Using Machine Learning on Google Cloud Platform to Transform Demand and Supply Balancing
Pluto7 Planning In A Box (PiAB) SaaS offering on Google Cloud Platform leverages machine learning to help customers forecast highly accurate demand to drive the most efficient supply management, leading to reduced operations costs and increased productivity.
Customer Challenges that PiAB solves:
- How to accurately predict demand for products.
- How to appropriately manage supply levels that are responsive to changes in demand.
- How to have exception-based notifications and alerts to minimize human decision making.
Our SaaS solution
- PiAB leverages Google Cloud’s unbeatable data processing, analytics and machine learning capability to provide its core functionality of highly accurate demand forecast and demand supply balancing.
- Easy for mobile interfaces for its exception based alerts and notifications.
- PiAB is built on a docker container model and uses Google Kubernetes Engine to deploy the SaaS as images to customers which helps avoid the drawbacks of a multi-tenant environment.
Benefits of Planning In A Box
- Delivers the most accurate demand forecasts to retailers (Typically experience 85%+ accuracy)
- Enables company to focus on new features and product development, instead of production problems
- Builds new features, such as chatbots, quickly, in one day, instead of many
- Grows business by leveraging a reliable, scalable platform to deliver greater value (eg delivered 50% improvement in inventory for a retail customer)
“The drive toward Google Cloud Platform was to get beyond performance bottlenecks and leverage Google machine learning on a cloud platform that scales and is cost effective.”
—Salil Amonkar, GCP Manufacturing Practice Head, Pluto7
PiAB was targeted for medium-sized retailers and enterprises that relied on distribution. The typical customer might be a seller on Amazon or have an e-commerce site and is faced by the challenge to accurately predict demand for their products to help ensure they have the right amount of available inventory when it’s needed. If they buy more inventory than demand warrants, they lose money. If they didn’t buy enough product to meet demand, they also lose money.
Playing it safe by regularly carrying low inventories does not work well since this will impact negatively on their Amazon ranking and their listings could get pushed down, making it harder for potential customers to find them.
These customers need a demand supply management solution that is easy to setup and use so that they can easily implement without much technical support to drive their business. In other words, they truly need a planning in a box solution which is exactly what the Planning In A Box team decided to solve.
The Planning In A Box solution:
Solving the inventory dilemma is what Pluto7 Planning In A Box does for its small and mid-sized retail customers. The software as a service (SaaS) offering helps sellers on Amazon, Shopify, eBay, and other online channels accurately forecast their demand, weeks and months in advance, to manage their inventories accordingly.
Planning In A Box engineers knew they needed to evolve the solution to being simple enough to use in order to grow and stay competitive. In early 2017, the Planning In A Box team began migrating to Google Cloud Platform. The team was particularly interested in discovering how Google Cloud Machine Learning Engine could help deliver the most accurate demand forecasts to their retail customers. Leveraging all the important features of GCP such as Cloud Dataflow based data processing (Cloud Dataflow), advanced storage and data warehousing (Cloud SQL, BigQuery), docker-container model, Google Kubernete Engine, Google Data Studio to help design, develop and deploy an easy to use demand supply balancing solution.
Behind the scenes of the Planning In A Box offering, multiple Google services, including Google Kubernetes Engine, Google Cloud SQL, and Google BigQuery, “work together seamlessly to automate demand forecasting using a growing amount of datasets, while delivering increasingly detailed reporting analytics and more,” notes Salil Amonkar, GCP Manufacturing Practice Head for Pluto7.
Planning In A Box is built on a Docker container model and uses Google Kubernetes Engine to deploy the SaaS as images to customers. As a result, users can access the SaaS front-end, as well as applications that run in their own environment on the back-end, as separate instances—which helps them avoid the drawbacks of a multi-tenant environment.
“The drive toward Google Cloud Platform was to get beyond performance bottlenecks and leverage Google machine learning on a cloud platform that scales and is cost effective,” Salil adds.
The Planning In A Box team could leverage Google Cloud Platform quickly and were soon able to release using an agile approach the demand forecasting solution which met the required level of demand forecasting accuracy in a Proof of Concept. Using a time-series forecasting model, forecast accuracy was significantly improved, especially compared to the statistical average forecasts the company’s previous solution provided. This is an example of a best practice that the team leveraged for enhancements and other professional service delivery of use cases related to demand forecasting.
Using the docker-container model and leveraging the Google Kubernetes Engine for deployment was another example of a best practice that was leveraged by the team to avoid typical pitfalls associated with multi-tenancy models.
“Google Cloud Platform is far more reliable than the previous platform Planning In A Box ran on, which would crash periodically,” says John Nikhil, Head of Growth & Sales, Planning In A Box, Pluto7. “We’d have to restart the server, get the data back in, and then do the forecasts again, and it was a drain on our resources that required two data scientists to keep the machine learning server running.”
Having to re-input data after a crash could also affect the accuracy of forecasts.
Google Cloud Machine Learning Engine continuously works around the clock and it never crashes. As a result, forecast data can be requested at any time and without crashes there is no worry about the data being incorrect due to system interruptions.
“It was so easy to deploy and test Google Cloud Machine Learning Engine without making any commitments or investments, especially since Google sweetens the deal with a credit for trying its services.”
—John Nikhil, Head of Growth & Sales, Planning in a Box, Pluto7
Everything In A Box
The freedom to experiment, along with the stability, scalability, cost-effectiveness, and breadth of services available in Google Cloud Platform, encourage Pluto7 to expand its vision of Planning In A Box services. The goal is to become a total inventory management solution, giving customers a full view of their inventory in warehouses, manufacturing, and stores, and integrating all their online channels, including Amazon (in the US and internationally), Shopify and eBay.
“We want them to see, at the click of a button, their complete inventory across the world, including the ability to manage their distribution, all in one place,” John says. Or, as the product name implies, to truly have all their "Planning In A Box."
Benefits of Planning In A Box:
Highly accurate demand forecasting (typically 85%+)
- Since Google Cloud Platform deployment was completed in Q2 2017, Planning In A Box customers have already experienced the benefits of more accurate forecasts—notably during the recent busy holiday shopping season. Planning In A Box delivered a forecast in August to one of its customers, an Amazon seller who specializes in kites. The forecast predicted demand for the seller’s kites would increase nearly 300% during the holiday season.
Drives optimized inventory (Some customers saw 50% improvement)
- Based on that prediction, the seller geared up its manufacturing and inventory months in advance to prepare for the holiday demand.
Driving higher revenue
- Due to an extremely accurate forecast that was delivered well in advance of the peak seasonality, the seller maximized revenues for the holiday shopping season which is notoriously challenging to predict.