Guides / Demand Sensing
This guide provides an introduction to Demand Sensing in the Supply Chain and introduces key concepts
Demand Sensing is, in a nutshell, a technology to help you develop the most accurate possible understanding of future demand across your products. This technology is continuously analyzing the impact of various real-world factors that affect demand, from seasonal patterns to internal promotions to external variables such as weather, to enable you to develop the most accurate forecast. These Demand Sensing forecasts are typically leveraged across the short-term forecast horizon, otherwise known as the operational horizon, which is between Week 1 and Week 12.
For Retailers, Demand Sensing allow Demand Planners to receive Forecasts that are updated at a high-frequency, typically once a day, and provide them a view of demand with the most up-to-date short-term data such as Promotional Plans, weather, inventory levels, stock in transit, web-traffic and events. The benefits of this type of short-term forecast are numerous for most retailers, it typically provides significant accuracy improvements when compared to traditional forecasting methods, it allows them to identify short-term supply imbalances, it helps them make better allocation decisions and it helps them react to the dynamic changes in the market more quickly.
In conclusion, an accurate demand sensing forecast is the foundation by which a company can understand which products are going to be sold at which location and on which day.
It is Machine Learning that has enabled Demand Sensing. Machine Learning in this context is an umbrella term for a group of algorithms that can take in significant amounts of data and automatically infer their importance to a prediction. Whereas with previous forecasting approaches, where the Demand Planners used to overlay seasonal curves and trend lines to a baseline forecast, with Machine Learning, the algorithms do this automatically and, more importantly, also ingest a wide variety of data streams such as weather, promotions and price data and then automatically identify the relationship of each of these data streams to your demand patterns and the relationship of the different Datastreams to each other. These algorithms are especially good at understanding the relationship between the different demand drivers. For example, they can identify that an Ice cream Product sells more on a Weekend than on Weekday, but it can also identify that Sales significantly increase if it is both a Weekend and the Temperatures are higher than average.
That being said, Machine Learning/Demand Sensing is not a silver-bullet. It’s important to note that the successful implementation of Demand Sensing Technology for a retailer is contingent upon data quality. Although a lot of retailers are swimming in data, it is often limited in a few key areas:
1. Clean Historical Promotions Data
2. Historical Stock on Hand by Day by Location
3. Master Data on Product Display
If important data streams, such as those above, are unclean or non-existent, it can jeopardize Demand Sensing models effectiveness as they’re inherently subject to issues with Garbage In / Garbage out. Typically without these data streams, it’s been shown that these Demand Sensing algorithms achieve accuracies on par (or slightly better) with traditional statistical models. But if retailers do have these key data streams, then Demand Sensing models significantly outperform statistical approaches. These improvements can range from 10% to 40% accuracy improvement.
So in summary, Machine Learning enables retailers to incorporate a large number of data streams that capture key demand drivers, and then model their impact and their relationship to each other. For most retailers this drives significant value
Demand Sensing, as opposed to Demand Forecasting, is typically focused on the short-term forecast horizon (i.e. <12 Weeks) and leverages additional demand drivers, including weather and competitor activities to improve the forecast accuracy when compared to traditional forecasting models. Because the Demand Sensing Models rely on data streams that project a forward view, such as promotion plans, and these forward plans typically only extend forward to a maximum of 12 weeks in the future, Demand Sensing technologies do not typically deliver significant performance improvement beyond the 12 week forecast horizon.
Demand forecasting traditionally describes the use of statistical time series models to forecast the future demand across a 0-24 month forecast horizon. These statistical time series models leverage historical demand, seasonality and trend to provide a forecast. Demand forecasting has been a common business function since the 1980s.
Figure 1. Demand Sensing is leveraged across the short-term forecast horizon, as it provides higher accuracy as is updated at a higher frequency. Beyond this the forecast is managed by traditional forecasting methodologies and processes such as S&OP
Because of this, solutions such as Remi AI offer a hybrid of Demand Sensing for the Short-Term forecast horizon and Demand Forecasting models for the Long-Term Horizon.Nowadays Demand Forecasting has become an umbrella term that encapsulates both traditional Demand Forecasting techniques and the newer, more advanced Demand Sensing Technologies.
There are number business activities that have significant impacts of demand, including price changes and promotions. These business activities can have such significant impact on sales volumes, it is imperative that they are including in the Demand Sensing inputs.
Figure 1. Demand Sensing enables Demand Planners to factor the different business activities into the Demand Forecasts automatically
Demand Sensing is capable of modelling the price elasticity of your products, enabling you quickly understand how much a change in price will impact that product's rate of sale. This means that Retailers can start to automatically feed their Past and Future Price Changes into their Demand Sensing Software and get immediate feedback on its impact across their products.
It can even provide Retailers reports on which of your past discounts were effective in delivering increases in revenue and which of them were ineffective and saw you discount margins without increasing revenue.
Alot of Retailers plan their Promotions and Discounts days, if not, weeks in advance. Yet, despite the fact that they're planned well in advance, they're often not automatically forecasting their impact on demand.
Demand Sensing can accurately predict the impact of promotions, including complex factors such as:
1. The type of the Promotion, such as Online Coupon, Discount, Bundle
2. The size and length of the discount
3. Marketing for the Promotion, such as Brochures, Digital Campaigns, Email Marketing and others
4. In Store Display (for Brick and Mortar) or Homepage (for Online)
Accurately adjusting the forecasts for all these factors is just not feasible for a lot of retailers because the number of products to adjust is simply too high. But Demand Sensing is smart enough to intelligently identify the impact of each of these factors and adjust the forecast. This can drive significant value for the Retailer, both by reducing the time spent by Demand Planners adjusting forecasts and also driving more accurate forecasts for promotional activities. This reduces this risk of running out of stock during the promotional period and overstocking after the promotion
In summary, demand sensing offers retailers a significant opportunity to improve their operations by better anticipating demand and therefore enable more effective planning.
Demand Sensing is driving real change for a lot of retailers: A good demand sensing tool can identify the key demand drivers and not only provide more accurate demand forecasts but also provide insights into their inter-relationship.
That being said, Demand Sensing is only as good as the Data you can provide it. So if you have the Key Data streams available and in a clean state, you can look to implement Demand Sensing to liberate your Demand Planners and Managers from the depths of data analysis so they can actually use their expertise, making the critical decisions that a computer can't and making your customers love you.