Alegion, an Austin, Texas-based provider of labeling and annotation services for enterprise data science teams, today announced that it has secured $12 million in series A funding contributed by RHS Investments. Founder and CEO Nathaniel Gates said the fresh capital will accelerate both the expansion of its 75-person team and the integration of new technologies into its product suite.
“Just as assembly lines incorporate power tools and robotics to enable scale, [machine learning] model development will require machines training machines to achieve the highest levels of model confidence,” said Gates, who founded Alegion in 2012. “Our customers can first leverage human judgement to train their model, and then watch as newly trained machines are incorporated that allow unprecedented scaling.”
Alegion offers data labeling services powered by a mix of automated systems and human workers, tailored for tasks such as AI model training, testing, and exception handling in domains like computer vision, natural language processing, and entity resolution. (Think object detection, meaning extraction, and sentiment analysis.) It assembles and vets a workforce from a global pool to satisfy any number of project requirements, and the team then uses tools to enrich and compile data into large-scale corpora on which AI models train. Alegion’s validation suite evaluates model accuracy at every training stage, and post-deployment, the company employs human-assisted exception handling of edge cases to further improve performance.
Humans only enter the loop when necessary, Alegion stresses — machine learning algorithms handle the bulk of conditional logic, multi-stage workflows, and quality control routing. And when humans are tapped for complex or subjective judgments beyond the capabilities of AI, configurable policies ensure data is only made accessible to prequalified teams via signed URLs.
Alegion’s crowdworkers use task-specific workflows that break up chores like segmenting images, extracting inferences from transcripts, identifying overlapping objects in videos, and more into digestible microtasks. A range of annotation tools are at the team’s disposable, including bounding boxes, key points, polygons, and splines, as well as text-based classification functions for NLP and named entity recognition.
Adaptive quality controls support best-practice approaches like judgment consensus and domain-expert reviews, ensuring a baseline level of accuracy. Moreover, users can see all the data related to their jobs, enabling them to run their own process audits and analyses, and they’re able to interact with workers and provide bonuses to top performers.
Alegion claims this approach results in data sets with 99% accuracy compared with an industry average of 40-60%, potentially reducing training time and project development. Investors like RHS’ Hank Seale attribute Alegion’s success to its focus on precision: The company currently serves Fortune 500 customers across verticals such as retail, finance, technology, automotive, defense, agriculture, and health care — including Airbnb, Microsoft, the Home Depot, and Walmart.
“Artificial Intelligence’s insatiable demand for accurate training data can’t be provided through human power alone,” said Seale. “Alegion’s ability to supplement human effort with machine learning is strongly differentiating.”
The company competes with a number of heavy hitters in a data annotation tools market expected to be worth $1.6 billion by 2025, according to Grand View Research. There’s Mighty AI, Hive, Appen, Cloud Factory, Samasource, and Amazon’s Mechanical Turk, to name a few, plus Scale AI, which raised $100 million for its data labeling services this week.