Linear Squared develops world’s first AI planner for apparel manufacturing

AI significantly reduces planning time and frees up capacity that can be used to generate additional revenue. The Sri Lankan startup behind this technology is now building steam to take its AI product global

Linear Squared, a technology startup specializing in developing sophisticated algorithms, has developed an AI solution to boost productivity at apparel manufacturing plants, a world-first according to Google India Vice President Rajan Anandan. Anandan recently became the chairman at Linear Squared after the venture capital firm he co-founded, BOV Capital, invested in the startup after recognizing its potential to develop AI solutions for global markets out of Sri Lanka. “They have created the word’s first AI-driven factory planner. It will first be sold to large Sri Lankan companies like Brandix and MAS, and then they will scale all around the world,” he says.

Consumers are pickier than ever. Growing demand for personalized and unique products and offers over standardized ones, and  expectations for cheap speedy delivery, are forcing companies to adopt technology to optimize production efficiencies and gather, analyze and predict consumer trends.

Globally, manufacturers are turning to artificial intelligence (AI), machine learning (ML) and big data analytics to survive, and this is driving the Fourth Industrial Revolution. Few industries experience the full brunt of fickle consumer expectations than clothing  manufacturers who are constantly expected to deliver quality and speed at low margins. Sri Lanka’s apparel industry, dominated by family-owned manufacturers like Brandix, MAS and Hirdaramani, has a global reputation for doing just that. These companies produce high margin, specialized garments investing in R&D and consumer data. Lean management, sophisticated machinery and process innovations keep costs under control, giving them the flexibility to adapt to changing consumer tastes quickly. They’re so good at all this that some of these companies offer their management expertise to manufacturers in other countries as well.

Capacity Squared doesn’t just save time and minimize human errors, it’s creating additional capacity and revenue

While process efficiencies can always improve, companies like Brandix and MAS have achieved so much that it’s difficult to fathom what else they can do. While AI evokes notions of robotics and self-driving cars, it specifically refers to a computer system that can perform processes that typically require a substantial amount of human intelligence. It’s not about taking over repetitive tasks: AI empowered networks can learn from experience, gather data from multiple sources and analyze huge volumes of data, all these at high levels of accuracy and speed. However, finding a use for AI in the production floors of Sri Lanka’s accomplished apparel manufacturers was a challenge for Linear Squared.

“This is why it took us more than year to understand the apparel industry and develop an AI algorithm than can deliver a significant gain,” Dr. Sankha Muthu-Poruthotage, a co-founder of Linear Squared, says. The startup worked closely with one of the leading apparel  manufacturers to identify a problem that AI can fix. They found an opportunity in an area called capacity planning. “In fact, they led us to the problem and was a part of the solution all along,” according to Sankha.

Capacity planning in an apparel manufacturing plant requires significant amount of human skill and experience. An apparel manufacturing plant is arranged into several production lines or modules. However, these modules are not homogenous by nature. The skill level and experience vary among different modules. Each module has a different ‘efficiency ladder’, a concept which is related to the learning rate of each module. This learning rate can depend on the skill level of the module, as well as the kind of styles they have produced in the recent past. It is the planners who decide which style will be produced on which line, on which date, considering order quantities, raw material availability, delivery deadlines, learning rates, machine-related constraints and embellishments-related parameters.

“To come up with a capacity plan that satisfies all these conditions is a monumental task by itself. However, it does not guarantee that the entire plant is run at its maximum efficiency,” Sankha says. Linear Squared developed an AI algorithm called Capacity Squared to tackle this complex problem, and designed it to achieve optimal plant-level efficiency in under five minutes. “In the process, it eradicates human errors and biases, and cuts down the planning time by hours,” he says. Capacity Squared has an AI algorithm that is powered by a neural network. A neural network tries to mimic the learning process of a human brain, where it continuously learns with each new experience. At present, planners do this by their own experience. The neural network can replicate the thinking process of a highly skilled planner and consider a multitude of subtleties humans can’t possibly keep track of before deciding on the most appropriate plan. The system can be connected to an ERP system and other data sources, so planning is done at just a click of a button. This also means the plant can be more responsive to even minor changes in requirements, which would otherwise take hours to fix.

Capacity Squared is powered by a neural network which mimics the learning process of a human brain

“Our vision is to optimize the entire process, from sourcing to delivery. We think this is a good start since we believe that we have solved the most complex problem out of the lot,” Sankha says. Capacity Squared doesn’t just save time and minimize human errors, it’s creating additional capacity and revenue. The current estimate is around $25,000-30,000 a month for a mid-sized plant of 15-20 production lines. That’s over Rs50 million a year in additional revenue.

Linear Squared was incorporated in 2015 by Dr. Sankha Muthu-Poruthotage and Rajith Munasinghe, both with strong academic credentials in mathematics and statistics, having studied and worked overseas for several years. At the time, big data, AI and machine learning were already hot topics globally, but there was a gap they realized they could fill.

For big data, AI and machine learning to work, businesses had to have three elements in place: capabilities to develop algorithms, appropriate technology and a deep understanding of the industries they were in. “There is a reason why globally big data, AI and machine learning are not being used as widely as they ought to be: most businesses fail to understand these elements that needs to be in place,” Sankha says.

Companies may have the wherewithal to invest in the technology, but no expertise to optimize them. Even if they had data scientistst o develop algorithms, the effectiveness and actionability of their work depends on how deeply they understood the industry the business was in.

Linear Squared is building capabilities around these verticals. The co-founders are grounded in mathematics and statistics crucial for developing algorithms, and they initially built a team in that area. Next, they developed in-house capabilities around the technology and went on to expand capabilities in domains such as finance, FMCG, telco and apparel. The startup’s mission statement reads: ‘to provide end-to-end data-driven solutions to complex business problems, leveraging sophisticated algorithms, cutting-edge technology and domain expertise’.

The startup first focused on providing big data and machine learning solutions acquiring a customer portfolio that included heavy-weights in telco, IT, FMCG and apparel. Traditional analytics rely on what happened in the past. For example, businesses know who has bought their product, whereas sophisticated algorithms such as ML and AI tell businesses who will buy their product in the future.

“Businesses tend to group their customers into broad segments, each containing several thousand people. However, big data can create a segment of its own for each individual so businesses can make better decisions and strategize better. Our algorithms don’t compromise on accuracy, and we offer speed. A process that took weeks or even months to gather data, process and take action can now be done in real time,” says Rajith.

Big data, AI and machine learning don’t have to be out of reach of smaller companies. Businesses don’t have to make a big investment in infrastructure. The Linear Squared team insists that the starting point for any company is doing what they can with available resources and drive amazing results. There are only a few companies in Sri Lanka with big data. But the rest can work with their existing infrastructure, which Linear Squared recommends most of the time. Cloud-based solutions are available, scalable and you pay for what you use, which is a good option for any company.

The company is also developing an offering for FMCG businesses. Many of them forecast sales. However, this takes a very rudimentary shape, on an excel sheet, for example. Often, these forecasts are extremely inaccurate and unreliable for objective decision making since they do not factor complex and dynamic market factors and macroeconomic conditions. “Forecasts can be off by as much as 20-40%. The forecasts our algorithms make were tested and were 90-95% accurate,” says Rajith.

Traditional analytics rely on the past, whereas ML and AI make accurate predictions about the future

Linear Squared has set the entire process of forecasting into an algorithm. Typically, a data scientist needs weeks to analyze data before making an accurate forecast. Linear Squared has taken the data scientist out of the equation, for quicker, more accurate and scalable forecasts.

“The solutions we’re building to solve specific problems are unique in a global sense, such as Capacity Squared for which we’ve filed a patent. So, we expect global demand for our solutions. Right now, the strategy is to build the team and deepen our expertise to be on par with any global company in big data, AI and machine learning. In six to twelve months we will be ready to scale out of Sri Lanka. BOV Capital invested in us because they believe we can,” says Sankha.