Analytics Section of INFORMS NewsStanford team wins Syngenta Crop Challenge
Syngenta and the Analytics Society of INFORMS named Xiaocheng Li, Huaiyang Zhong and associate professors David Lobell and Stefano Ermon – a team from Stanford University – as the winners of the inaugural Syngenta Crop Challenge in Analytics. The team was awarded a $5,000 prize for their entry, “Hierarchy modeling of soybean variety yield and decision making for future planting plan,” which modeled a system for predicting soybean seed variety selection.Read More
Analytics Section of INFORMS NewsInnovative Applications in Analytics Award
The Innovative Applications in Analytics Award (IAAA), which recognizes creative and unique developments, applications or combinations of analytical techniques used in practice, has garnered significant interest in its brief life. The IAAA for 2016 was awarded to an MIT-led team for its submission entitled, “An Analytics Approach to the Clock Drawing Test for Cognitive Impairment,” during the recent INFORMS Conference on Analytics & Operations Research in Orlando. Fla.Read More
Analytics Section of INFORMS NewsWelcome to the Analytics Society!
The Analytics Section has become INFORMS’ newest society, and we are now the Analytics Society of INFORMS. Achieving society status recognizes our significantly increased membership base and our vastly expanded scope of activities, all of which were on display at the 2016 INFORMS Conference on Business Analytics & Operations Research in Orlando, Fla. We are still the fastest growing subdivision of the Institute!Read More
Collaborative forecasting: from vision to reality
Collaborative forecasting enables companies to transition from periodic, disparate and isolated forecasting activities to a single, real-time enterprise forecasting process.
By Brian Lewis
Forecasting is the foundation of virtually all business planning. Scheduling production requires a demand forecast. Deciding between short- and long-term investments requires a cash flow forecast. You need a customer usage forecast in order to buy enough servers to support a growing cloud-based software application. The list of forecasts that a company must prepare is endless and, moreover, the forecasts are intertwined.
One of the biggest problems when forecasts are inextricably linked is the lack of communication between stakeholders. No one knows which assumptions have been incorporated, whether they are up to date or which data is relevant. The result is incomplete information and potentially obsolete forecasts. This uncertainty leads to confusion and mistrust in the forecast. Stakeholders start looking for alternative forecasts they can confidently base key decisions on; sometimes this means building their own models, spreadsheets or local data systems. Other times it means relying on their experience and gut-feel alone. While they may feel more confident, they still have no assurance that data and assumptions are correct and up to date.
Because forecasting is so pervasive and so important, it should be treated as the “plumbing application” that it truly is. Imagine if all forecasting activities were integrated into a single, living enterprise forecast. Data is automatically analyzed as soon as it becomes available. People with relevant insight about the business and knowledge about planning decisions input this information as soon as it changes. The impacts of changes to data and assumptions automatically flow through the system in real time, updating all relevant forecasts. You can see the impacts on all areas of the organization as changes happen. You can test and understand the implications of what-if scenarios and alternative decisions. You can respond and modify your plans in real time.
Collaborative forecasting applications are turning this vision into reality. These technologies are based on a few fundamental concepts: balancing analytics and insight, real-time updates of data and assumptions, and openness and transparency.
Balancing Analytics and Insight
Every forecasting process falls on a spectrum that is bounded by two extreme approaches – pure analytics on one side and pure insight on the other. Collaborative forecasting combines both approaches, incorporating the best aspects of each.
Analytics-based forecasting relies on forecasting “engines” in which statistical and other mathematical models are applied to data to automatically find trends, patterns, seasonalities, outliers, shifts, correlations, etc. Essentially you push data in and forecasts come out. These engines employ analytics that find and extract information from data that humans could not easily find themselves (if at all). Computing power is dirt cheap these days, and automating the analysis of massive amounts of data in a short period of time is extremely cost effective.
However, analytics-based forecasts are only as smart as their underlying models and algorithms and are highly dependent on the input data. Forecasting engines cannot find information – such as the effects of a promotion – unless that information is somehow represented in the data and the model knows to look for it. The data itself might be nonexistent (e.g., no historical sales for a new product) or irrelevant (e.g., historical sales do not match a customer’s new buying pattern). In any of these cases, analytics-based forecasting is not an option.
In contrast, insight-based forecasting relies exclusively on human judgement and gut-feel. Key stakeholders apply their experience, expertise, knowledge of business operations, expectations of customer actions, sense of market conditions, etc. to project future outcomes. This approach has benefits, such as the incorporation of timely first-hand knowledge, but it has flaws as well. People are not particularly adept at analyzing data for trends, patterns, etc. and therefore miss out on this contribution to the forecasts. The approach is not scalable or easily replicable, which is a problem if your company needs to forecast many items. People are also subject to many sources of conscious and subconscious biases, such as the optimism of a sales team, which can inappropriately skew forecasts.
Collaborative forecasting software can balance analytics and insight by applying analytics to find good baseline forecasts and then adjusting and overriding them based on insight. For each item being forecasted, collaborative forecasting applications must allow you to select the right mix and application of analytics and insight. As anyone who has worked with large enterprise software applications can attest, flexibility is a rarity. Fortunately, new collaborative forecasting applications are designed more as forecasting platforms than forecasting tools and offer users the high degree of flexibility that they require.
Real-Time Updates of Data and Assumptions
As a concept, collaborative forecasting is a “living” process in which forecasts are always up to date. As soon as someone changes an assumption or data is updated, the forecasts should incorporate the changes across all departments in real-time. If the forecasts do not reflect the latest information, then any decisions you base on those forecasts are immediately obsolete.
Collaborative forecasting therefore requires a central forecasting application that pulls in data from disparate systems spread throughout the organization. The application must provide mechanisms for key stakeholders and knowledgeable people to record their assumptions and link them directly to the forecasts. The application cannot simply document and store insights in some spreadsheet or database. Instead, assumptions must be modeled quantitatively and linked to a forecast. As the assumptions change, the forecasts change in response.
For example, suppose your marketing department plans to run a new promotion that they believe will lift demand by 10 percent next month. Rather than writing a document or sending an e-mail with these details, the marketing department should be able to input key assumptions in the collaborative forecasting application, such as the lift percentage, the launch date and duration, the type of adjustment (i.e., multiply the lift by the baseline demand), and which forecasts the promotion should adjust. If marketing decides to delay the promotion by one month, the demand forecast will automatically reflect this shift as soon as they update the launch date.
Even simple decisions such as delaying a promotion can have widespread operational and financial implications. The cost to implement the promotion shifts by one month, which affects financial forecasts. Did you need additional headcount to support the promotion? If so, this affects headcount forecast. The shifted promotional demand impacts your revenue forecast and potentially procurement, inventory and production forecasts. Remember, all forecasts are intertwined.
With real-time updates of data and assumptions, collaborative forecasting allows you to see the impacts of changes as they happen and be more responsive in planning.
Openness and Transparency
People trust processes that are transparent and inclusive. Rather than tracking down and reconciling information from Marketing, Sales, Operations, Finance or HR, let them take direct ownership of the information they know best and link to the information that they need. Collaborative forecasting allows each stakeholder to add new information, propose changes, test scenarios and share insights in a single, shared system. Their rationale and contributions are tracked, time stamped and documented for complete openness and transparency.
Documenting the rationale for assumptions is a critical form of communication between stakeholders. While there still might disagreement about the numbers themselves, stakeholders can have more productive discussions if they know the rationale behind them. Perhaps this leads to changes, perhaps not. But access to information and open discourse builds trust in the process. Without it you run the risk of introducing uncertainty, confusion and mistrust, and this is something you clearly want to avoid.
Another reason for documenting rationale is for performance management. Everyone knows why they are updating their assumptions today, but they may not remember the details in a few days, let alone a few weeks or months from now. By time-stamping your changes and documentation, you have an audit trail of every change that was made. When you look back at older forecasts, you can see the old data and assumptions, who was responsible, their rationale for making changes and so on. By knowing what was done and whether or not it was successful, you can design forecast accuracy and process improvements that incorporate the good and eliminate the bad.
Implementing Collaborative Forecasting
Three potential hurdles exist to implementing a collaborative forecasting process: technology, awareness and readiness. Collaborative forecasting software now exists in many forms. Some of the newest and most promising are comprehensive and support all of the fundamental concepts of collaborative forecasting. Others are more limited, for example focusing more on the analytics and the forecasting engine approach. With new technology comes new awareness of what is possible and the interest in collaborative forecasting is increasing rapidly. However, even with technology and awareness, some companies may feel they are not quite ready to implement a collaborative forecasting process. You know your process isn’t great and your forecast accuracy is only so-so, but process changes can be challenging at your company. That may be true, but the financial and competitive advantages of collaborative forecasting are substantial enough that you cannot afford to wait any longer.
Brian Lewis, Ph.D. (firstname.lastname@example.org), is vice president of Professional Services for Vanguard Software, an enterprise forecasting software company with more than 2,000 customer in 60 countries, including 33 of the Fortune 100. He is a senior INFORMS member.