Rules Don’t Work: AI Predictions For Manufacturing And Supply Chain
The interview with Kumar Srivastava, Vice President of Engineering at Noodle.ai, a company that builds AI solutions that bring radical efficiency to manufacturing and supply chains across industries.
To start off, can you give us a synopsis of who you are and what you are responsible for at Noodle.ai?
I started my career out of grad school at Microsoft in Washington, focused on machine learning-driven spam detection. I started looking into leveraging advanced predictive analytics for building users’ analytics and prediction systems so you could determine whether somebody was trustworthy or not before you enable them to access various products and services from Microsoft.
Later I moved to the Bay Area, spent some time in the digital transformation of the company called Apigee, where we were enabling enterprises to expose their data and services through managed APIs and then leverage the power of data to make better online services. At a startup called Clear Story Data, I focused on data harmonization, which is the ability to automatically connect datasets to get a broader picture of security, which leads to better analysis, and better insights. After that, I spent some time at the Bank of New York Mellon Innovation Center in Silicon Valley, focusing on building a set of capabilities for financial services applications.
Currently, I am at Noodle.ai, where I lead engineering for an ERP Supply Chain suite of products where our focus is to enable enterprises to get the right product to the right location at the right time while minimizing cost, errors, and wasted resources. I’m responsible for product development, delivery, maintenance, support, operations, security – essentially anything that has to do with technology.
Kumar Srivastava
VP of Engineering at Noodle.ai
Kumar, there is a lack of AI talent on the market. Do you offer Data Science/AI education at Noodle.ai?
There are two kinds of talent shortages on the market. The first one is the shortage of people who have the ability to build machine learning AI-driven products and services. And then, in the business operations space, there is a shortage of people who understand how to leverage the output of machine learning, any AI products, who understand how the jobs transform when you are faced with a different set of products and services operating at a much higher level than you used to. There’s a lack of people who understand how these products connect dots and create predictions that would have typically been somebody’s gut feel or would be not present because the users are using historical data to estimate what the future might look like. The problem with that approach is that they are not leveraging the power of AI signals that exist in their data to make a predictive call.
AI changes the job for all industries and all users because the tools you are using are different. The way you leverage those tools, the information these tools provide to help you make better decisions is also radically different. We need to train the industries on how to leverage these capabilities.
Going back to your question, at Noodle.ai, we do not offer AI education, but a big part of the way we operate, the way we deliver our software, and our intelligence to our customers is the focus on designing and building our products to be educational.
We design it so that people can understand, leverage, and extend their capabilities or behavior based on the data we are providing. We help customer success differently because customer success in a machine learning-powered world is different from the traditional software world. After all, your machine learning model is constantly changing when it is getting adapted to what’s happening. The output it produces also can vary over time.
In our perception, customer success is about understanding the customer’s environment and always being prepared to help them understand the predictions in a way that gives them confidence in the output so that they can make the right decisions.
What kind of people do you look to bring to your teams to foster the innovation culture?
The answer is similar to how we build good teams – you look for diversity of thought and experience. We want people to be skilled in what they do. We look for people who are the best in the field – it could be data management, data processing, or creating machine learning data models. In general, we are looking for people who have a solid background to get the job done.
And because there is a skills shortage, you can’t find everyone at the level you need; we focus a lot on training and the ability to be trained and learn on the job. We want our people to learn and go beyond what is in front of them.
We want people who have demonstrated the ability to pick up the depth of the technology, not just being able to write “Hello world” with new technology, but they can push the envelope by spending enough time focusing on developing their craft.
The second part is we look for people with diverse backgrounds. You can’t find someone with ten years of advanced AI experience in the market. There are very few such people, and they are probably in academia. Whereas in the industry, what you’re looking for is people who have worked with new technologies and can deal with various industries’ challenges. We’ve been focusing on the supply chain when many of our fundamentals are coming across many different industries. And so, we look for people who work in financial services or healthcare or manufacturing, or it could be retail or e-commerce. If you can put that experience in a discussion where you are trying to decide about technology or a direction, you are more likely to make better decisions.
That is a remarkably interesting approach. Innovations in Enterprise AI have been a hot topic for the past few years now. What does the innovation frontier look like at Noodle.ai?
Just like beauty is in the eyes of the beholder, innovation has value in the eyes of the user. The way I look at innovation is how close we are to the exact problem that we promise our customers to solve for them. Innovation should be grounded and guided by those problems.
I believe innovation starts with the user or the customer that you are working with. When you are doing enterprise AI, you are trying to create value through innovation that is specific, easily demonstrable, and usable by the customer.
Just as I said, at Noodle.ai, we have three big problems: get the right product at the right location, at the right time, minimizing waste mistakes, make sure that we are enabling manufacturers to build products of the highest quality. We want to let our customers know upfront about how they can deal with issues and then make sure that their machinery is operating at the highest level of certainty and accuracy.
Innovation can happen in multiple layers.
We have specific problems that we go after, and we restrict ourselves to solving these problems. Those could be our techniques to detect or predict certain scenarios that the customer wants to act on. It could also be innovation in the area of serving multiple customers or different user patterns. You can innovate on the technical side and enable customers to get the value of the products and services we provide. In the innovation process, we focus on generating the value and then delivering the value.
The third bucket of innovation is constantly looking for the patterns that exist across multiple customers and using that to find out what’s the feel of this problem and then focus on generating solutions to the problems that keep bubbling up and are common or relatively common across many customers, many different industries. We try to innovate by focusing on solving problems through open-ended activities like research, prototyping, comparing, and contrasting baseline.
Can you provide a successful case study for our readers to better understand your products and services?
Sure. At Noodle.ai, we have looked at different slicing and dicing of the same problem as part of operations, trying to find AI-powered solutions to reduce production waste to enable our customers to maximize production.
A good example of our work could be pasta sauce. It has a shelf life; it cannot sit on a shelf for too long. Suppose you are the manufacturer with distribution centers in 10 locations around the world, and you have five factories distributed across the world. In that case, you will have a planner that will ensure that if there is a demand for a hundred units in Europe and a hundred units in Japan, that you are producing 200 units for that horizon, and then you are shipping a hundred units to each location.
The challenge happens when you say, “Europe and Japan will always need a hundred units every month.” And if for some reason, it could be internal issues – you introduce a new type of pasta sauce, and that starts taking off in Europe, or external issues – your competitor does something or the world changes because of a pandemic when you rely on a fixed-number model that says it’s always going to be a hundred units for these locations, you might not realize that numbers have changed and now every month in Europe you have the demand for only 80 units. You’re shipping 20 units that are going to stay on the shelf, and then you have to get rid of them. And that’s a major waste in the scope of the broader market.
If in Japan, the demand has changed for 120 units, but you are shipping only a hundred – you leave the revenue and the profits from 20 units on the table. That is something that you could have had if you could predict that the demand was going to be 120, as opposed to the rules-based on the approach of saying, “I’m going to send a fixed number of units to the same location every month or every week.”
At Noodle.ai, we want to reduce waste with advanced predictive analytics data radically, so you are not hiring trucks; you are not hiring ships and planes to send stuff where it’s not needed.
With the right data, you are not oversupplying, but you are estimating and predicting the demand so that when you get there, you are not going to lose money. If the demand is predicted to be X, you are shipping as much stock to that location of that particular type to meet the demand, which translates into meeting all the revenue that is possible.
Until now, a lot of these decisions have been dealt with in very deterministic math, applying rules that say, “if my last three months the demand was X, then I’m going to use that as my number.” Whereas what we are saying is you can use a lot more internal signals that already exist in your enterprise data systems. You can also use signals from external datasets such as mobile phone movements in a certain area or weather conditions.
At Noodle.ai, we come up with a specific estimate or prediction for the demand for a particular product, for a particular stop unit in a particular location, and then you fulfill that demand. You might want to use your machine learning model to tell you what the demand will be and then make the correct decisions, corrective actions to either manufacture more or deliver more or redirect deliveries in such a way that you would be reducing waste for yourself and maximizing the value that you can get.
Kumar, thank you for the conversation and your vision for building AI solutions that change the way industries operate.
Stay tuned for the next interviews!