Today, we’re excited to announce our investment in Abacus, a cloud-based AI platform that makes deep learning simpler and easier to use. Every executive can imagine how AI can help us all better understand and predict the underlying health of our business, but few have been able to capitalize on its benefits in the real world because of the cost and technical resources required. Abacus is working to democratize AI by offering a turnkey solution for the most widely needed business use cases such as forecasting, recommendations, classification, and anomaly detection. In doing so Abacus enables customers to bypass the most technically difficult problems faced when using cloud platforms such as Amazon, Microsoft, and Google to bring the promise of deep learning into any application or production system.
We asked Gus Shahin, CIO of Flex and member of Decibel’s Customer Innovation Council, about the immediate results Abacus has created for the nearly $25 billion global manufacturer in a Q&A with Bindu Reddy, founder and CEO of Abacus.
Gus: Flex is one of the largest global manufacturers in the world - we have 170,000 people who work every day at over 100 sites in 30 countries. We produce products for the transportation, technology, healthcare, and consumer industries and have to manage the global supply chains for tens of thousands of products and hundreds of thousands of underlying components. The world has been moving faster every day since I joined Flex nearly 20 years ago. In the past, we used to forecast our entire business using spreadsheets and historical knowledge but we acknowledged years ago that this had to change. We have had to digitize our entire supply chain, so that everyone has an end to end picture in real-time, and invest in ML and AI to help us all better forecast what demand will be for everything we manufacture around the world.
Gus: Every tech company and startup talks about how they use AI. We have been approached by everyone who would like to help us solve our demand forecasting challenges, but most of the solutions in the market, including those in the major cloud platforms, are generic tools that require us to invest in deep learning expertise in every unique domain in which we operate. Initially, we were going down this route but changed course when we met the founding team of Abacus. Bindu had experience with the AWS AI products herself, and her co-founders had run the ML infrastructure at Google Adsense and had written one of the largest cloud databases, Google BigQuery. They had their own frustration with AI tools and had a vision for making deep learning accessible for non-expert users. The founders of Abacus were some of the brightest minds in AI and were frustrated with the existing tools - you can imagine just how much more difficult this problem seemed for us at Flex. We felt the Abacus approach would enable us to finally scale AI-powered forecasting across a supply chain as diverse as ours.
Bindu: Abacus is unique in that we provide the power of deep learning for customers, but do not require them to have any ML or AI expertise to get started. This approach is very different from other platforms which generally provide tools to help very technical users create their own machine learning and deep learning systems. First, we take customer data “as-is” which removes the friction of having to clean and wrangle imperfect data sources. We can then automate the entire AI pipeline so that a customer can get best-in-class results for their use case that is readily maintained for them in production. We found that most people don’t have the luxury of building an expensive AI system to get a prediction they need - at Abacus the service is turnkey so customers can get immediate value without having to build and run the AI system itself. This may sound like a minor detail but for a customer this feels like the difference between USING Google’s search engine, and having to BUILD Google’s search engine.
Bindu: Demand forecasting is one of the hardest problems to address in manufacturing - there are so many different underlying SKUs within a product line, and each one has to be forecasted to make your supply chain work. Most traditional manufacturers use statistical methods that look at historical demand, but today’s world is dynamic and ever changing with seasonal, spiky, and unpredictable shifts. There is a lot of frustration for a technical team when your system is constantly changing - you need to clean up your data, retrain your models, and make sure they continuously perform in production. The Abacus approach streamlines all of this to make AI scalable. Now, we can generate millions of accurate forecasts on hundreds of thousands of SKUs and what used to take months to plan can now be done in minutes. This is the power of deep learning and it enables us to get to the best answer as quickly as possible.
Gus: The biggest roadblock for supply chains in both the pre and post-COVID era has been the ability to accurately forecast demand. So many things go right in your business when you can manufacture to your actual customer needs - you can order the right amount of material at every stage and make sure you are producing and fulfilling products without any production bottlenecks or supply shortages. But in today’s world everyone that manages a product line struggles with forecasting what they will need, and they are terrified about being short or product, particularly for critical components. This results in over ordering across the board, which for us can mean billions of excess working capital and inventory throughout the year. In our first major product line using Abacus, we are already saving tens of millions of dollars by having more demand forecasts with much greater accuracy which enable us to order and fulfill on a much shorter cycle time. And what we imagined would take 20-50 data scientists and engineers to manage can now be done with a single platform.
Bindu: Our team has experience with almost every major AI application and our goal was to make each of these services as simple and accessible to use as possible so we could ultimately address the most common use cases within large customers. At Flex, we have already talked to their CISO about using Abacus for IT security and detecting anomalies on system logs. And through Decibel’s Customer Innovation Council, we have discussed wide-ranging applications such as predicting cloud cost for fast growing DevOps teams, detecting fraud in financial services, and classifying customers for upsell opportunities or potential churn. The field of AI is expanding rapidly and our only constraint is our customer’s creativity for how they can transform their business with deep learning and data.