As technology continues to develop and the demand for products that are available and quick to arrive grows, you’ll need precise predictions of what customers will want in the short term. Since we can’t always predict what today’s and tomorrow’s customers will do by past behaviour alone, Demand Sensing is a perfect tool for bridging that gap. By anticipating changes in demand in the short term - typically less than ten weeks - businesses can reap a multitude of benefits, from reducing aged inventory, overstocks, and lost sales to increasing on-shelf availability. But, before we get into all of the reasons why you should include Demand Sensing in your supply chain management, let’s break down what it is and how it works.
What is Demand Sensing?
In a nutshell, Demand Sensing is a subset of Demand Forecasting that is used to predict future demand for the short term, usually on the hourly, or daily scale for up to 10 weeks into the future. This means that you can pick up on short-term dips and inclines in demand that would otherwise be smoothed out in long-term forecasts. Sounds straightforward, right?
Well, not exactly. Let’s get (somewhat) metaphorical.
Let’s imagine that we’re making some alcohol. We throw in our ingredients - some yeast, sugar, and water (our inputs) - and use distillation (our Machine Learning algorithms) to separate out our desired product, i.e. the demand signals that are useful for predicting demand for a given item, location, or time horizon. To distill out these useful components, we want to apply heat (the algorithms) to separate the delicious and slightly warm alcohol (our predictive demand signals) from the methanol and yeasty sugar-water (our noisy data). And, once we’ve distilled our alcohol, the mix of sugar, water and yeast that’s left behind can’t be used again, just like how our real-time data is no longer useful once the period that we’re forecasting for has passed. Where this metaphor falls short is that the Machine Learning algorithms have the ability to learn and identify how the effects of our demand signals on demand flux over time.
Unlike other types of Demand Forecasting which primarily use historical sales data and are better suited for long-term forecasting, Demand Sensing is very accurate in the short-term because it can use real-time data. Sources of this data vary and can include:
Demand signals - think customer, POS, and channel data
Web analytics
Social media trends
Environmental data (such as weather forecasts)
Historical shipments and future orders
Competitor pricing and promotions
Mobility data, including how many people come in store or even use public transport
To make sense of all of this information, Demand Sensing uses Artificial Intelligence to identify patterns in the data and determine which variables affect demand and our predictions (as mentioned in the earlier example), which further improves the accuracy of forecasts. It’s also important to note that the forecasts you produce are only as good as the data that you feed into your model. Specifically, the data that you use needs to be useful and accurate for the period that you want to forecast for. For instance, a supermarket will tend to find that weather forecasts are a useful demand signal for predicting demand for hot or cold items, while a furniture retailer will often see that weather is just part of the noise.
Why do we need Demand Sensing?
While Machine Learning has been applied to Demand Forecasting before (and with great success), the great thing about Demand Sensing is that it can capture how customer and consumer behaviour can change on a dime in response to a variety of different events. Examples include company decisions - particularly promotional and competitive activities - or events outside of your control - from social media to economic changes, disruptions arising from major crises such as COVID-19.
If you’re wondering why we can’t just use real-time data with other Demand Forecasting methods and avoid using Demand Sensing altogether, we’ve got you covered. This is primarily due to differences in the kinds of calculations needed for historical data (which time series and other methods rely on) compared to those needed for real-time data.
Limitations
Demand Sensing is incredibly useful, but it isn’t the answer to all of your forecasting problems. The main issue, (as mentioned in our brewing metaphor) is that the data we’re using is only really useful for forecasting up to 10 weeks into the future and, once we start to forecast beyond this period and our data is getting older, Demand Sensing loses the edge it has over other methods. To avoid this, you’ll need to pivot to a method that is better suited to this length of forecasting, such as seasonal forecasting models. Additionally, focusing on short-term forecasting and neglecting your medium and long-term plans can result in breakdowns in other parts of the supply chain.
Final Thoughts
The need for accurate, short-term forecasting (and, thus, the need for Demand Sensing) will continue to grow and become increasingly crucial for supply chain management. With benefits like improved accuracy and the ability to use more data-rich sources than those available for long-term forecasting methods also makes a convincing case for deploying it in your forecasting strategy.
Want to find out more about Demand Forecasting? Why not have a read through of our case studies, where you’ll find out how we’ve used demand forecasting to help increase stock availability and improve the accuracy of forecasts. Or, check out our blog for the latest AI reads and the answers the oft posed question, ‘What’s the difference between Demand Forecasting and Demand Prediction?’ here.
Or, if you’re ready to chat about Demand Forecasting and our platform, drop us a line here.