LXT, a provider of industry-leading AI data solutions, recently announced that it has completed the technology integration of Clickworker to deliver a single platform for comprehensive, high-quality AI data solutions to power generative and other AI applications. Clickworker is a global crowdsourcing platform that enables freelancers to perform microtasks in areas such as writing, translating, research, and data processing. With over 8 million active users, it offers flexible, scalable solutions for AI model training.
Clickworker has become a key infrastructure provider for AI model training, offering labelled text, images, speech, and sentiment data at scale. Could you walk us through how Clickworker operates, the markets you serve, and the clients you cater to?
Clickworker, founded 20 years ago in Germany, started as a crowdsourcing platform. We built our own technology and built a scalable platform to access the paid crowdsourcing market. Today, we have more than 8 million workers globally. Initially, we focused on text writing through desktop environments, but five years ago, the rise of AI prompted us to diversify.
We launched the Clickworker app, which made it possible to engage freelance workers in regions like Africa and Asia, where access to personal computers is limited. This expansion has led to 200,000 new users joining each month. For many workers in developed countries, it is side-income, but in lower-wage countries, it is a significant source. Workers can choose their hours and select projects, with most tasks revolving around data collection and annotation—audio recordings for speech recognition training, or selfie photos and videos for AI training.
What types of clients purchase data in large volumes from Clickworker?
The majority of our clients are based in Silicon Valley, with companies focused on AI and improving speech recognition or training large language models (LLMs). Additionally, businesses in product development use our services to gather real-world data from store visits or supermarkets. While most of our clients come from the high-tech sector, we are seeing an increase in demand from Europe as AI investment there continues to rise.
How has the demand for your services evolved with the rise of AI?
Data is critical for AI companies. As AI technology advances, the need for diverse, high-quality data grows. Initially, data collection focused on basic tasks like training computer vision and voice recognition systems. We collected selfies for aging detection or object identification, such as recognizing animals or cars.
We have also seen a trend towards more complex scenarios, such as training surveillance cameras for security applications or creating video content with interactions between people. But with LLMs, the demand has shifted. While foundational knowledge is built from internet data, there is now a need for high-quality data to train edge cases, improve and validate industry knowledge of models.
How do regional differences in digital access and tech education impact your ability to scale high-quality data annotation?
We have implemented several measures to ensure scalability across our global workforce. Since we do not meet workers in person, we have developed workflows to quality-assure every piece of work submitted. We use trusted workers or agencies to verify the data. Additionally, we have invested in fraud management to identify fake devices or duplicate user accounts. We assess language skills and industry expertise to ensure that only qualified workers are engaged for specific tasks. By segmenting workers based on demographics and skills, we can assign the right workers to the right projects.
There is growing demand for these workers, but the skill requirements are evolving.
As AI becomes more embedded in different industries, domain-specific expertise is becoming essential.
For example, we are focusing on building a crowd with skills in specialized areas like manufacturing, legal, and software development. While the amount of work is increasing, the key challenge is whether it will be suitable for general workers or more specialized roles. This is something we are continuing to explore.
What is the average age of your Clickworkers, and does it vary across different regions?
The average age of a clickworker is about 35, which is older than the 18-25 range we originally saw. While we still recruit younger workers, we have noticed that crowdsourcing is attracting a broader demographic. However, we still face challenges in reaching older age groups, which can affect certain projects. Balancing demographics is crucial for ensuring unbiased data collection.
With millions of freelance clickworkers, gig labor ethics has been a hot topic, particularly in relation to low pay and limited protections. How does Clickworker address these concerns?
Clickworker has formed a working group with other platforms and unions in Germany to tackle these issues. We ensure workers receive at least the legal minimum wage, especially in low-wage countries. Fraud management is critical to prevent workers from accessing higher-paid work inappropriately. We offer 24/7 support to ensure workers can perform their tasks without feeling isolated. We also have an arbitration board where workers can report any unfair treatment. Fair treatment is directly linked to the quality of work, as better treatment results in better data.
In an ecosystem increasingly dominated by big tech platforms, Clickworker has emphasized that data and model interoperability can level the AI field for mid-sized players. Can you tell us a bit more about why that is important?
The merger with LXT exemplifies why this move made sense for Clickworker. It expanded our capacity and reach in a market where competition is often well-funded and fast-moving. Joining forces with LXT allowed us to focus on secure data annotation, offering services for highly confidential data, such as in the medical or legal sectors. Our combined capabilities enable us to offer managed, secure or crowd-based solutions as well as more complex, multimodal data (video, audio, text) and specialized training for edge cases and domain expertise.
Synthetic data and strict AI regulations are emerging as major forces reshaping the AI business. How do you expect these to impact your business in the next few years, and what other trends are on your radar?
Synthetic data is a key trend, allowing us to multiply real-world data and simulate scenarios that are hard to replicate in real life, like dangerous situations. This brings cost efficiencies and increases data volume. Additionally, we see a growing demand for human involvement in fine-tuning AI models, particularly large language models (LLMs), due to concerns over hallucinations and compliance issues. As enterprises deploy their own LLMs, we expect more companies to turn to human-powered data quality assurance.
Two major trends are currently driving AI disruption. First, robotics is evolving as the real world is being simulated to improve robotic performance. Second, agentic AI, where AI can make decisions and take actions autonomously, is gaining attention. Alongside reinforcement learning, these trends are fundamentally reshaping the industry.
