Can Machine Learning (ML) Help Create More Accurate Plans and Forecasts?

Modern planning solutions now offer (or claim to offer) Machine Learning capabilities. But what is the practical benefit to your planning and forecasting process?  Is Machine Learning real and beneficial, or just more marketing fluff?

We’re hearing a lot about Machine Learning (ML) and Artificial Intelligence (AI) these days.  First, let’s quickly differentiate ML from AI and remove some of the mystique.  The purpose of AI is to simulate natural intelligence to solve a complex problem and its goal is generally to find an optimal solution.  ML is a subset of AI.  It integrates input and output in a feedback loop that it uses to automatically learn and improve predictions without explicit programming.  Put another way, ML’s goal is to increase accuracy by finding patterns in data to then come up with a prediction.

The application of ML to financial planning has the most potential in helping to create more accurate forecasts by quickly combing through large data sets to find the patterns that reveal hidden drivers.  These drivers can then be incorporated into the planning process to create more accurate projections.

Practical examples of how ML can be applied are found everywhere. In an article on the FP&A Trends web site, one of the authors describes a company’s project to predict how many people, for how many hours would be required to operate a gas station.  Locations, prices, day of the week and other factors were used to find the underlying patterns and predict future personnel needs based on assumptions.

The holy grail of ML is to be able to incorporate operational data and relevant outside data to hone predictions and create more accurate forecasts while fostering a deeper understanding of the levers that impact your organization’s most important performance metrics.

But are we there yet? ML doing this for you without an in-house team of computer scientists is not quite a reality, but it’s coming.  And basic ML capabilities are indeed being baked into planning systems with a point-and-click user experience in mind.

Adaptive Planning’s ML functionality uses at least two years of historical data (three to four years would be optimal) to automatically create a predictive forecast version which can then be compared to your human-input forecast to highlight anomalies.  This can be especially useful for alerting you to areas in the human-input forecast that might require further scrutiny.  The upper and lower bounds used to flag an entered forecast value as an anomaly to the ML prediction can be customized (the default is 45%).

The medium-term roadmap for Adaptive Planning’s use of ML is more ambitious.  They plan to offer Intelligent Demand Planning, where you can add additional levers such as weather, health-related data and more to help predict demand for products, raw materials, and sales and service personnel.  Sales Territory Optimization is in development as well, which would use various outside factors to determine where to focus sales reps for maximum efficiency and revenue impact.

Longer-term, Adaptive Planning will further leverage ML to provide Target Guidance using Monte Carlo Analysis, and Tendency Analysis which uses various statistical techniques to reveal long-term biases in your plans (i.e., where your actuals are usually off from your plans).

These ML capabilities help with two main things: creating more accurate forecasts and saving you time. This frees up time to allow your organization to focus on the value-add tasks of better aligning forecasts to organizational goals, evaluating performance, and taking corrective action when it can make a difference rather than just reviewing last year’s misses.

So, what’s the bottom line?  Although it’s still relatively new, ML currently has its use in anomaly detection, while the future is looking very promising indeed.  The leading modern planning solutions can bring some of the benefits of ML to your planning process now, and you’ll be ready when the upgraded capabilities become available.  However, it’s important to note that the effectiveness of ML depends on having more, more relevant, and more accurate data to work with.

Want to learn more about how Adaptive Planning’s ML capabilities work?  Here are a couple of resources:

How could Adaptive Planning help streamline your planning, reporting and analytics?  We’d love to brainstorm with you, so feel free to reach out to us!