Boosting sales prediction’s accuracy and reliability with AI
Could artificial intelligence (AI) help your business build resilience during a crisis? Find out more as we share key insights from our podcast episode with Hugues Foltz from Vooban.
Winter is fast approaching.
Once again, some countries are bracing for high gas prices due to ongoing geopolitical conflicts. To keep up with customer needs, manufacturing companies like SBI need to predict the supply and demand for wood and pellet stoves around the world as accurately as possible.
In this blog, we discuss how AI has changed the way SBI predicts its sales for the better. These key insights are based on our conversation with Vooban’s Executive VP, Hugues Foltz.
Listen to his interview on the AI and Digital Transformation Podcast here.
What is sales forecasting?
In general, forecasting is a way for people to “see” the future as we often see in the form of weather reports. This approach is also applied to business operations.
Sales forecasting is a process that allows companies to estimate their sales revenue for a specific time frame (i.e. month, quarter, year). It predicts how much companies may potentially sell in the future. Using this information, finance and operations teams will be able to prepare budget and purchasing plans with their suppliers. Moreover, these forecasts let them anticipate possible irregularities in their complex supply and demand value chains.
Without the right tools, operating in the dark can be dangerous
“If you don’t master your supply chain, you cannot optimize that much of what you are doing, nor optimize your operation [as a whole].”
Can miners survive a difficult operation without their headlamps? The same principle applies to businesses.
Before 2016, SBI’s sales team had a sales forecasting accuracy of roughly 50% — the same probability one expects when flipping a coin.
Low sales prediction accuracy brings more harm than good as the manufacturing and shipping of stoves risk getting delayed once they receive an unexpected request from an international client.
To fix this issue, SBI asked Vooban to develop a sales prediction tool that could outperform its predecessor’s accuracy level. During the first four months, Vooban trained the algorithm using historical data retrieved from SBI, as well as external data.
The sales prediction’s accuracy immediately shot up to 75%2 during the tool’s initial run. It later peaked at 85 to 87% after more than a year’s worth of training, as well as adjustments to the database and data set.
But how did it perform at the height of the COVID-19 pandemic?
In 2020, the algorithm logged an accuracy level of roughly 78% — a commendable outcome considering SBI had never encountered a crisis of this type and magnitude before.
Things to consider:
Sales prediction accuracy can be measured by comparing the sales estimation made by the sales team in contrast to the actual sales made for that same year.
For context, a reliable prediction would have an accuracy level between 80 to 90%.
Building a robust and efficient sales prediction algorithm for a volatile market
“One of the reasons why the use case is interesting is because it’s about stoves, “One of the reasons why the use case is interesting is because it’s about stoves, which implies that wood is a part of the logistics chain. As we all know, the Ukrainian war disrupted the supply of wood and prices skyrocketed. So, I’m curious to know how you dealt with this kind of black swan?” ”
Supply chains are very tricky to deal with as they vary significantly based on the businesses and processes involved. Its complexity closely resembles board games like “Ticket to Ride.” With one misstep, the rest of the supply chain could collapse instantaneously, compromising businesses’ operations and reliability.
To address this volatility, SBI’s sales prediction algorithm needed to be antifragile. It must be able to detect the data drift and adapt automatically.
For Vooban to achieve this outcome, their AI team needed to know fundamental details such as: (1) the size of the company, (2) number of third-party suppliers involved, (3) regulations concerning the pipeline or product and (4) expected demand, to name a few.
Using this information, they were able to designate the right weight for each variable, thus ensuring the accuracy of the re-trained algorithm during an unexpected event like a pandemic.
Having proven its merit as a tried and tested solution, SBI continues to use Vooban’s sales prediction tool at present.
