Recent advances in generative AI have accelerated its adoption across industries, prompting organizations to explore new ways to enhance productivity and operational efficiency. At the same time, an equally important question has emerged: in the highly regulated and risk-sensitive biopharmaceutical sector—particularly within GxP environments where the principle of "document what you do and do what you document" is fundamental—does AI represent a transformative opportunity, or does it introduce new compliance risks that must be carefully managed?
At Asymchem Clinical (Clin-Nov), the answer is becoming increasingly clear. According to Zhexin Liu, Executive Director of Digital Innovation, the key lies in treating compliance not as a constraint to AI adoption, but as a foundational principle for innovation. By strategically combining low-code platforms with large language models (LLMs), Asymchem Clinical has been exploring practical approaches to deploying AI within a stringent GxP framework. These efforts demonstrate how AI can evolve beyond a simple productivity tool to become a catalyst for business innovation and operational enablement.
Drawing on Liu's practical experience in digital transformation and AI implementation, this article explores how AI applications can be rapidly deployed within regulated environments, how the value of historical data can be unlocked, and how frontline teams can be empowered through digital innovation—all while preserving the integrity, traceability, and compliance standards required by GxP-regulated operations.
Figure: Zhexin Liu
Executive Director of Digital Innovation, Asymchem Clinical (Clin-Nov)
1. Strategic Positioning: AI as an Enabler, Not a Disruptor
AI adoption in the enterprise is often associated with sweeping process transformations, workforce reductions, or the replacement of established ways of working. At Asymchem Clinical (Clin-Nov), however, AI serves a fundamentally different purpose within the organization's digital strategy.
"Rather than being viewed as a disruptive force, AI is positioned primarily as a business enabler and an assistive tool," said Zhexin Liu, Executive Director of Digital Innovation. In his view, the greatest challenge to AI adoption is rarely the technology itself, but user behavior and established work practices. "Most users are not looking to fundamentally change how they work. The role of AI is therefore to support existing workflows, enhance productivity, and simplify routine tasks—not to force users into entirely new ways of operating."
This philosophy has shaped a distinctly human-centered approach to AI implementation at Asymchem Clinical. Whether preserving familiar, Word-like interfaces for scientists or using conversational AI to guide less experienced employees through document and template selection, the underlying objective remains the same: adapting technology to users rather than requiring users to adapt to technology.
The same principle applies when evaluating the value of AI initiatives. In highly regulated GxP environments, accelerating timelines and mitigating risk are often perceived as competing priorities. Liu argues that they should instead be pursued in parallel, as speed and quality are inherently interconnected.
"AI enables us to generate clinical documentation more efficiently and automate the completion of Site Start-Up (SSU) forms, significantly reducing execution timelines," Liu explained. "At the same time, embedded tools such as our regulatory SOP Bot can automatically assess user inputs against predefined compliance requirements, helping reduce errors and compliance risks, particularly among less experienced team members."
Rather than forcing a trade-off between efficiency and compliance, this approach leverages AI to strengthen both. By embedding intelligence directly into operational workflows, AI functions not only as an accelerator of productivity, but also as a safeguard for quality and regulatory compliance.
2. Technology Strategy: Why Low-Code Platforms and Large Language Models Are a Powerful Combination
Once the strategic direction was established, the next challenge was selecting the right technology approach. Rather than investing heavily in off-the-shelf AI solutions or pursuing a fully custom-built platform from the ground up, Asymchem Clinical (Clin-Nov) adopted a different path—integrating large language models (LLMs) with applications built on Microsoft Power Platform.
According to Zhexin Liu, Executive Director of Digital Innovation, this decision was driven by four key considerations: security, agility, cost efficiency, and compliance.
Security came first. For a CRO organization handling large volumes of sensitive clinical and operational data, safeguarding information assets is a fundamental requirement rather than a technical preference. "Business stakeholders place significant emphasis on data security, particularly when cloud-based technologies are involved," Liu noted. "Established technology providers such as Microsoft offer a level of assurance and governance that is especially important in highly regulated GxP environments."
Agility was another critical factor. As business requirements continue to evolve, the ability to rapidly translate ideas into working solutions has become a competitive advantage. "Innovation cycles are becoming increasingly compressed. If we cannot respond quickly enough to emerging needs, we risk falling behind," Liu explained. By leveraging low-code development capabilities, applications can be assembled and deployed rapidly, enabling teams to address operational challenges without lengthy development cycles.
Cost effectiveness also played an important role in the decision-making process. Compared with fully custom-built systems, which often require substantial investments of time, resources, and specialized talent, low-code platforms significantly reduce development overhead while accelerating time to value.
Perhaps most importantly, the approach provides greater flexibility in balancing innovation with compliance. Traditional enterprise SaaS solutions are often built around standardized functionality supplemented by customized extensions. While effective in many scenarios, this model can become restrictive in a rapidly evolving AI landscape, where technological capabilities, use cases, and regulatory expectations continue to change at an unprecedented pace.
"As AI technologies evolve, organizations need practical and adaptable solutions that can address business needs quickly," Liu said. "It is important not to become constrained by traditional software architectures when more direct and efficient approaches are available."
This flexibility is particularly valuable in GxP-regulated environments, where software validation and qualification remain essential requirements. Liu believes that low-code applications offer a distinct advantage because every feature is developed to address a specific business need, reducing unnecessary complexity and creating a clearer path for demonstrating the compliance of AI-enabled functionalities.
"Regulatory agencies and industry guidelines increasingly recognize that the adoption of new technologies is inevitable," Liu observed. "When supported by appropriate controls and governance mechanisms, the benefits can significantly outweigh the associated risks. The regulatory framework is evolving alongside these technologies, providing a clear foundation for responsible implementation."
3. Translating Business Needs into Practical AI Applications
AI delivers the greatest value when it addresses clearly defined business challenges. At Asymchem Clinical (Clin-Nov), current AI implementations focus on areas such as GLP documentation, GCP study start-up activities, and GMP equipment validation. Despite serving different functions, these use cases share several common characteristics: they involve highly repetitive tasks, rely heavily on institutional knowledge and prior experience, and carry a significant risk of human error.
One example is the generation of Site Start-Up (SSU) documentation, a critical yet highly labor-intensive component of clinical trial initiation. SSU teams are often required to complete site-specific application forms for hundreds of hospitals, each with its own documentation requirements and formatting conventions. Managing updates and amendments can be equally demanding, particularly when revision histories must be maintained in accordance with regulatory expectations.
"AI enables us to automate large portions of document population, comparison, and revision tracking," Liu explained. "This allows employees to spend less time on administrative work and focus more on activities that require professional judgment, such as content review and quality oversight."
In GMP equipment validation, the challenge is different. The primary obstacle is not document volume, but the variability of experience across teams. Given the diversity and complexity of equipment types, less experienced personnel may struggle to interpret regulatory requirements consistently, leading to inefficiencies and increased risk of errors.
To address this challenge, Asymchem Clinical (Clin-Nov) has incorporated equipment classifications and most routine change scenarios into its digital workflow. AI can automatically identify comparable historical validation projects and use the most relevant validation protocol as a starting point for new activities.
"By making prior knowledge more accessible and reusable, we can reduce variability across teams and help ensure that validation activities meet internal quality expectations from the outset," Liu noted.
Ensuring the reliability of AI-generated content is equally important in GxP-regulated environments. As a result, multiple control mechanisms have been embedded directly into operational workflows.
Human oversight remains a core requirement. After completing a task, AI generates a summary report that clearly identifies incomplete fields, outstanding items, and areas requiring user verification. Employees remain responsible for reviewing, supplementing, and approving the final output.
Automated comparison capabilities provide an additional layer of quality control. When protocol modifications or process changes occur, AI can identify discrepancies, highlight affected sections, and flag content requiring updates, reducing the likelihood of omissions during document revision.
In GMP validation workflows, regulatory intelligence has also been integrated directly into the user experience. Relevant regulations and guidance documents can be referenced automatically during report preparation, while AI evaluates entered values against predefined acceptance criteria and alerts users when results fall outside expected ranges or warrant additional review.
This combination of AI-driven automation and human accountability reflects one of the most widely adopted principles in modern AI deployment: the Human-in-the-Loop approach. Rather than replacing human expertise, AI enhances consistency, accelerates execution, and supports better decision-making while ensuring that ultimate responsibility remains with qualified personnel.
4. Building Trust in AI Through Traceability and Audit Readiness
In GxP-regulated environments, the value of any digital system is ultimately measured not only by efficiency gains, but also by its ability to maintain data integrity, traceability, and audit readiness. As the industry often emphasizes, activities that cannot be documented cannot be demonstrated. For AI-enabled workflows, ensuring transparency and accountability is therefore a fundamental requirement rather than an optional feature.
According to Liu Zhexin, these considerations were incorporated into the platform architecture from the outset. Within Asymchem Clinical's AI-enabled framework, experimental and validation content is structured and stored in Microsoft Dataverse at the section level, allowing version histories to be preserved and tracked throughout the document lifecycle. Every modification can be reviewed, traced, and, when necessary, rolled back to a previous version.
This traceability extends beyond document management. User interactions with AI-generated content are also captured and analyzed, creating a continuous feedback loop that supports both process improvement and model optimization. As employees revise, supplement, or refine AI-generated outputs, those interactions provide valuable insights into how future recommendations and responses can be improved.
In GMP validation workflows, newly generated content, approved revisions, and validated protocols are continuously incorporated into the organizational knowledge base. This enables historical knowledge to be preserved, reused, and expanded over time, creating a self-reinforcing cycle of learning and operational improvement.
The discussion surrounding AI governance inevitably extends to ethical oversight. While AI ethics has become an increasingly prominent topic across industries, Liu advocates a pragmatic approach within regulated environments.
"At this stage, ethical review should remain integrated within existing governance and review processes," Liu explained. "Human involvement continues to be essential, particularly when evaluating decisions that may have quality, compliance, or patient-related implications."
This perspective reflects a broader principle guiding AI adoption at Asymchem Clinical. Rather than creating parallel governance structures for emerging technologies, the organization seeks to incorporate AI capabilities into established quality systems, validation frameworks, and oversight mechanisms. In doing so, innovation can be introduced without compromising the rigor, accountability, and compliance standards that define GxP-regulated operations.
5. Driving Adoption by Adapting Technology to Users
Technology alone does not determine the success of AI adoption—people do. Rather than investing heavily in developing a large workforce of AI specialists, Asymchem Clinical (Clin-Nov) has focused on making AI accessible to existing business users.
According to Zhexin Liu, the objective is straightforward: adapt the technology to the user, not the other way around. Scientists continue to work within familiar, Word-like interfaces, while less experienced team members can rely on conversational AI to identify appropriate templates, navigate procedures, and quickly access relevant SOPs and regulatory requirements.
"Organizations do not need everyone to become a prompt engineer," Liu noted. "What matters is enabling employees to benefit from AI naturally within their existing workflows."
This theory reflects a broader respect for domain expertise. The value of scientists and operational teams lies in their ability to interpret data, exercise professional judgment, and make informed decisions—not in spending time learning how to operate increasingly complex digital tools.
6. Looking Ahead: From Efficiency Gains to Unlocking the Value of Data Assets
Looking ahead, Asymchem Clinical (Clin-Nov) plans to expand AI applications into additional areas, including pharmacovigilance (PV) data source management, signal detection, and natural language-based data querying. Beyond these initiatives, Zhexin Liu highlighted the development of a clinical data warehouse as a key strategic priority.
"In the AI era, the most valuable assets an organization possesses are its historical data and accumulated knowledge," Liu noted. "Traditional document-based approaches often create information silos, and valuable experience can be lost as employees move between roles or leave the organization."
To address this challenge, Asymchem Clinical is investing in the development and expansion of a clinical data warehouse. By providing AI with study objectives and contextual information, the system will be able to leverage large volumes of clinical standards, metadata, and historical project knowledge to identify relevant information and generate insights. "This is an area with tremendous potential that deserves further exploration," Liu said.
In Liu's view, this represents the long-term vision behind Asymchem Clinical's AI strategy. Near-term applications such as report generation and form completion primarily address operational efficiency and productivity. The broader opportunity lies in activating decades of accumulated clinical data and organizational knowledge to support higher-quality decision-making and create new value for drug development programs.
When discussing how generative AI may reshape service delivery within the CRO industry, Liu emphasized the importance of managing and governing organizational process assets. Organizations that can effectively use vector databases to organize and retrieve tens of thousands of similar study templates, historical validation protocols, and accumulated project experience will be better positioned to connect activities across the development lifecycle—from preclinical reporting to GMP validation workflows.
According to Liu, this capability has the potential to become a significant competitive differentiator, enabling CROs to improve both the speed and quality of service delivery while establishing barriers that are difficult for competitors to replicate.
Conclusion
Zhexin Liu's perspective reflects more than a CRO's AI journey—it highlights a practical path for digital transformation in highly regulated environments.
Rather than treating AI as a disruptive force, Asymchem Clinical (Clin-Nov) has adopted an enablement-focused approach, combining the flexibility of low-code platforms with the power of large language models while maintaining strict GxP compliance.
From addressing immediate operational challenges such as SSU documentation to building a clinical data warehouse that unlocks the value of historical knowledge, the company's strategy balances long-term vision with practical execution. As the role of AI continues to evolve, this combination of innovation, governance, and knowledge management may become a key differentiator for CROs in the years ahead.
