Recent advances in AI, such as foundation models, make it possible for smaller companies to build custom models to make predictions, reduce uncertainty, and gain business advantage.
Time series forecasting is a powerful machine learning method that leverages historical time-stamped data to predict future events and help reduce uncertainty from business conditions — for example, to accurately predict sales, inventory levels, and even manufacturing data.
Much of the data your company has is already time-stamped. It’s probably sitting in Excel spreadsheets, brimming with potential. Here are five ways you could use that data for time series forecasting.
Turn Excel spreadsheets into future knowledge about your business
You’ve been collecting information about your business for years, all stored neatly in an Excel spreadsheet. That data tells the story of where your business has been, but you can also use it to predict what will happen, what demand will look like, the cost of materials, or how shipping times might change. Times series forecasting utilizes time-stamped data — whether that is dates, years, hours, minutes, or seconds — to analyze past temporal patterns and make predictions about the future relevant to your business. If you’re just starting with time series forecasting, new out-of-the-box foundation models let you get started immediately. Foundation models are already pre-trained on large data sets, so during inference, you can directly input your data and quickly see predictions without further training. Options for these foundational models include Nixtla TimeGPT-1, Amazon Chronos, Google TimesFM, Salesforce Moirai, Lag-Llama, and MOMENT. TimeGPT has an Excel plug-in that lets you do the forecasting from within Excel.
Use information about external factors to predict things about your business
You’re a global lawn care company. Fertilizer demand differs in the northern and southern hemispheres according to the weather. The timing of when seasons begin has been changing, and soil types vary. A few days’ shift, or ingredients in soil conditioner, makes the difference between a successful season or not. You need to use data from previous years’ purchasing trends, such as when they started, how they’ve shifted, and how demand changes once the season starts. Combine this data with time series forecasting to enhance your forecasts and make better decisions. These external factors are called exogenous variables, and they’re crucial in time series forecasting as they provide additional information that might influence the prediction. These variables could include holiday markers, marketing spending, weather data, or any other external data that correlates with the time series you are forecasting. You can incorporate these exogenous variables into many of the foundation models.
Measure uncertainty with probabilistic forecasts
All of us are betting on the future, trying to figure out what to do now to maximize desired outcomes. This is true whether you are planning insurance portfolios or determining airplane fuel rates. Probabilistic models can predict the distribution of future outcomes — how likely each outcome is — and therefore provide valuable information for decision-makers to weigh the risks and rewards associated with different actions. Here, foundation models will give you a good start, but you may want to try different models to determine which will be most accurate for your use case. There are many models! A great overview is Rob J. Hyndman and George Athanasopoulos’s book Forecasting: Principles and Practice. Fortunately, many of the models have been implemented in Python and R, so you can fine-tune them using these tools.
Evaluate different future scenarios
You can use TimeGPT to forecast a set time series, such as the demand for a retail product. But what if you want to evaluate different pricing scenarios for that product? Performing such a scenario analysis can help you better understand how pricing affects product demand and can aid in decision-making. This is another way of using exogenous variables, but we change how we use them.
Detect anomalies that will affect your predictions or identify unusual patterns
Anomaly detection involves identifying unusual observations that don’t follow the expected patterns in the data set. Anomalies, also known as outliers, can be caused by various factors, such as errors in the data collection process, sudden changes in the underlying data generating process, or unexpected events. They can pose problems for many forecasting models since they can distort trends, seasonal patterns, or autocorrelation estimates. As a result, anomalies can have a significant impact on the accuracy of the forecasts, and for this reason, it is essential to be able to identify them. Furthermore, anomaly detection has many applications across different industries, such as detecting fraud in financial data, monitoring the performance of online services, or identifying usual patterns in energy usage.
The biggest companies in the world employ large teams of machine learning engineers who build custom models to leverage time series data to make predictions. Recent advances in AI, such as foundation models, have made it possible for smaller companies to do the same with fewer technical requirements and expertise. Data has become the DNA of business. Make sure you’re leveraging what you already have to gain the best advantage.
Cristian Challu is co-founder and chief science officer at Nixtla.
—
New Tech Forum provides a venue for technology leaders—including vendors and other outside contributors—to explore and discuss emerging enterprise technology in unprecedented depth and breadth. The selection is subjective, based on our pick of the technologies we believe to be important and of greatest interest to InfoWorld readers. InfoWorld does not accept marketing collateral for publication and reserves the right to edit all contributed content. Send all inquiries to doug_dineley@foundryco.com.
