Solving A Hard AI Problem: Generating Forecasts In Sparse Data Environments!

Gus Shahin Blog Image

Anyone who’s worked in a manufacturing space can tell you that one of the most significant challenges they face is managing supply chains of the materials and speciality items used to fabricate products. Without clear foresight into demand, how, when and from where they’ll get these vital items, the manufacturing process can quickly become unwieldy and untenable. The problem is exacerbated when the manufacturer creates new SKUS and there isn’t a lot of historical data to create forecasts.

Flextronics (Flex) was founded in 1969 to address the needs of high tech manufacturers in fields as diverse as automotive, computing, and healthcare. In the 50 years since, they’ve advanced the methods by which our global economy coordinates to create vital technology. Their expertise at streamlining supply chains and facilitating communication between manufacturers and suppliers has cemented them as a gold standard in their field. 

One of the key challenges for Flex is predicting how much of a particular SKU they need to manufacture to fulfill customer demand. Since Flex deals with complex electronic systems, the problem is an extremely hard one as SKUs change regularly, with new ones being introduced and old ones being retired.

While a number of methods have been employed over the years to forecast trends in manufacturing, few have shown as much promise as deep learning. Neural Networks are capable of learning patterns across SKUs and are capable of handling the ‘cold start’ problem by understanding the relationships between SKUS based on the meta-data of each SKU. Any additional variables or factors that might influence the forecast can also be considered by Abacus.AI’s deep-learning service.

synthetic data example
Synthetic Data Created Using a GAN

To that end, Flex CIO, Gus Shahin and the rest of the Flex team partnered with us here at Abacus.AI to generate accurate AI models with actionable forecasting. Gus is well known in the industry for being an early adopter of new products and services and has helped Flex adopt cutting edge technologies and optimize it’s business processes and customer experiences. Using our plug-and-play deep learning engine, Flex is able to gain insights that would otherwise be elusive.

For example, not every client Flex works with will have troves of data going back years. Some are young companies, or some may specialize in more boutique items. This often means their datasets are sparse and noisy. Getting useful results from such datasets is a significant problem.

However, Abacus.AI has been able to provide Flex with solutions that augment small datasets. Methods such as Generative Adversarial Networks (GANs) use sparse amounts of data to generate synthetic data that has shown high levels of accuracy and fidelity. This in turn allows for reliable forecasting that typically wouldn’t be possible with the given information.

Flex’s first major product line using Abacus.AI has allowed them to generate more demand forecasts with greater accuracy than before. This has saved them tens of millions of dollars and enabled them to order and fulfill on a much shorter cycle. All this has been done with Abacus.AI’s single platform rather than the dozens of data scientists they thought they’d need to dedicate to the project.

jon sakoda
Jon Sakoda – Founder and Managing Director at Decibel

While working with Flex, we were introduced to Jon Sakoda from Decibel Ventures. We were extremely impressed by Jon and Decibel’s philosophy of working closely with their portfolio companies and helping them forge long lasting relationships with their customers. We are honored to count Decibel as an investor and Jon as an excellent champion and friend!

Related posts

Becoming An AI First Organization: The 1-800-Flowers Journey


How AI Is Helping Us With The Coronavirus Pandemic


Experience the magic of Generative Adversarial Networks

Leave a Reply

%d bloggers like this: