AI data services company Appen is accelerating its transition from traditional data annotation to large model data services.
Starting from the 2023 annual report, Appen has consolidated projects such as LLM training data, multimodal generative AI, video generation data, and model safety and evaluation into its generative AI-related business (hereinafter referred to as GenAI business). It has officially established this as a strategic business category and explicitly proposed the transition to GenAI.
According to financial reports, the proportion of Appen's GenAI business revenue increased from 4% in 2023 to 33% in 2025, reaching 44.1% in Q4 2025; in H1 2026, it was 29.0%, an increase of 4.4 percentage points year-over-year. During the same period, the group's revenue grew by 17.5% year-over-year, and the adjusted EBITDA returned to profitability.
From business structure to operating performance, Appen's GenAI business transformation has begun to show results. However, further breakdown reveals significant differences in performance between the China business and the overseas business.
In H1 2026, the Appen China segment (covering China, Japan, and South Korea, hereinafter referred to as the China business) contributed USD 76.2 million in revenue, accounting for 63.6% of the group's revenue, with an adjusted EBITDA of USD 12.1 million; during the same period, the Appen Global segment (primarily targeting the US and European markets, hereinafter referred to as the overseas business) generated USD 43.7 million in revenue, with an adjusted EBITDA loss of USD 4.5 million.
The China business is the most positive part of Appen's current operating performance. In H1 2026, revenue from the China business grew by 80.4% year-over-year, and adjusted EBITDA increased by 315.5% year-over-year. The profit margin improved from 6.9% to 15.9%, achieving simultaneous growth in revenue and profitability.
According to disclosures from the earnings call, the growth drivers for the China business include the expansion of GenAI projects, Chinese tech companies expanding overseas, growing demand for consumer-grade AI applications, and increased revenue from pre-built datasets. In contrast, although the overseas business has secured large model-related projects in coding, finance, and healthcare, the new business has not yet fully offset the contraction of traditional business.
Therefore, the focus of this research on Appen is twofold: first, the core drivers behind the high growth and profitability improvement of the China business; second, the main challenges faced by the overseas business in its transition to GenAI. Based on this, we evaluate the overall business adjustment progress of the Appen Group.
01 GenAI Business Revenue Reaches 33% of Total, Business Adjustment Makes Progress
Appen's business transformation is highly synchronized with the changes in AI market demand.
During the traditional machine learning stage, Appen primarily provided data collection, annotation, and evaluation services for scenarios such as search, voice, and computer vision. In 2023, traditional model testing and search relevance business accounted for 75.9% of Appen's revenue. Among them, a single customer, Google, contributed USD 82.8 million, accounting for 30.3% of the group's revenue. At that time, traditional AI data services were its main source of revenue.
Given the rapid development of large models, customers' data needs and budget priorities have begun to shift. On the one hand, traditional annotation demand continues to shrink; in 2023, Appen's revenue dropped by 29.7% year-over-year, and the termination of cooperation with Google in 2024 further impacted the traditional business. On the other hand, large model training, alignment, and evaluation have started to generate new data demands, and Appen's business focus has gradually shifted to the GenAI business.
This change has been clearly reflected in the revenue structure. In 2023, the proportion of GenAI business-related revenue was only 4%; in 2024, it accounted for 22% of revenue excluding the impact of Google; and in 2025, it further rose to 33%. Meanwhile, service content has gradually expanded from image bounding, speech transcription, and search judgment to preference data, code and math reasoning, model evaluation, reinforcement learning environments, agent operation trajectories, and robot training data.
Table 1: Appen GenAI Business Growth and Group Performance Changes
In 2025, Appen's revenue still declined by 1.5% year-over-year, but it has resumed growth after excluding the impact of the termination of the Google contract; in H1 2026, the group's operating revenue grew by 17.5% year-over-year. The GenAI business has become an important increment to make up for the revenue gap in the traditional business. However, the group's growth is mainly contributed by the China business, while the overseas business revenue still declined year-over-year. The transformation progress in the two regions shows significant differences.
02 China Business Grows and Profitability Improves, Transformation Results Begin to Emerge
In terms of revenue growth, growth drivers, and profitability performance, GenAI-related demand has become an important growth driver for the China business. In 2025, the China business revenue reached USD 102.9 million, a year-over-year increase of 74.8%, with an adjusted EBITDA of USD 10.6 million.
In H1 2026, the China business revenue further increased to USD 76.2 million, a year-over-year growth of 80.4%; the adjusted EBITDA increased from USD 2.9 million in the same period last year to USD 12.1 million, a year-over-year increase of 315.5%, and the adjusted EBITDA margin improved from 6.9% to 15.9%.
Table 2: China Business Becomes the Main Source of Appen's Revenue Growth and Profit
The high growth of the China business mainly comes from large model-related demand. According to the H1 2026 earnings call, Appen stated that the improvement in the China business profit margin mainly comes from three aspects: an increased proportion of generative AI projects, revenue growth from pre-built datasets, and improved operational efficiency brought by revenue expansion.
From the perspective of the business model, the transformation of Appen's China business is mainly reflected in the change of service content, rather than the reconstruction of the charging model. According to research by iFenxi, some projects are still settled based on personnel scale or service cycle. Appen organizes experts and delivery teams to continuously undertake model training, evaluation, and iteration tasks. This model can adapt to the characteristics of continuously changing customer needs for large models, and also helps maintain personnel utilization and project profitability.
Based on the comprehensive analysis of revenue sources and operating performance, the China business has made relatively obvious transformation progress. GenAI-related projects have become the main business increment, driving simultaneous growth in revenue and profit. However, its revenue still mainly comes from project delivery and personnel services. Although standardized products such as pre-built datasets have grown, they are still in the expansion stage. In contrast, the transformation progress of the overseas business is significantly slower than that of the China business.
In H1 2026, the overseas business revenue declined by 26.9% year-over-year, and the adjusted EBITDA remained at a loss. Although Appen has entered professional data fields such as coding, finance, and healthcare, and secured new large model-related projects, traditional search and content-related businesses are still shrinking, and the new business has not yet fully offset the decline in the original business. It can be seen from this that the transformation results in different regions of Appen have shown significant divergence: the China business has achieved revenue growth and profitability improvement through the GenAI business; although the overseas business has completed the adjustment of business direction, it is still in the stage of switching between old and new businesses.
03 Expert Data Has Higher Value, but Profit Margin May Not Necessarily Increase
After large model data services enter professional fields such as coding, finance, and healthcare, although the value of a single task has increased, the profit margin may not necessarily increase synchronously. Appen's management stated in the H1 2026 earnings call that the gross profit margins of high-value projects such as coding, finance, and healthcare are generally similar to those of traditional projects. This indicates that expert data raises customers' requirements for data quality and professional capabilities, and correspondingly increases the costs of expert organization, task design, and quality control. The cost structure of expert data is significantly different from that of traditional tasks.
Traditional tasks such as image bounding, speech transcription, and search relevance judgment can be standardized and decomposed according to workload and accuracy; expert data in coding, math, finance, and healthcare require the participation of personnel with professional knowledge, and task rules need to be continuously adjusted according to model capabilities and training objectives.
Project delivery not only includes data production but also involves expert certification, question design, answer calibration, model testing, and multiple rounds of iteration. Therefore, for AI data service providers, what determines the profit margin is not just the unit price of data, but whether they can improve personnel utilization and unit delivery efficiency through expert resource management, automation tools, project pricing, and delivery processes.
Figure 1: Expert Data Increases Project Value but Also Raises Delivery Costs
According to research by iFenxi, the focus of competition in AI data services is shifting from personnel quantity to expert capabilities, task understanding, and high-quality delivery. Long-term large customers, stable expert resources, and a mature delivery system remain important foundations.
Appen is improving the productization level of its business through pre-built datasets. In H1 2026, Appen expanded datasets such as reinforcement learning tasks and verifiers, code repositories, enterprise agent training data, medical data, and STEM Q&A. Some datasets can be sold directly or combined with Managed Services to complete customer customization.
The "pre-built datasets + customized services" model helps reduce the cost of repeatedly organizing and producing data for some projects, but this model is still in the development stage. Currently, Appen's large model-related revenue still mainly comes from customized projects and Managed Services, and pre-built datasets are more of a supplement to the existing service model.
04 In the GenAI Data Services Market, the Key to Competition is the Share of Top Customers
The performance differences between Appen's China business and overseas business come not only from operational efficiency but also reflect the differences in customer demand, existing business structure, and project delivery models between the two markets. The outstanding performance of the China business is firstly benefited from the rapid growth of large model data demand. Chinese foundation model providers, tech companies, and AI application companies are still in the stage of model capability improvement and application exploration, with continuously increasing demand for training, alignment, evaluation, and professional data. Meanwhile, Chinese tech companies expanding into overseas markets further drive the demand for multilingual data, model evaluation, and localized data.
Appen disclosed that its China business growth is mainly driven by new and expanded large model projects, indicating that Chinese customers are investing more budgets into GenAI data services.
Secondly, Appen's China business has established a delivery model adapted to market demand. According to research by iFenxi, some customers do not purchase datasets at once, but rather procure experts and delivery teams that can continuously undertake model training and evaluation tasks. Settling based on personnel scale or service cycle allows flexible adjustment of tasks according to the model iteration progress, which not only meets customers' continuously changing data needs but also improves the personnel utilization of the delivery team.
Thirdly, stable large customer relationships and expert organization capabilities are also important foundations for the China business to quickly meet demand. Large model data projects usually require organizing personnel with professional capabilities in coding, math, finance, and healthcare within a short time, and repeatedly adjusting tasks according to model performance.
Appen's China business has a relatively mature project management and delivery system, which can quickly convert new demand into revenue. Pre-built datasets further supplement customized projects and improve the utilization efficiency of existing data resources.
Figure 2: Divergence in Transformation Results Between Appen's China and Overseas Businesses
The situation faced by the overseas business is different. The key to the slower transformation of the overseas business lies in the fact that Appen has not yet secured a sufficiently large core share in the data procurement of leading model providers in the US. The main budgets in the GenAI data services market are concentrated in leading model providers such as OpenAI, Anthropic, and Google. These customers have large individual procurement scales and can also drive data service providers to achieve sustained growth.
A new generation of data service providers represented by Scale AI and Surge AI entered the R&D processes of leading model providers earlier, participating in core links such as model training, alignment, and evaluation, and gradually established direct cooperative relationships with researchers. Appen's management also stated at the H1 2026 earnings call that some fast-growing competitors have deeply entered the customers' internal operations, gained the trust of researchers, and customers can procure directly from them without going through the complete bidding process. Therefore, the core of competition in the overseas market is not price.
Appen's management stated that when leading AI labs choose data service providers, they pay more attention to data quality, delivery speed, and trust in the supplier. Once a supplier enters the stable cooperation system of researchers, it is easier to obtain subsequent projects and expand procurement, which also increases the difficulty for latecomers to penetrate core projects.
In addition, the China business and the overseas business have relatively independent customer, data, and operational systems. The customer relationships and data resources accumulated by the China business are difficult to directly replicate in the overseas market, and the overseas team still needs to independently establish cooperative relationships with leading model providers. Therefore, the transformation experience of the China business cannot directly drive the overseas business to complete its transformation.
Appen's case illustrates that transitioning from traditional data annotation to large model data services, securing new projects is only the first step. What truly determines the effectiveness of the transformation is whether it can enter the core R&D processes of leading model providers and convert pilot projects into sustained large-scale procurement. The China business has already demonstrated operational improvements brought by the matching of demand growth and delivery capabilities; the overseas business, however, still needs to break through the core customer share to complete the substitution of traditional business.