Client Overview
Data Sleek recently had the opportunity to work with HyperWolf, a fast-growing cannabis delivery company, to optimize their delivery mechanisms. They operate in a competitive environment and know that it’s important to keep delivery operations running smoothly. This project culminated in the targeted application of Demand Forecasting providing valuable insights for multi-region support. The overarching goal of our work was to help HyperWolf provide affordable products while minimizing time to delivery. This would be based on insights about consumption gleaned from historical data. The team at HyperWolf prides themselves on getting orders to customer doorsteps as quickly as possible. One of the ways HyperWolf tries to accomplish their customer service mission is to anticipate needs. This way they can fulfill orders quickly.
Collaborating to Implement a Demand Forecasting Model
Our team worked with HyperWolf to optimize the development of quantitative decisions about stocking inventory. The goal is to stock in warehouse locations to suit geographically dispersed fulfillment needs. Data Sleek used an econometric approach, building demand forecasting methods to enhance heuristic insights about the data that HyperWolf’s team was already bringing to their analysis. We determined such conditions might also play into the decision making model that HyperWolf would need. With this on hand, the demand forecasting process would enhance that area of insight, as well.
We also hoped to alleviate HyperWolf from the burdensome and time-consuming task of producing KML files by hand.
What is Demand Forecasting?
Demand forecasting is a process that allows data scientists and analysts to estimate the size of demand. This is done for a period of time in the future by using historical data. At a high level, this is done by looking at the interactions of overarching trends and repeating patterns. This type of predictive analysis helps companies to make smart decisions about how much inventory to keep in stock. Demand forecasting can also be useful in planning for other considerations. Examples are related to marketing, human resources, expansion plans, and distribution. Demand forecasting is also helpful in detecting anomalies.
Trends at play may be the result of a growing or shrinking market, macroeconomics, or changes in brand awareness. Patterns are also important and they tend to reflect cyclical rise and fall associated with seasons. Those patterns of seasonality don’t have to be aligned with the seasons of the year. They can be observed in cycles over the course of a day, a week, or a month. The kinds of patterns you observe in data over time will depend on the nature of the demand for the product. The demand forecast would take all of these patterns into consideration. This would also allow the company to work hand in hand with partnering manufacturers supplying the inventory.
Considering Patterns and Seasonality
Business-specific time series often have multi-period seasonality because that mirrors the patterns humans tend to operate with. For instance, analysis of one product might be heavily influenced by a 5-day work week and that would be reflected in the time series data. Patterns related to another product might be more aligned with vacation schedules or special events.
At its core, forecasting is a tool that helps business leaders to manage risk while capitalizing on opportunities.
HyperWolf’s Use-Case
Data Sleek’s work with HyperWolf was actually a multi-step process in which we helped them to figure out the most efficient way to break down their geographic distribution regions. We helped them build a KNN model, as a way to address segmentation, and then determine where those clusters observed in the model would be overlaid geographically.
As mentioned in the introduction, a time forecast aims to observe the changes of independent variables over time and how those independent variables impact a target variable. In this case, the target variable was sales volume and our analysis was able to help this company not only to predict the volume of sales overall, but also to predict volume of sales for each individual region. With that information, HyperWolf is able to make more informed decisions about how much inventory to stock for each region.
Opportunities for Growth
Before this analysis, the company had been making decisions about how to distribute stock manually and heuristically, which can be time-consuming. They saw an opportunity to make more data-informed decisions during their collaboration with Data Sleek. This became increasingly important during HyperWolf’s growth in sales. They subsequently and saw the need for more efficient business operations that they implemented programmatically. To support the growth they saw, and to drive further business growth, HyperWolf sought dynamically generated KML regions for its drivers to maximize the business’s operation efficiency and profitability. Emphatically, this is a clear opportunity to differentiate themselves among the competition.
Data Sleek took a multistep approach for generating a dynamic and optimized geographic data file. First, we implemented data warehousing according to the client’s requirements, then pre-process the resulting dataset. After that, we were able to geographically cluster historic orders data using a K-Means clustering method which then allowed us to utilize a geospatial data package to allocate clustered regions on a grid map and form new KML regions. This resulted in a KML file based on the variables that made the most sense for this particular business’ use-case. Finally, using Prophet, a forecasting model developed by Facebook, we generated a demand forecast for each region in the KML file.
One way that this type of demand forecasting stands out as a method for observing demand forecasts is that it operates based on a few assumptions. Notably, one of those assumptions is that there is an analyst in the loop. An expert in the operations at a specific business can interact with tunable hyperparameters that are human-interpretable. Retail businesses are often watching for trends and related to sales and special events which is especially straightforward to adjust for by adding them manually.
Making Predictions Using Historical Data
Now they have a better idea of how to break down the business they do into discrete geographic regions. Using this forecasting method, HyperWolf can now use their ever-growing bank of warehoused historical data to predict the number of sales in each region with the demand forecast, enabling business operations.
Most importantly, HyperWolf can now predict sales volume by region, several weeks in advance.
Alternate Applications
In addition to being able to plan for routine operations, HyperWolf can use those same demand forecasting methods to address other business challenges because Data Sleek helped them to optimize their warehoused data.
An example of where HyperWolf could go on to apply demand forecasting is in holiday sales. They may want to be able to predict that there would be a large demand for their products and prepare for it in terms of staffing, inventory, and marketing plans.
Outcome Summary
By consulting with Data Sleek, HyperWolf was able to leverage Machine Learning. They synergistically increased the effectiveness of their own expertise in the cannabis delivery market. Their expertise in this niche gives them an intuitive understanding of how they can tune the model’s human interpretable parameters. This offers more granular control over how they make future-forward and well-informed decisions as they grow.
Could Your Business Benefit From Demand Forecasting?
A similar approach could be used to optimize operations whether your company is focused on working within local economic conditions or is running at a large, international scale.
If you’re ready to incorporate similar strategies in Data Science applications at your business, set up a meeting to consult with a Data Sleek pro today.