient sources of supply.

What Has Changed?

What has changed you ask? Think about the pace of change in business today. Uber, Netflix, Airbnb, Amazon, to name a few, have disruptive business models. Successful companies encourage managed risk, embrace change and focus on speed of decision making. They are passionately aware of their customers’ needs. They respond rapidly to new opportunities, constantly evolving and redefining their markets. Their entire operation is geared to respond quickly to deliver new services and products at disruptive prices. AmazonFresh is a great example of this.

The value of forecasting has been elevated with digitization

Companies that believe that facts are the best guide to decisions have made data and analytics an integral part of their business, supporting their competitive strategy. For example, Netflix is using analytics to predict customer viewing preferences. UPS uses prescriptive analytics to tell drivers which route they should follow to make deliveries in the least amount of time with the least fuel.

Digitalization Improves Your Ability To Forecast… And Delight Your Customers

Large data sets are improving the ability to analyze trends and buying behaviours. Companies are using data analytics to help them sense and respond. With data such as point of sale data, channel data, weather patterns, and buying behaviour, early warning signals provide insight into what is really happening versus the plan or forecast. This allows companies to respond faster to deviations.

There is also an increased use and accuracy in predictive and prescriptive analytics to generate forecasts and identify out of tolerance forecast exceptions. And with machine learning and artificial intelligence the machine will create the statistical models.

 Increased collaboration has increased transparency, reducing time to respond to demand changes – you can thank millennials for that

The emphasis on collaboration in organizations, influenced by millennials, has increased transparency throughout functions in the organization. This has resulted in companies reducing their time to respond to demand changes. It is well understood that the accuracy of the forecast diminishes over time; therefore, a faster response can be based on a more accurate forecast.

How Do You Get There? Top Ten Tips

1.  Encourage innovation and disruptive thinking. Allow employees to test and learn in innovation labs.
2. Foster education on digitization; big data and analytics and how it can positively impact your forecasting.
3. Focus on group decision making when creating a consensus forecast. This requires cooperation versus competitiveness.
4. Monitor the value add of the collaborative process versus just the statistical and naïve forecasts.
5. Use range forecasting to allow for upper and lower alternative forecasts to also be reviewed and analyzed with multiple forecast scenarios.
6. Perform risk analysis to proactively understand and mitigate the risk on supply planning, production, revenue, and margin of the forecast scenarios.
7. Implement process improvements to reduce reaction time, cycle time and inventory. Taking time out of your process allows you to act on a better forecast and respond more effectively to changes.
8. Improve your response time and the accuracy of your response to the customer for demand changes by working in collaborative teams versus siloed decision making.
9. Direct your employees to be objective driven versus functionally driven.
10. Last but not least, become intimate with your customer. Understand their buying behaviour. Apply segmentation strategies to improve your service and delight your customer, every time.

[Ed: to set up machine learning in your organization, check out this guide to getting started with machine learning]

At Nespresso, we started the Collaborative Demand Planning (CDP) initiative two years ago, with the aim of forecasting Nespresso sales in a data-driven way. The goal of CDP is to reduce demand planning bias, minimize working capital and set up automated statistical forecasting. Switzerland, Brazil and France have been selected as pilot markets. These countries are very useful to study ways of working for the three Nespresso categories: Coffee, Machines and Accessories.

Coffee, the most important and mature category, has been selected for a full roll-out of statistical forecasting. As of February, the HQ Demand Planning team provides sales forecasts to 39 Nespresso markets. Nespresso HQ is acting as a hub, generating, analyzing and sharing forecasts for those Markets. The CDP initiative is summarized in the following image.

Scope of the Collaborative Demand Planning initiative

We use a SAS solution to generate both topline and SKU-level forecasts for each market. While we define specific ARIMA models at topline level, we let the tool automate forecasts at SKU level. In addition to sales history, we include causal variables such as key promotions, customer data, number of boutiques and B2B sales force information. This allows us to take into account domain knowledge from each market.

Questions We Asked When Establishing The Process

In the last two years, we learnt a lot about key questions related to business forecasting. Here are the important topics we had to review while implementing the project:

• What do we want to forecast (warehouse output vs. final sales to the consumer)?
• What is the forecast horizon?
• Should we forecast weekly or monthly?
• Which reconciliation strategy should we use (top-down vs. bottom-up)?
• What family of models is better in our case (ESM, ARIMA etc.)?
• Which causal variables should we track and include in our forecasting models?

How We Use Customer Data

Since Nespresso is mainly selling directly to the final consumer, we can use customer data to improve forecast accuracy. Examples of customer data includes member base and average consumption. To avoid multicollinearity issues, we perform a careful selection of variables based on a correlation analysis with sales. To evaluate the success of this initiative, we are tracking metrics such as sales forecast accuracy, adoption and DPA. Research activities on forecasting Machines and Accessories are performed in parallel.

Having completed the implementation and roll-out, we are now working with teams in individual markets to build trust and increase forecast adoption. This is where the expertise of the HQ Demand Planners is coming into play. Each month, the team is reviewing, analyzing and sharing forecasts with the markets. At this point, we see the limitations of what the tool can do. The forecasting tool is perfect to find models that minimize forecast error but several proposed models have been found to be unusable and unrealistic.

Filtering Out Unusable Forecast Models

An example of an unusable model is one proposing a flat forecast (i.e. each month the same value). Even though it may minimize the error, it won’t be trusted by stakeholders. We consider models as unrealistic when the proposed growth is unlikely to happen in reality. We found that reviewing both the shape of the forecast and the YTD vs. YTG are crucial to propose forecasts that are trusted and adopted by our stakeholders.

Alan Milliken

Many firms have no strategy for dealing with unforecastables and some even pretend that all items are statistically forecastable.  Statistical tests need to be used to identify those articles which cannot be forecasted with conventional techniques, such as intermittent demand and demand with high variances.

For example, low volume and erratic demand may indicate that an item should not be offered in the current product configuration or at a particular location.  A key question that must be answered in such cases is whether or not the product is profitable and to what extent? If the product is profitable, the corrective action may be to move the customer to a more popular package size or assign the customer to a different warehouse with more demand for the product.

So, I am not talking about Mickey Mouse and Donald Duck.  I am talking about the upcoming IBF conference in October where you can learn the “magic” of better demand planning.

You will have many opportunities to learn how to improve both the quantitative and qualitative aspects of the Demand Planning process.  I will be presenting on how the world’s leading chemical company, BASF, develops strategies to improve forecasting and better manage unforecastables.  This includes methodologies for aligning product offerings with forecasting strategies and determining when conventional statistical tools are the right answer.  Mine and other presentations will provide you with the knowledge and understanding to optimize the interaction of people-process-technology within the demand planning function.

If you are looking for a little “magic” to improve your demand planning & forecasting, join us in Orlando.  You will hear about the use of volume & variance analysis to develop demand planning strategies and much more.

Of course, your comments here would be greatly appreciated.  How do you handle unforecastables?  It would be great to hear from you.

Alan L. Milliken
BASF Corporation

See ALAN L. MILLIKEN Speak at The IBF’S:

$695 (USD) for 3 Full Days!

October 12-14, 2009
Orlando Florida USA