Related Expertise: Finance Function Excellence, Digital Transformation, Data and Analytics
This is the first in a series of articles by Boston Consulting Group and Daimler Mobility discussing the concept of forward-looking financial steering. Here, we introduce the concept and explain how companies can use it. Subsequent articles will address implementation challenges related to people and technology. The insights are derived from Daimler Mobility’s successful deployment, with BCG’s support, of forward-looking steering in its global operations.
People don’t steer their cars solely on the basis of what they see in the rearview mirror, yet that is essentially how most business leaders steer their companies: they look backward to decide how to move forward. This method makes it hard for companies to cope with the ever-increasing levels of uncertainty in today’s business environment. To keep up to speed, companies need an approach to financial steering that permits rapid and effective course corrections in anticipation of future developments. Companies should spend far less time developing detailed plans and far more time taking action to counter threats and capture opportunities.
To make that happen, the paradigm for steering must fully shift its focus from backward looking to forward looking. Backward-looking steering entails analyzing deviations between plan targets and actual performance. Forward-looking steering entails comparing targets with forecasts of how KPIs will evolve over specific time horizons. To truly adopt forward-looking steering (as described in this article), a company must use algorithmically derived forecasts.
Although it is common for companies to produce forecasts manually, few companies use algorithms. Algorithmically derived forecasts allow the focus to shift from periodically reporting results to accurately forecasting the development of KPIs—faster and with less effort. Armed with foresight into how conditions will change, companies can take action to preempt unfavorable outcomes and promote competitive advantage.
Adopting algorithm-based, forward-looking steering is not easy, however. A company must enrich its traditional manual processes with a data-driven, automated approach to generating forecasts and performance reports. Among the many challenges are assembling a team that has statistical capabilities, setting up a new technical infrastructure, and building people’s trust in technology.
“To master the digital transformation, a company must take a comprehensive approach to algorithm-based forward-looking steering,” says Stephan Unger, Daimler Mobility’s Chief Financial Officer (CFO). “This includes not only advanced analytical methods, new technologies, and the right expertise, but also an engaging approach to change management.”
BCG’s CFO Excellence Index, a benchmarking survey of more than 200 companies, found that “excellence in forecasting” is a key differentiator of top-performing finance functions. The survey also found that forecasting capabilities are an important factor in promoting both efficiency in the finance function and satisfaction with the function across the broader organization.
Yet despite these recognized benefits, most finance functions have not adopted a state-of-the-art forecasting approach to their steering processes. At the same time, the processes that they use in preparing their annual budget and in reviewing their midterm plan tend to be long and cumbersome. (See “The Steering Process at BackwardCo.”) For half a year, organizations devote significant resources to collecting numbers. In many instances, the derived targets are already outdated by the time a company completes the process.
To understand the shortcomings of the traditional paradigm for steering, consider a company we call BackwardCo. It spends nearly a year creating a very detailed plan that documents its stakeholders’ agreement on the expected results for the upcoming fiscal year or years. Then it uses this plan during the ensuing year as a reference to assess performance. Here is how this approach plays out for 2020:
If quarterly performance does not meet capital market expectations—on either the upside or the downside—capital market investors will be unpleasantly surprised. Because the announced targets establish investors’ expectations, BackwardCo will need to upgrade or downgrade its quarterly guidance to investors.
Subsequent performance reviews look backward, making proportional comparisons of static financial KPIs for the month, quarter, half-year, or year with the midterm plan. A plan that is developed through this process reflects individuals’ personal biases and becomes obsolete as soon as the ink is dry.
Forward-looking steering entails analyzing algorithmically derived forecasts and defining and deciding on course corrections. (See “The Steering Process at ForwardCo.”) It provides early guidance on the likely development of KPIs under different scenarios and on the corresponding impact on future results. Decision makers can use this foresight to assess the attractiveness of alternative pathways that the company might take. This enables them to make decisions to exploit opportunities and to avoid adverse developments much faster than their competitors can. The model allows leaders to address several types of steering activities. (See Exhibit 1.)
Each month, a company we call ForwardCo produces an algorithmically derived forecast for the next 18 months. Here is how the company uses those monthly forecasts to derive a midterm plan for 2020 and to conduct forward-looking steering throughout the year:
The initial inputs to the forward-looking model are the company’s historical actuals, which accurately reflect the level, trend, and seasonal pattern of each KPI. Users can enrich these inputs with market data, macroeconomic indicators, or any other metric series that correlates to the KPI being forecast. Uncertainty decreases over time as more monthly actuals arise and go into the model. If subsequent forecasts point in a similar direction, the organization gains confidence in the projected quarterly and year-end results.
The company relies on a series of drill-down analyses to understand the key drivers responsible for forecast fluctuations and to identify their root causes. An organization can use local business know-how to confirm the validity of the drivers. By automating and digitizing the process, companies can rapidly generate information about the future development of the most important KPIs.
The forward-looking model enriches the backward-looking model in the following ways:
The company uses the insights to determine the most attractive options available to it. In case of adverse developments, this means deriving and taking corrective action to preempt the realization of an unfavorable forecast.
It is important to note, however, that the quality of the insights generated depends on the quality of the source data. Moreover, an algorithm forecasts KPIs at specific levels of probability—not with absolute certainty. For these reasons, the business decision on the right path forward should remain with the leadership.
In their efforts to modernize steering processes, companies all too often fall into the trap of buying or developing a new digital tool without suitable support. They then deploy it in the organization without changing the philosophy of the steering model, building the necessary organizational capabilities, or taking people along on the journey. Relying solely on tools usually results in failure. To avoid this trap, Daimler Mobility has taken a comprehensive approach to forward-looking steering that includes five elements: steering philosophy, skills and collaboration, algorithms and data, visualizations and cloud services, and steering process and transformation. (See Exhibit 2.)
Steering Philosophy. Daimler Mobility has laid out the central philosophy of its enhanced steering model in a set of guiding principles, which it communicates to the broader organization. These guiding principles make several critical commitments:
Skills and Collaboration. A central pillar of transforming the steering approach at Daimler Mobility has involved the building of an analytics team with capabilities to develop, run, and maintain the algorithms. Because most data scientists do not have a financial background, the company needed a way to transfer financial knowledge quickly and efficiently. One innovative approach it used was to conduct a series of weekly question-and-answer sessions. In each session, which lasted two to three hours, an expert elaborated on a specific business topic, such as a business model and its respective data flows.
Daimler Mobility has strongly encouraged collaboration between teams of business controllers, IT experts, and data scientists. Controllers help data scientists gain hands-on experience in daily financial operations, and they establish channels of communication that enable the scientists to guide the business. Data scientists help controllers understand how algorithmic forecasts support their daily work, encouraging them to view the forecasts as effective tools rather than as a threat to their influence in the organization.
Algorithms and Data. Many algorithms for forecasting KPIs are available, but choosing which algorithms to use can be a daunting task because the information available to feed into them evolves over time. Daimler Mobility’s calculation engine (named “Merlin,” after the legendary wizard) incorporates a multitude of algorithms and decides on its own which to use—whether one algorithm or a combination of algorithms. The engine applies machine learning to generate a fully data-driven, unbiased forecast that presents the best possible predictions. Whenever new monthly data is available, Merlin produces millions of forecasts for selected KPIs and automatically validates the forecasts by comparing them with past performance. On the basis of this comparison, it selects the best algorithm or combination of algorithms for a particular KPI. To ensure continuous calibration, Merlin repeats this process each month and whenever the company introduces a new KPI.
As a rule of thumb, data preparation comprises up to 80% of the development time of an algorithm. The source data often includes outliers—data points in a time series that have extremely high or low values. Outliers are especially problematic because they can significantly distort forecast results. Daimler Mobility has developed a model that detects outliers and automatically makes appropriate value adjustments to avoid distortions.
Visualizations and Cloud Services. Daimler Mobility employs a number of tools to effectively communicate and apply insights. For example, it uses dashboards to present forecasts, deviations over time, and root-cause analyses to specific users at the corporate, regional, or country level. (See Exhibit 3.) The country-level forecasts are visually displayed in an application called MyForecast. The application also gives controllers basic analytical support, such as the ability to compare forecasts for different time periods or under different exchange rate regimes. To promote organizational acceptance of statistical forecasts, each function allows controllers at the country level to adjust the forecast to reflect their views on how KPIs will evolve. The application also displays these adjustments.
The business leadership uses simulated scenarios to discuss possible options and decide on any necessary action. A storytelling feature in the visualization application allows presenters to facilitate their discussions by bookmarking specific information on dashboards; the presenters can then tell management the most important aspects of their story by navigating through the logical sequence of these bookmarks. Presenters can easily deviate from the story and delve into details when discussions demand a deeper look.
Daimler Mobility runs forecast algorithms and subsequent information processing for dashboards on cloud services that can execute many calculations simultaneously.
Steering Process and Transformation. The enhanced steering process that Daimler Mobility implemented is faster than the traditional steering approach, and it involves more stakeholders. On the fifth working day of each month, the organization receives the actual KPI data for the previous month. On the same day, it generates more than 5,000 forecasts. Two days later, it derives implications and makes decisions. This process includes accounting for one-time-only events that the actual financials do not reflect and that the engine cannot detect or learn. In the final step of the process, the executive team meets in a videoconference to discuss the factors that determine future results, to consider what-if scenarios, and to identify the most promising paths forward. The executive team then makes a joint decision regarding the way forward.
“To successfully transform toward the enhanced steering approach, employees must understand its benefits,” says Daimler Mobility’s CFO Stephan Unger. “Predictions from rolling forecasts allow our experts to focus on value-added tasks and provide a quick, unbiased basis for decision-making by the executive team.”
Transitioning the organization to this enhanced steering approach and mindset has been no easy feat. One success factor has been frequent communication to the relevant stakeholders about how the enriched process differs from the traditional process. This effort includes highlighting new or changed activities and their sequence, as well as noting new or changed roles and responsibilities. Another success factor has involved convincing all stakeholders of the viability of an algorithmic approach and the accuracy of the forecasts. To achieve this result, the company has retroactively compared forecasts to actual developments.
Daimler Mobility’s experience highlights several challenges that a company must meet in order to transition to a forward-looking steering model:
In the next articles in this series, as well as in a video presentation, we will explore in greater depth how Daimler Mobility overcame these challenges, and we will discuss the lessons that other companies can learn from its experience.
Daimler Mobility’s algorithmically generated forecasts predict performance for the next 18 months for more than 50 business entities, each with approximately 100 KPIs. In 70% of the predictions, the statistical forecast has proved to be the same as or more accurate than experts’ judgment—and it achieved these results faster and with far less effort than the experts did. This success comes from a forecasting engine managed by data scientists, as well as from a rigorously designed and executed transformation. The effort has yielded an important reward: foresight that enables the company to shape the future by preempting negative outcomes and exploiting new opportunities.