In the AI Startup of the Week, the editorial staff of ai.nl is featuring promising AI startups, their innovations, solutions and challenges. In this twenty-first episode, we are taking a look at Cape Town-based AI startup Enlabeler. This South African startup is on a mission to merge the need for accurate video, text, image, and audio labelling with the urgent need for job creation.
Every company trying to become AI-fuelled is faced with usually the same challenge. While they are making rapid progress on AI skills, development, and even deployment, they are not making enough progress to completely uppend the workflow around labelling their data. Enlabeler is trying to solve this very challenge and is doing so with job creation in mind.
Building South Africa’s AI competitiveness
Enlabeler is not just building a startup that labels data for other companies. It is a startup with the grand ambition of turning South Africa’s economy into a competitive economy in the artificial intelligence and data industry. Every country is using AI to advance its economy and propel the future of its citizens by introducing jobs. Enlabeler seems to be aiming for a similar result.
Founded in 2019, the startup raised an undisclosed sum in the form of seed funding from E4EAfrica in 2020. The startup is operating in a relatively niche segment of artificial intelligence. It has reported serious growth in business over the last couple of years as companies race to label their data and push their progress on use of supervised learning.
For automation of business processes, companies need AI powered models and labelled data is crucial. Enlabeler says it aims to become the number one data labelling service provider for clients all around the world. It is doing that by creating a pan-African community of data labellers, analysts, and language specialists. While doing so, it is also creating AI-driven jobs supporting the African economy.
A competitive data labelling product
The rise of Enlabeler is on the back of a simple yet niche product that turns raw data into high-quality training datasets for your AI models. The startup says that eight out of ten AI projects fail due to lack of accurate training data. For startups as well as big tech companies, the access to accurate training data has become a clear challenge.
“Our platform turns raw, unlabelled data into high-quality training data. Our team of domain experts works with different data types, for a diverse range of industries. With this, Enlabeler creates flexible tech jobs and fights local unemployment across Africa,” says Esther Hoogstad, founder and chief executive officer of Enlabeler in an interview with Disrupt Africa.
The startup offers a suite of services that help companies create training datasets from unstructured image, video, audio, and text data. Some of the services offered by Enlabeler include image and video annotation models designed by Enlabeler that are used for computer vision models, transcriptions of audio files into text, translations of video and audio content to another language, and text classification and entity recognition to train models in the area of sentiment analysis.
Hoogstad explains that machine learning models and algorithms require big datasets to train the models. However, the data scientists or machine learning engineers “don’t have the time and capacity” to spend hours and hours creating, cleaning, and labelling datasets for their models. As a result, there is now an opportunity for a startup like Enlabeler to thrive by supporting companies looking for end to end data labelling solutions.
“Companies are looking for reliable and accurate data for their internal artificial intelligence (AI) and machine learning (ML) model,” she says.
Just like how speech to text services have emerged in this new Ai-driven economy, the data labelling industry should be seen as a close cousin. This industry also employs multiple data labellers, annotators and language specialists who painstakingly work on labelling the data.
“We now have a growing team of nine people and a database of over 350 data labellers, annotators and language specialists that often come from marginalised communities,” Hoogstad adds.
A mission to create jobs
While it is a known fact that every company wants access to accurate and labelled training data for their AI models, they don’t necessarily have the resources to do it themselves. Startups like Enlabeler are providing a valuable service in an industry where AI is making an impact but often held back by data related problems. However, this African startup is not without competition and to stay ahead, it is not only building a business model but doing so by creating jobs in Africa.
“Global competitors in the data labelling and annotation space are Sama, Labelbox, Labelfuse, Scale AI, and a few others. However, none of these are based in Africa, and none share the same mission to create and build datasets in Africa for domestic and international clients,” Hoogstad explains.
She adds, “Ultimately, it’s about empowering a whole new generation of professionals in the data industry that will gain experiences in the growing AI and ML space. Because of Enlabeler’s price point, customised service offering and quick turnaround times, we are able to compete with some of the more automated, large players based in the US.”
The growth of Enlabeler can be owed to the fact that it charges clients per dataset or per unit of labelling. The clients of Enlabeler also pay only for the output that meets their agreed quality standard. With additional services like data pipeline integration, building of customised APIs, dataset creation and cleaning, the startup ensures it offers an end-to-end business solution.
Hoogstad explains that the company created approximately 45 labelling jobs in 2020 but surpassed that number by more than 25 per cent in 2021. It has a mix of clients with 70 per cent based in South Africa, and 30 per cent being international. It has a large clientele of AI and ML companies but also big infrastructure players based out of the Netherlands, the US, and Canada.
With AI development accelerating and companies requiring big structured and clean datasets, Enlabeler is poised for success. As a fully remote startup, it is not impacted by the pandemic or associated lockdown. For Hoogstad and Enlabeler, the real challenge will come when there are more startups offering labelling as a service and offering at a much more competitive price. At that point, Enlabeler is aiming to be completely embedded in the South African economy and it seems likely to achieve that goal.