Forecast products, minimize losses and save money with Machine Learning
How a large semiconductor manufacturer leveraged tailored Machine Learning models to improve demand forecasting.

INDUSTRY SEMICONDUCTOR

The client

The client is a semiconductor manufacturer with over five billion devices used in consumer electronics such as smartphones, Google Home and Alexa. This client ignited the human interface revolution with their touch, display and biometrics products- all built on the company’s storied research and development, extensive intellectual property and global partnerships. With solutions designed to optimize the user experiences in the mobile, PC and automotive industries, the client combines ease of use, functionality and aesthetics to enable products that help make users' digital lives more productive, secure and enjoyable.

Overview

This client looked to Google Cloud and Pluto7 to help them proactively reduce excess and obsolete inventory through better demand forecasting. The thought process was that by using accurate forecast to drive the production schedules, they will produce products that were in line demand which results in reducing the amount of excess inventory.

By using Pluto7’s Demand ML™, the company achieved a demand forecast with greater than 90%+ accuracy compared to a baseline of 72%- successfully reaching their target. More importantly, 8 out of 10 product lines used AI and ML model and outperformed their current forecast models.

Challenges

Currently, business leads are hesitant to use automated forecasts for unique products since accuracy is low and unreliable

The client wanted more control over their supply management, especially around usage, multi-echelon inventory, and inventory postponement techniques by leveraging their forecast

The client is operating at excess and obsolete levels at 30% over build plan inventory which is costly

Strategy

Utilize Google Cloud Platform to collect all related data

Leverage Machine Learning and Artificial Intelligence to produce high accuracy forecasts for target products and compare these to the current generated forecasts

Results

ML and AI models were able to develop a demand forecast with over 90%+ accuracy that was utilized by product forecast team as an innovative solution

With this high level of accuracy, business leads are now realizing the paradigm shift and going through the change management driving adoption of ML generated forecasts, reducing the number of manual overrides

Over 4 weeks of parallel testing, the ML model outperformed existing forecast 8 out of 10 product lines.

“Earlier our demand planners did not believe the automated forecasts for the products that were chosen for the Proof of Concept and were constantly making manual overrides. With the high accuracy forecast and excellent results from the parallel testing following the POC, our demand planners have started to believe that they can rely on the automated forecast with ML.”

- Sr. Director of Applications, Semiconductor Company

The client has experienced rapid growth, with its products being used by many OEMs of mobile devices and related products. With various end users and a large variety of mobile products, the company has not been able to develop its own reliable demand forecast for each unique product. This has resulted in unreliable forecasts causing downstream supply chain issues like excess and obsolete inventory levels exceeding 30% of build plan which is very costly to the client.

What does it take to achieve a high accuracy product demand forecast?

The product demand forecast needs to be of sufficiently high accuracy in order to help drive optimized downstream supply chain planning and execution. To achieve a highly accurate forecast, multiple sets of data need to be used. Internal data such as regional aspects, sales orders, order shipments, product catalog, inventory levels and related information needs to be analyzed for trends and correlations. Achieving this requires the ability to pull all the data together and process it at business real-time processing speeds, while leveraging advanced predictive capabilities like those provided by Artificial Intelligence and Machine Learning. The key element here is that each product and customer has unique demand patterns which can mainly be identified and monitored efficiently through ML based forecasting models that learn continuously. This is a perfect problem for Google Cloud Platform- with its data storage, analytics at scale and associated Machine Learning- to solve.

“Since none of our existing automated forecasting techniques were working for the products in question we had to look at Machine Learning and Artificial Intelligence as a potential solution for automated demand forecasts that our demand planners could trust.”

- Sr. Director of Applications, Semiconductor Company

Now able to get high accuracy product demand forecasts

With these objectives in mind, this large semiconductor client partnered with Pluto7, leveraging Google Cloud Platform to transform their data. Utilizing analytics, Machine Learning and Artificial intelligence, Pluto7 and GCP helped them create product demand forecasts with over 90%+ accuracy, and one Machine Learning forecast that beat the current forecast 8 out of 10 product lines.

With Pluto7’s expertise in Advanced Analytics, Machine Learning and Artificial Intelligence, we were able to identify the right Google solutions and quickly develop a Proof of Concept that demonstrated how to use both internal and external data to produce a multi-attribute product forecast with high accuracy.

“Google Cloud Platform’s immense processing power and its state of the art machine learning capabilities led to a very high accurate demand forecast that could be delivered to the customer. It is a new paradigm in demand forecasting where you can build ML models with high level of granularity of your customer-product segmentation demand economically and plan a more realistic supply need.”

- Manju Devadas, CEO & Founder, Pluto7

Stronger through innovation

Working with Pluto7, this customer is experimenting with Google’s Machine Learning technology to leverage more data, analytics and Artificial intelligence. Ultimately, the objective is to achieve breakthrough innovation and business transformation by expanding the success of leveraging Google Cloud, AI and ML to related use cases and beyond.

Why Google

This client chose Google Cloud Platform because it:

  • Offers a flexible, scalable, ready-made infrastructure in the cloud.
  • Provides a cost-effective platform that’s easy for marketing team members to use
  • Delivers speed, security, reliability and flexible pricing

Products used

Google Cloud Platform

Google BigQuery

Google Cloud Storage

Google Cloud Dataprep

Google Machine Learning

Tensorflow