Everyone remembers Supahands as a company that would send out agents to help you with your errands or repetitive jobs, but it has been quite a while since the company pivoted from that to the much more lucrative task of data labelling to help companies train their machine learning systems.

As its chief executive officer (CEO) and co-founder Mark Koh tells Enterprise in an interview, while running the business, the company kept abreast of the latest technological trends and witnessed the rise in importance of artificial intelligence (AI) and machine learning. It quickly realised that in the machine learning process, the labelling of training data takes up a lot of the data scientist’s time.

“Labellling data for the purpose of training a machine learning model is a highly human-intensive task that also requires a level of skill, in addition to requiring the labeller to potentially go through thousands of images and labelling them individually.

“Considering that we had a few thousands SupaAgents already with us at the time, as well as an in-house platform that allowed us to manage them easily, it was just a great fit for our existing business model,” says Koh.

Basically, he says, Supahands accelerates the development and deployment of AI for its technology clients across the globe. “This is a growing industry and we still see a good response from them.”

Koh adds that because Supahands takes care of data labelling, its clients can focus on other aspects of their operations without having to worry about the quality of their training data, as well as the management of the data labelling process. 

“Having high-quality data plays a key role in the accuracy of a finished machine learning model. And as data labelling is such an important but time-consuming aspect of the ML (machine learning) life cycle, we support our clients by taking care of the entire process. 

“We consult our clients, determine their data labelling needs, put in place labelling guidelines, train our SupaAgents and ensure the data is labelled at high quality, managing the data labelling aspect end to end with speed, scale and accuracy,” he says.

Has it been affected by the movement control order (MCO)? Not as much, says Koh. “Yes, the response rate has been slower than usual with businesses being careful with their cash flow. But we have still been able to bring in new projects, renew contracts and acquire new clients during the MCO period.”

Koh says even now, there is definitely a market for its data labelling solution and untapped opportunities which the company is now “exploring aggressively”. “The AI market is continuously growing and, if anything, there’s more need for automation than ever before so the need for high-quality data to train machine learning models is not going away.”

He says the company just needs to identify the new demands quickly and adapt accordingly. “I think this goes for most companies in our position.”

He adds that the company never really saw this move as a pivot. “It felt like a natural evolution of our capabilities. We started with very simple data entry work to now being able to handle millions of data labelling tasks in a month. We just grew up!”

Koh points out that Supahands was born out of the idea that you could be more efficient if you could have other people support you by doing the more repetitive tasks in your life. “We started recruiting our remote team of SupaAgents to complete a variety of tasks for individuals from restaurant reservations to basic research.

“Then we started seeing businesses utilising our service for repetitive and manual tasks so that they could focus on the more strategic aspects of their organisation,” he says, adding that soon its corporate business outgrew its individual errand-servicing one.

But that was not the only change. The company soon realised that while the concept of a fully remote team working with it on a contractual basis had a lot of benefits, it was difficult to grow quickly when the tasks performed were so diverse. “It became difficult to train SupaAgents at scale, which impacted the speed of our output and quality control.”

Koh says Supahands then focused heavily on building the tools and platforms that would not only allow it to execute work specific to the AI/machine learning industry, but also perform the necessary quality checks.

“For example, SupaAnnotator was built to perform image annotation tasks, the output of which was used to train computer vision models. We also built SupaTutorial, which allowed our team of business operation managers to train SupaAgents at scale without manually screening through every individual on a project,” he says.

Koh says Supahands' promises to its clients are speed, scale and accuracy. “The whole journey has been about how we can deliver those promises with the powerful combination of technology and humans.”

He says demand for AI was a huge window of opportunity for the company. In addition, the rise of the gig economy and working from home or remotely also played a huge part in its growth, enabling it to recruit SupaAgents quickly from all over the world to support demand for its services.

“We now have over 13,000 SupaAgents, more than triple the number of agents we had just a year ago, and they themselves are a great community and an extension of the SupaTeam that powers our business,” he says.

Most of its agents have been working remotely from the beginning, so it was business as usual when the MCO was announced as far as their work in Supahands’ projects was concerned.

“Our in-house team here in Kuala Lumpur, on the other hand, started working from home a few days before the MCO. We had to quickly adapt some processes to make them remote-work-friendly, but otherwise the transition went quite smoothly.”

How has morale been since the MCO? Not bad, he says. “We try to keep our internal communications honest and transparent, and I think that has helped keep the team driven and motivated. 

“Business has not been easy. Our potential clients are based in affected countries, such as the US, but there are still opportunities to grow Supahands. Our team has been able to adapt to these challenges and has been more productive than ever.”

Other small and medium enterprises (SMEs), however, have been more adversely affected. Does Supahands have any solutions for them? Yes, says Koh. “We service a variety of retail sectors, including both offline and online players in the market. With e-commerce clients, we work to improve the accuracy of the product search function on their websites by ensuring that all the products that have been uploaded are tagged with the right sets of labels.”

For instance, he says, if a shopper types in “purple watch” into the search bar, what should come up are, well, purple watches, not listings of toothbrushes, make-up and toys.

Koh says Supahands supports SMEs by enabling them to refine their product offerings, so that they can drive better sales on their end. “We frequently encounter clients who may not know where to start when it comes to data labelling. We help by being really hands-on with their teams, consulting them on the best process, putting guidelines in place, training the labellers and providing high-quality data that they can actually use — all performed via our technology infrastructure.”

He adds that Suphands sees itself as a partner of businesses, rather than a vendor. “Our whole approach is different. We don’t sell; we help.”