In the past, it might have been true that human behavior was too complex and chaotic for mathematics to predict. But today, it’s absolutely true that we can use a data-based approach to predicting how humans will behave.
To do this, it’s necessary to alter our approach – Instead of looking at the way humans behave from our human perspective, we need to study the sales data. Even more crucial is understanding not just demand forecasting systems, but to be able to predict behavior in unpredictable and improbable circumstances, such as the rise of the Coronavirus pandemic, unforeseen economic turbulence, and similar situations.
What is Advanced Demand Forecasting?
The purpose of demand forecasting is to address these processes:
- Supplier relationship management. It becomes far easier to grow or shrink your number of suppliers or explore new supply chains using demand forecasting. With it, you can predict customer demand in quantifiable terms, allowing you to determine how much product to order at all times.
- Marketing campaigns. Demand forecasting can be used to tailor and tweak marketing campaigns and specific ads to increase sales. It’s even possible to leverage marketing data with complex machine learning models.
- Order fulfillment and logistics. Demand forecasting systems can smooth out the supply chain. This means both avoiding situations where a product is out of stock and times when unsold goods waste excess retail space.
- Manufacturing flow management. Demand forecasting fits within an overall ERP system, predicting your production needs by projecting how many goods you produce will be sold over time.
- Customer relationship management. Customers who are planning to make a purchase expect immediate availability of their products. A Demand Forecasting system can help project what products will require purchasing in a given time from certain locations. As a result, customers are happier and more committed to your brand.
Now with the new machine learning algorithms arrived, most businesses are seeking to take advantage of them in conjunction with big data to automate and optimize business processes.
Techniques utilizing machine learning technology offer predictive power over the quantity of services or products an enterprise will need to buy during any bounded future time frame. Machine learning offers the following advantages over traditional forecasting techniques:
Gives a forecast with greater accuracy
- Is capable of analyzing larger quantities of data
- Processes data at a quicker speed
- Is able to discover patterns hidden within data
- Uses recent data to automate updates to existing forecasts
- Cultivates a robust system capable of dealing with a variety of circumstances
- Is able to adjust to changes both in short and long term
It’s worthwhile to touch on some of the real-world applications of advanced demand forecasting. Note that the below companies are not the only ones to use it successfully. Many retail enterprises have found success with an ML-powered approach to demand forecasting.
Walmart has focused on the supply chain with its demand forecasting tools. They begin with identifying which products to sell, locate vendors and create equitable deals for those products. Walmart projects customer demand using a blend of historical data, promotions and sales data, and the behavior of its competitors and any changes in trends.
Amazon filed a patent in the field of anticipatory shipping, allowing them to accurately forecast demand at the intersection of product and neighborhood/city using AI technology.
IBM has created IBM Planning Analytics, a hybrid analysis, forecasting, and planning platform capable of assisting in predicting customer demand both in the present and in response to future trends. IBM uses its TM1 technology here, allowing for incredibly detailed what-if analysis and sophisticated dimensional calculations.
Avercast was founded by professionals in the field of supply chain management and has created a cloud-based system that takes advantage of a suite of 208 forecasting algorithms. Avercast’s solution offers ABC analysis, management of promotions, optimization of containers, calculations of safety stock, regression factor and adjusting forecasts on the fly. The system can integrate with CRM and ERP solutions, facilitating communication between sales staff and vendors via web portal and secured smart devices.
Smart Software has developed Smart IP&O, a system that calculates expenditures along with operational performance to get a clearer picture of classification metrics and inventory segmentation. Smart IP&O takes advantage of being hosted on AWS (Amazon Web Services) as a cloud-based solution, transforming the planning process into a far more collaborative platform than traditional methods.
How to get an accurate demand forecast?
One of the most critical components of accurate forecasting is quality data in the appropriate quantities. Any data we collect must be cleaned, checked for any gaps, analyzed for its relevance and finally restored.
Many machine learning models exist, and the correct choice for any given system depends on the business goal in mind, the time period of desired forecasts, and the quality, quantity, and type of data available. If you’ve done some reading in the field of demand forecasting, you may be familiar with approaches like Linear Regression, Random Forest, Time Series, and Feature Engineering. These approaches are functional in the majority of demand forecasting use cases.
Metrics for success provide a tangible definition of value within a demand forecasting context. A representative message might say something to the effect of:
“I need a solution using machine learning capable of projecting demand for x products over a span of the next week/month/six months/full year with an accuracy threshold of y percent.”
Anomalies in Demand Forecasting Systems
It’s necessary to be aware that any demand forecasting system is vulnerable to anomalous events, such as the Coronavirus pandemic.
A model for forecasting demand is generally working off of historical data and has no way of knowing that an unlikely external factor has radically transformed demand. For instance, the demand indicators for antiviral drugs or facemasks are utterly different this year than they were at the same time last year.
In this type of circumstance, a few ways of generating an accurate forecast exist.
The first is to take some time collecting further data on the new market’s behavior and then build a new model from the beginning. The second is to take an approach involving feature engineering, making use of price index, the state of the market, events in the news, exchange rates and related economic concepts.
Machine learning has a wide range of uses beyond demand forecasting, and the future of machine learning is bounded only by the limits of how we use and develop it. A decade ago, we wouldn’t have been able to fathom using a statistics-driven approach to such a myriad of solutions. Now, we create demand forecasting systems with a more mature understanding of the value it brings to the modern-day business.
Predicting customer demand is hard! There’s no crystal ball. What do your customers really want? And not just what they say they want, but what they will pay for?
Most product managers go through the usual channels to guess at this – talk to the sales team, read an analyst report. This is kind of LAZY, not to mention the data you collect from these sources are _trailing_ indicators, reflective of the past. To “skate where the puck will be” as the saying goes, more on-the-ground effort is needed.
I wrote a handbook on the different methodologies for getting leading indicators of customer demand in your market here:
1. Cold outreach and networking
2. Online forums
3. Ask on social media
4. Sit in with the SDRs
5. Win/loss interviews to predict the future
6. Customer success log
7. Not-so-secret shopping
8. Growth experiments
9. Product Analytics
10. Adjacent Trends