Instead of providing a single predicted value, probabilistic forecasting presents you with a range of possible outcomes. Find out how this grants you the insights to make better decisions as an Amazon Seller.

What is Probabilistic Forecasting?

Tackling Forecasting first, simply put, it is predicting an event which has not yet occurred. Of course, this would ideally be done using all the information at your disposal—including what you have observed so far, plus any additional factors like knowledge of future events which you think might have an effect on what you’re attempting to forecast. All this knowledge is typically consolidated to arrive at a single outcome. This is what a majority of Forecasting solutions do, and this single outcome is known as a Deterministic or a Point Forecast.

Of course, the reality is: a lot of different eventualities could occur, and the occurrence of any one of these would have a ripple effect on the forecast. Probabilistic Forecasting is a method which acknowledges the uncertainty of forecasting, and accounts for the possibility of these different outcomes. This method generates a forecast which assigns probabilities to each possible outcome, and gives you—in addition to the Point Forecast mentioned above—a range of all the other possible outcomes, known as the Prediction Interval.

What Exactly is a Prediction Interval?

Let’s take a simple example of your weekend plans. If someone asked you right now what you were going to be getting up to this Saturday, there’s a very good chance you’d know exactly what you’d be doing, or at least be able to narrow it down to a relatively small subset of possible activities. How? You would most likely use all the information about what you’d like to do, what the weather would be, conversations and plans you’d made with friends, etc.

If someone asked you the same question but the time period was instead this weekend a year from now, it stands to reason you would feel less confident giving an answer. Who knows what could happen in the next 365 days? Maybe you would’ve even taken up a new hobby like hang gliding by that point, which is perhaps not an option now, but would be in your “subset of possible activities” by then. Forecasting should be approached the exact same way, and Probabilistic Forecasting is how we achieve this.

A Prediction Interval (PI) is a range of values within which you expect that future value to fall, with a Confidence Level (CL) in percentage with which you expect that to occur. In our real-world “weekend plans” example, you could give a response like, “I’m 95% confident I will be going to the movies or grabbing lunch with my friends on Saturday. We haven’t decided exactly which one yet, but these are the options we’ve discussed so far.”

Contextualizing Uncertainty and Risk as an Amazon Seller

As an Amazon Seller, in a Demand Forecasting and Inventory Management scenario, what you will be trying to forecast is how many units of a given ASIN/Product you will sell on a given marketplace at set points in the future: next week, the week after, etc, for a Forecast Horizon you specify (perhaps looking as far as 52 weeks ahead for financial reasons). You would be managing this across the tens, hundreds or thousands of listed ASINs in your inventory.

With a Point Forecast, you would only get a single forecasted volume for each week with no context whatsoever around the forecasting model’s confidence, how much uncertainty there is, or how far off the reality could be once actualized. So all you would know is, for example, you should expect to sell about 240 units of ASIN X next week. How far off can you expect to be? No idea. What’s your worst-case/best-case scenario? No idea. You would be completely oblivious about these additional pieces of data which should be informing your FBA Inventory decisions.

With Probabilistic Forecasting on the other hand, you would be able to view a more insightful forecast for all your ASINs with the following data:

• A 95% Confidence Level: This makes you aware there is some level of uncertainty, but assures you that of all the possible outcomes generated by the forecasting model, the system is 95% confident the actual volume will be in the given range of predicted volumes.

• The Forecast’s Lower Bound: The lowest number of units the model states you should expect to sell, or your worst-case scenario with a 95% level of confidence. e.g. the least you can expect to sell is 230 units.

• The Forecast’s Upper Bound: The highest number of units the model states you should expect to sell, or your best-case scenario with a 95% level of confidence. e.g. the most you can expect to sell is 250 units.

• The Forecast Volume: The most likely volume the model states you can expect to sell, taking the average of all the possible outcomes simulated.

How Ronin Uses Probabilistic Forecasts to Improve your Amazon Inventory Management Process

Even though a configured Safety (or Buffer) Stock setting would be in place to minimize your chances of a stockout in your Amazon inventory, getting your Demand Forecast right in the first place should be your main priority. Having a robust and insightful forecast is a proactive approach to FBA Inventory Management and your Safety Stock is a preventative backstop for FBA stockouts.

Ronin’s proprietary forecasting engine, FarSight, generates these Probabilistic Forecasts which help you manage your Amazon stock better by:

• Capturing Uncertainty: Acknowledging there is always some level of uncertainty in forecasting, Ronin’s solution presenting the range of possible values allows you to better understand how likely (or unlikely) the point prediction is. Additionally, just as in our “weekend plans” example further up, the further out the forecast, the wider the Prediction Interval, and this is valuable data you just cannot extract from a Point Forecast.

• Improving Decision-making: Being able to see your best and worst-case scenarios make you better prepared for every eventuality. This will help to reduce stockouts (losing sales due to underestimating demand) and minimize excess inventory (lowering your FBA storage and holding costs), which leads to the next point.

• Facilitating Better Communication of Risk: Understanding the range of possible outcomes allows you to paint a more vivid picture to colleagues, stakeholders or even your suppliers—adding a greater level of detail to any planning conversations.

• Enhancing Pricing or Promotion Strategies: When you have a clearer outlook of the future, you will easily and more confidently decide whether or not you should be offering discounts, creating bundles or special promotions during upcoming periods of low volume, or reducing redundant spend to maximize your margins during high-volume periods.