Quality Match raises $6 million to build better AI datasets

Quality Match, a Heidelberg, Germany-based quality data annotation provider, today announced that it raised a €5 million ($6 million) seed round from LEA Partners. The company says it’ll use the proceeds to expand its team and accelerate product development.

Training AI and machine learning algorithms requires plenty of annotated data. But data rarely comes with annotations. The bulk of the work often falls to human labelers, whose efforts tend to be expensive, imperfect, and slow. It’s estimated most enterprises that adopt machine learning spend over 80% of their time on data labeling and management. In fact, in a recent survey conducted by startup CloudFlower, data scientists said that they spend 60% of the time just organizing and cleaning data compared with 4% on refining algorithms.

Quality Match, which was bootstrapped in 2019 by a team of former Pallas Ludens, Apple, Google, Microsoft engineers, aims to improve the speed and quality of data labeling processes by disambiguating the potential sources of error. The platform explains the sources of errors in datasets, highlighting where edge cases originate and providing strategies on how to improve the data.

There’s no shortage of data labeling startups competing with Quality Match — the market was valued at $1.3 billion in 2020, according to Grand View Research. For instance, Scale AI has raised over $100 million for its suite of data annotation services. There’s also CloudFactory, which says it offers labelers growth opportunities and “metric-driven” bonuses. Hive, Alegion, Appen, SuperAnnotate, Dataloop, Cognizant, and Labelbox are other rivals of note.

But Quality Match uniquely begins building or enhancing datasets by optimizing the representativeness and diversity of the samples, ensuring they’re representative of the real world and contain difficult edge cases sprinkled throughout. Then, the platform exposes and quantifies ambiguity in the datasets before breaking the taxonomies into small, intuitive questions that form a fully automated decision tree. Quality Match runs multiple repeats of this decision tree to provide confidence scores on all of the annotations.

Moreover, Quality Match provides metrics including geometric, label, and definition accuracy that are intended to inform about wrong tags or attributes as well as missed or spurious detections of annotations. It also shows how factors like taxonomy version changes over time and varying criteria for quality scoring might be contributing to imbalances in the datasets.

“During the pandemic, our industrial customers, in particular, have increasingly realized that they will have to rely more on high-tech solutions in the future because, in these times, large groups of people can no longer work together in one room,” said cofounder and managing director Daniel Kondermann, who told VentureBeat that the goal this year is to reach €1 million in revenues. “To be successful, companies must adapt, which leads to an increasing demand for automation and therefore AI across a wide range of industries. Quality Match also started entering the market of medical technology. An industry that got even stronger due to the pandemic and therefore continued to develop new and improve existing AIs. All these industries are asking for our datasets and profiting from our work which is why so far, we have managed this pandemic very well.”

Twenty-employee Quality Match, which counts among its customers Mapillary, Bosch, and other companies engaged in health, 3D maps, autonomous driving, AR/VR, retail, and construction, received the whole of its latest funding from LEA Partners. Kondermann says that the immediate focus will be on hiring talent.