How can you use AI to better understand how causal forecasting can help your business? Logility’s own, Chris Mason, shares his opinion and expertise on how business leaders can leverage causal forecasting software to be better prepared for quickly changing demand.
“Are you using all your available data to improve your forecast?
Most of us in 2023 have heard that Artificial Intelligence, (AI) can help manufactures and supply chains optimise inventory plans. This piece is for those us wanting actual examples to understand and relate to our own lives.
Statistical forecasting is a well-established method for anticipating demand to base an inventory plan around. The approach considers historical sales of a single data point e.g., a SKU, and uses predictive analytics to estimate future sales. Useful but limited. Whereas an AI engine can consider multiple data points at the same time and look for correlations between them which translate into new insights.
A go to example of AI in action is a driverless car, but realistically any machine, which sends usage information back to its manufacture, (think IoT) is a relevant data source for Artificial Intelligence. If your company makes decisions on how many spare parts to put in market of anything from a car to a kitchen appliance like coffee machine, or a tool used in construction, or your garden, then causal forecasting can turn data into meaningful information.
By capturing cycle counts in machines you can gain insight into how long a given part will last. So, if you then consider the product life cycle statistic, in the context of both the size, and age of your install base, you have more intelligence about how many spare parts to a put into market.
Advanced analytics with AI can refine your understanding further with scenario planning that uses more data points such as regional bias, the weather, related events and competitive pricing.
What’s the difference between AI, machine learning, and causal forecasting?
With the different acronym’s, this subject can get confusing. It is not unusual to hear about machine learning and artificial intelligence in the same sentence and then comes along Causal Forecasting. Geeks for Geeks.org, defines AI as a broad term that describing intelligent machines that can perform tasks that would otherwise be carried out by people. Machine Learning, (ML) is a subset that relates to teaching machines to learn without being “explicitly programmed.” ML algorithms are used to identify patterns and make data led predictions. Causal Forecasting is a technique that seeks to forecast events in the marketplace based on factors which have a high probability of influencing the future movements within that market.
How do you decide if a casual factor is really a cause to consider?
Weather is a commonly cited causal factor, but is it always useful? If you live in the UK, the weather can change so suddenly that the weather forecast is a national conversation. Some causal factors are simply more useful to some applications than others and it’s helpful to be mindful of the limitations.
For example. If you’re a producer of beer and the 10-day summer weather forecast is good and there is also a major sporting event on, the safe money says produce more beer to satisfy a forecasted peak in demand. How high the peak will be and how long it will last, will be influenced by other factors, such as how far a sports team goes in the event.
Another variable worth consideration in the scenario planning are the betting odds – if the local team is tipped to do well, it makes sense that more people will be excited by the prospect of watching that game with a beer in their hands.
How the long-range weather forecast impacts the sale of convertible cars is likely to be more complicated and arguable less useful. Does last year’s heat wave, or this year’s forecast have a bigger impact on sales? Or is it the case that it is macro-economic factors like inflation and consumer confidence, are the factors that really move the dials?
If you have a gut feeling that a causal factor like the weather or an event may improve your forecast accuracy, and you want to test the theory without risk, you can use a causal engine. This engine will validate the theory by running multiple items from a product database through the AI Causal algorithm over time to prove the correlation.
How much time and processing power required to run the AI will depend on the application, but when considering demand forecasts, you want it to be long enough for ancillary sales to be captured. In the world of FMCG for example, establishing the impact of competitive pricing on your sales is desirable to know.
Can AI help me understand if a maintenance, or a breakdown and repair strategy is more useful?
Users of systems like Logility’ s Casual Forecasting can tap into leveraging cycle counts which provides insight into how long parts last. Here causal forecasting is used to evaluate the cost of the repair strategy vs maintenance strategy based on variables such as supply chain disruption.
If your goal is to reduce inventory at a location, then a break fix strategy is appealing. You only need to order, pay for and store the part when you need and not before. However, this needs to be balanced against the cost and disruption of down time whilst the replacement part is sourced.
If the supply chain is uncertain, you might want to have a maintenance plan, so you do not have to worry about parts being on back order or have to absorb down time associated with a broken machine. Individual parts may have different strategies depending on how critical the part and its function are. For instance, if the part is a blade that wears out over time it will behave differently to a motor which just keeps going until one day it will not start because the wiring burns out.
The data led insights don’t stop there, because the usage data from cycle counts can also inform you as to when is a part likely to fail.
You can then model how many spare parts are needed in a given location and when
It is one thing to use time series modelling to look at part failure’s month over month, but looking at this data in isolation doesn’t really turn data into usable information. With AI one can take into account the pace of the growing install base and consider multiple data sets such as how many parts are out in the market needing to be maintained, how old the various machines are and how much they have been used. It becomes easier to determine how many spare parts will be needed, where and when. Something that sounds simple in theory but has been causing the spare parts industry big headaches for a long time.”
Chris Mason is a seasoned growth marketer at Logility who draws on his 20 years of experience helping clients optimize their supply chains. Inspired by innovation, Chris is passionate about empowering sustainability initiatives and helping clients reduce inefficiencies in their complex global supply chains. Chris is a prolific writer on trending supply chain