Forecasting at the forefront: Can I trust my forecast?

Mitch Prevett, SVP Head of Quantium Americas and Forecasting Specialist, Glen Maisey discuss the game changing results when CPGs move away from manual forecasting and adopt an AI based forecasting approach.

By Mitch Prevett and Glen Maisey

Today, forecasting in CPGs is the anti-thesis of trust. There are too many considerations for a person to process, and yet AI based approaches are ignored in favour of manual.

The machines just don’t think enough to be trusted. Control over the forecast, even with the pure intent to improve accuracy, is a necessity. Whilst most demand planners will agree the true intent is to improve accuracy, there is often the seemingly justified need to bias the forecast to help meet KPIs. After all, when your KPI is linked to sales you need to maximise your chance of meeting targets, even at the expense of the wider result.

Generally, promotion forecasts are manually created by the sales team using the trusty, “make a small adjustment on what we sold last time” method. Sales teams do not have the time to work through seasonality, moving events, promotional fatigue, forward-buy and cross customer cannibalisation. Even if they did, isolating each and how it is likely to impact the future is guesswork. Given that a single forecast can’t be trusted, any scenario analysis or trade spend optimisation is out the window.

Meanwhile, demand planners look after baseline forecasts in a separate process and take on the challenge of dealing with the complexity imposed from a broken system. For example, promotions are forecast at the level of a group of products together, which are then allocated to each individual product – but promotions don’t impact them all in the same way.

The sell-out forecast is then translated into a supplier order (ex-factory) forecast using manually created phasing allocations. These can vary significantly depending on the customer, promotion depth, supporting media, seasonality and event timing. Once again, thinking through the many combinations of effects for each customer is a difficult task to approach manually.

To keep this all moving, the S&OP process has been created to ensure demand planning and sales are on the same page. Most teams won’t fully align and settle for making sure the largest differences are accounted for. Often meeting rituals are onerous and a significant amount of time is invested across sales and supply chain functions in processes that are needlessly complex and prone to error. The common mistake is to target an optimistic sell-out forecast on the sales side and not order appropriately.

If algorithmic forecasts could be trusted, what would change? How many meetings, processes and errors could we avoid? How comprehensive, streamlined and up to date could they be? What would we focus on as a business instead? What is the true impact of having unbiased forecasts on KPIs like sales, waste, inventory and cost to serve?

In two recent “Proof Of Trust” exercises we conducted with CPGs, forecast error was improved by 1300bps for a wholesaler and 700bps in a major grocer in a very seasonal / high-low environment. Most importantly, these benefits are pure AI driven without any manual touchpoints.

Forecasting at its core is about learning how to interpret past combinations of drivers and projecting these outcomes onto future combinations, both seen and unseen. To put it another way, we can’t be bound by the training set we happen to have observed. What is the impact of a 30% discount on an ice-cream going out of season three weeks after a 50% discount in peak season and one week after we put on a 50% promotion in the other major supermarket, the combination of which we have never seen before? I don’t know. Our algorithm does.

We can now trust algorithms.

The results are game changing leading to decreases in inventory in combination with lower waste and higher sales. The trifecta. Upstream, late changes to production planning minimises and uses less costly overtime resource. Improved forecasts mean a smoother job, rather than jumping at shadows to address gaps in forecasts that aren’t in strategies or sales plans.

CPGs will be undertaking the journey to realise the benefit of trusted AI based forecasting over the next five years. Innovative businesses are pushing for the change now, to better service customers and end consumers with smarter promotions, ample stock on shelf, and ultimately savings in their pockets.

We implore any CPG looking for software to streamline or improve their forecasting to ask an important question: Can I trust the forecast?

Sign up for articles to your inbox