How AI is making the future clearer for business forecasting

If you could choose one superpower to take your business to the next level, what would it be?

Perhaps it would be hiring the perfect person for every role you have on the first try. Or you may want to have the ability to never feel tired.

Or maybe you’d want the ability to see into the future.

After all, being one step ahead has always been critical for any business that wants to lead its respective industry. But while literally being able to see the future is entirely fictional, we’ve got the next best thing: business forecasting. And with current advancements in AI, these predictions have become even more powerful.

Seeing what humans can’t

With AI added to the mix, business forecasts that rely on machine learning (ML) models have a huge advantage over more traditional methods. These include moving average analysis – a statistics-based method used to predict long-term trends – and exponential smoothing, which is similar to moving average analysis but assigns different weightages to each data point depending on when it was recorded.

In supply chain management, for example, ML-based forecasting can reduce errors by 20% to 50% as compared to traditional spreadsheet-based analytics methods, according to McKinsey.

Photo credit: imagesource / 123RF

Subsequently, this can lessen lost sales and product unavailability by up to 65%, lower warehousing costs by 5% to 10%, and slash administrative expenses by 25% to 40%.

“Traditional forecasting methods are based on analyzing data with characteristics that have finite, countable, and explainable predictors,” says Chandrika Kadirvel Mani, customer engineer and AI/ML technology specialist for Southeast Asia at Google Cloud.

ML, on the other hand, can “identify patterns that involve a larger number of features and are too complex to be detected by humans,” Mani continues.

Picking a model

So we know the benefits of using ML models, but which ones are businesses turning to?

A common one is multilayer perceptron (MLP), a type of neural network that works with a set of complex, non-linear data and can be used to recognize and classify patterns within the data set.

It’s been shown to be rather accurate for business forecasting. A study of MLP’s use to predict occupancy rates in Turkey’s hospitality establishments revealed a 91.85% success rate when accounting for factors such as total number of guests, total overnight stays, and average length of stay.

Forecasting occupancy rates helps establishments make decisions concerning pricing and promotion, among other uses / Photo credit: 123RF

In contrast, a moving average analysis for US hospitality establishments had a mean average percentage error (MAPE) of anywhere between 10.6% to 44.2% depending on the city. For exponential smoothing, the MAPE ranged between 6.6% to 39.9%.

Another ML model that’s often used in business forecasting is the recurrent neural network (RNN), which comes in handy when dealing with sequential information and data.

For firms, RNN models can be extremely useful when trying to predict future trends and other purchasing decisions based on customer transaction histories. It’s considered a top-performing model according to a study published in the International Journal of Research in Marketing.

Stumbling blocks

That said, ML technology isn’t a magic solution. One major challenge with ML-based forecasting is ensuring what’s known as “coherence in the data.”

Take for example a company that wants to predict the sales volume of their entire inventory. Its inventory can be divided based on the categories of the products that it has, which can then be further divided into the individual products themselves.

As such, an ML model can only be accurate if the forecast for each subcategory adds up to the total overall forecast for the entire inventory – this is known as coherence. Getting this coherence can be a challenge because each subcategory can be forecasted based on its own characteristics, which may not account for its relationship with other sub-groups or the overall category, thus introducing discrepancies.

Supermarkets are one example of having a wide range and variety of categories and subcategories that can make coherence difficult to achieve / Photo credit: 123RF

Additionally, there are many different types of ML models available, which can make it difficult for businesses and engineers to figure out which one best fits their forecasting needs.

“It takes a lot of time – at minimum, three months – as well as effort and collaboration,” Mani shares.

Even when a business has decided on the right model, there are other challenges that remain, such as getting it into production on time, monitoring the model to ensure that predictions don’t degrade over time due to situational changes, as well as retraining the models in a timely manner.

Getting it right

To overcome this, it’s crucial that businesses – and their engineers – have the right tools. One example is Google Cloud’s Vertex Forecast, which is a solution powered by Google’s AI research team Google Brain.

Using the Google Brain team’s findings, Vertex Forecast’s platform helps businesses evaluate and choose the most suitable ML models as quickly as possible. According to Google Cloud, this can slash the time it takes to build demand forecast models to as little as two hours.

Vertex Forecast’s algorithm is also adept at ensuring coherence, as Google Cloud uses an AI architecture that transforms historical information into a set of data points known as vectors. The model then generates predictions based on these data points.

These advantages have benefited several companies, one of them being US-based retail firm Lowe’s. Using Vertex Forecast, Lowe’s created more accurate models and predictions that could account for the subcategories within its inventory data. These included its store-level, SKU-level, and region-level inventory.

Another company, Brazilian retailer Magalu, also used Vertex Forecast to great effect. The firm used the solution to reduce inventory prediction errors, which helped it allocate and replenish inventory more efficiently and lower inventory management costs.

Reacting to major changes

Going forward, companies may lean toward the use of AI and ML in business forecasting especially following black swan events like Covid-19. That’s because traditional models have a tendency to rely on historical data, and large, violent changes as seen in the pandemic make such trends essentially useless.

A quote in a report by E2open illustrates this: “Traditional forecasting is based on the presumption that history repeats itself.” However, during the pandemic, “it completely failed.”

Covid-19 isn’t the first world-changing event and it surely won’t be the last. As such, the speed at which companies can put ML-based forecasting models into production will be a crucial advantage in the coming years.

According to Mani, “It is necessary for companies to capture such variables and quickly come up with accurate forecasting to proactively address their business needs.”

Google Cloud accelerates every organization’s ability to digitally transform its business. The company delivers enterprise-grade solutions that leverage Google’s cutting-edge technology – all on the cleanest cloud in the industry. Customers in more than 200 countries and territories turn to Google Cloud as their trusted partner to enable growth and solve their most critical business problems.

Google Cloud celebrated the fifth anniversary of its Singapore region on August 23 and will be hosting a digital broadcast of the celebration on September 15. The broadcast will highlight Google Cloud’s contribution to Singapore’s development in the past five years and how it will move forward with sustainability and AI at its core in the next five years.

Sign up for the digital broadcast here.

This content was produced by Tech in Asia Studios, which connects brands with Asia’s tech community. Learn more about partnering with Tech in Asia Studios.