Data annotation platform provider Encord, officially known as Cord Technologies Inc., said today it has closed a $30 million Series B investment led by Next47, with participation from previous backers Y Combinator, CRV, and Crane Venture Partners.
The startup has created what it likes to call a "data development platform" that's designed to make artificial intelligence models easier to train. It explains that the effectiveness of any AI model is directly tied to the data that it was trained on, and this means that data preparation and annotation are critical tasks for AI developers.
The challenge is that teams are forced to deal with tons of siloed, uncurated and unformatted information. Traditionally, data preparation and annotation has always been done manually, creating a major bottleneck for AI development.
While automated data annotation systems are not new, traditional ones have relied heavily on human supervision. Encord doesn't do this, instead automating the entire process by using AI itself to supervise it, helping teams to get large datasets ready for AI training much faster than was possible before.
Encord says its software addresses the four key steps involved in creating AI training datasets, encompassing data management, data curation, model evaluation and annotation. By consolidating these tasks into one platform, it creates a trail that developers can use to dig into their AI models and understand why they generate specific outputs. They can then work out how to improve their models and retrain them for better outcomes.
The startup is trying to compete in a data annotation and labeling industry that's expected to grow to more than $3.6 billion a year by 2027, and it faces a lot of competition. Its biggest rival is probably Scale AI Inc., and there are plenty of other startups pushing for a piece of that pie, such as Datasaur Inc., Dataloop Inc. and Soda Data NV.
However, Eric Landau, co-founder and Chief Executive of Encord, told TechCrunch that his company's platform is more versatile than those of its rivals. The advantage, he said, is that it's uniquely able to help AI developers explore and visualize their training datasets, no matter if they contain images, text, video or voice recordings. It also provides tools for comparing the output and performance of different models trained on the same datasets and detecting model accuracy issues. In addition, it can even suggest what kinds of additional training data should be added to a set to help fix any issues it identifies.
"Encord lets you consolidate all your data workflows in one platform," Landau explained. "Through this consolidation, companies gain traceability that sheds light on the often opaque 'black box' of AI, helping to understand why a model makes specific decisions."
Of course, its only to be expected that Encord believes its data development platform is the best, but there's no denying that its product has been well received. It counts more than 120 signed up customers so far, including organizations such as Synthesia Ltd., Koninklijke Philips N.V., Zoopla Ltd., Cedars-Sinai Medical Center and Northwell Health LLC.
T.J. Rylander, general partner at Next47, said Encord grabbed his attention because he has seen tons of innovation at the AI model and compute layers, but little progress in terms of the data layer. "Encord's forward-thinking platform addresses one of the biggest challenges in AI today - understanding and managing the data that will give enterprises high quality, reliable outcomes from their AI applications, therefore lowering the risk of AI implementation," he said.
