CEO-Led AI Governance Driving an 87% Adoption Rate
The result of reducing conflict and KYC (Know Your Customer) review tasks—which previously took up to 8 hours—to just a few minutes after the introduction of Codex led to an enterprise-wide AI adoption rate of 87%. As of June 2026, the active user rate of 87% is more than double the typical adoption rate for internal tools, and this figure includes partners and general staff as well as the technical department. This demonstrates how the productivity floor of an entire organization rises when AI tools become an enterprise standard rather than the exclusive domain of a few early adopters.
CEO Sam Nickless led the adoption by directly sharing his own actual work cases with employees. He clearly conveyed the message that AI is not "cheating," defining it not as an improper shortcut, but as a tool to apply human professional judgment more efficiently. The key was that the leader first disclosed how to use the tool and its limitations, thereby removing the psychological resistance and ethical burden employees might feel.
The Business Transformation team directly demonstrated customized workflows tailored to the characteristics of each department, such as marketing, finance, and recruitment. Rather than simply explaining features, they chose a method of visually showing the specific paths through which actual departmental tasks are automated. The strategy of translating technical possibilities into the practical language of individual departments led to a substantial increase in usage rates.
During the implementation process, no mandatory targets for usage were set. Instead, while the scope of what the technology could perform was clearly demonstrated, a principle was strictly applied that the human remains entirely responsible for the final output. By clarifying the locus of responsibility for the results rather than forcing the use of the tool, they prevented AI from becoming a means of evading professional accountability.
Now, they are transitioning from ChatGPT, a simple auxiliary tool, to Codex, which directly executes multi-step workflows. Codex is a code generation and execution model that builds a system to automatically complete several defined steps, moving beyond the level where individual users handle one-off tasks. This has raised the level of automation by expanding the role of AI from a simple assistant to an operational entity that processes defined workflows.
This confirms that strong leadership support and a shift in perception are essential prerequisites for enterprise-wide AI adoption. In practical terms, securing psychological safety regarding AI use, rather than technical training, becomes the key variable determining the actual speed of adoption. However, such governance is a strategy that is only effective in an organizational environment where transparency and leadership influence are secured to the extent that the CEO can directly share work cases.
Digital Twin GPTs and Departmental Operational Support Systems
A custom GPT (a chatbot for a specific purpose) was built by training it on the CEO's writing style, priorities, and background knowledge. Executives utilize a "digital twin" structure (a model that replicates a real-world object in a virtual world) by inputting ideas into this chatbot for pre-validation before reporting directly to the CEO. When a hypothesis is entered into the model that replicates the leader's way of thinking, the model suggests expected questions or areas for improvement from the CEO's perspective. This is the result of adding a pre-filtering process to adjust the direction of proposals and refine logic before the reporting stage.
The marketing and sales teams use ChatGPT to synthesize pitch materials—amounting to 400–500 cases annually—and write customized responses. By utilizing a vast amount of previously written proposal data, the system automatically extracts the most appropriate logic and style when a new business opportunity arises. They have secured both the quantity and quality of proposals by reducing writing time while maintaining the company's defined writing standards. This is a structure that immediately combines the company's past assets with current sales strategies, moving beyond simple text generation.
The finance team applied AI to spreadsheet management and data analysis tasks to increase numerical processing efficiency. AI performs the process of writing complex formulas or finding specific patterns in large datasets and converting them into statistical figures for reports. The workflow was reorganized so that humans review the analysis results of structured data. This is an attempt to increase the accuracy of financial operations by expanding the scope of AI utilization from unstructured text to the realm of structured data analysis.
The technical team deployed AI for creating internal guides, writing scripts, and system documentation. AI is exclusively responsible for tasks such as writing technical documents that explain new feature implementations or generating repetitive server management scripts. This allowed engineers to spend less time on ancillary administrative tasks like documentation and more time on core architecture design and implementation. This is an example of removing operational overhead with AI to increase development productivity.
The deployment of custom GPTs trained on the context of specific roles and the leader's way of thinking is a practical approach to transforming general-purpose AI into an enterprise-specific knowledge base. The core is lowering internal communication costs by modeling the judgment criteria of decision-makers within the organization, moving beyond a simple Q&A tool. However, this system is optimized for operational support and documentation tasks and has not yet reached the stage of replacing core areas where human responsibility and professional intuition are essential, such as legal judgments or final strategic decisions.
Transitioning from Simple Assistance (ChatGPT) to Execution (Codex)
Codex (a code generation and execution model) automatically performs defined multi-step workflows, distinguishing it from ChatGPT, which helps with one-off tasks. While ChatGPT is an auxiliary tool that answers individual user questions or creates drafts, Codex is an execution system that produces actual results according to set procedures. This transition has expanded the role of AI from a simple advisor to a worker that completes actual tasks.
Codex removed one day's worth of manual labor from the overall workflow of creating audit reports for 300 entities (legal corporations). Previously, humans had to check data individually and enter it according to the report format, but Codex reduced this time by directly performing the multi-step operational process. This demonstrates the efficiency that occurs when AI possesses actual execution power in administrative tasks that are complex in rules but repetitive.
The task of renaming 1,100 files and uploading them to another system, which previously took several days, was also completed within minutes through Codex. Codex solved irregular file naming rules or exception scenarios—which are difficult to implement with general automation tools—by processing them via code. This is an example of expanding the scope of automation by applying a code generation model to irregular tasks that are difficult to turn into standardized processes.
The Business Transformation team implemented actual working Python applications based on requirement specifications and existing specification documents. Codex interprets requirements written by humans in natural language and converts them into executable code, allowing internal tools to be built quickly. The development cycle was shortened as developers no longer wrote all the code themselves but instead reviewed and modified the code generated by AI.
The DevOps (integration of development and operations) lead in the technical team used Codex to build a "Watchtower" system for monitoring the AWS (Amazon Web Services) environment. This system integrates scattered operational signals into one, supports diagnosis and recovery actions when failures occur, and performs escalation (reporting and transferring to a higher-level manager) at the point where human intervention is required. This is an AI-based automation system that goes beyond simple notification functions to include the execution stages of diagnosis and action.
An execution system based on Codex provides higher productivity than a simple chatbot, but it must be predicated on human review. Because the code generated by AI moves to an execution stage that directly affects actual infrastructure or data, a verification process at the final approval stage is essential. Therefore, the practical value of this model lies not in full automation, but in the draft execution stage that drastically reduces the amount of work humans must review.
Quantitative Results: Reducing 8-Hour Tasks to Minutes
Review times for conflicts, KYC (Know Your Customer), and AML (Anti-Money Laundering) were reduced from 3–8 hours to a few minutes. KYC is the process by which financial institutions verify a customer's identity, and AML is a system to prevent the laundering of illegal funds. This is the result of automating the repetitive cross-checking tasks and conflict review processes—verifying whether conflicts of interest exist—that occur during the compliance stage of a law firm. By reducing the time spent on simple verification tasks, practitioners were able to focus more on higher-level work, such as judging the appropriateness of the review results.
By applying AI to recruitment research and data extraction workflows, tasks that took approximately 4 hours were reduced to 20 minutes. In the process of handling candidate references (reputation checks), approximately 25 minutes per person were saved. This is an example of removing bottlenecks in the simple repetitive process of collecting and refining numerous resumes and external data to classify them according to specific criteria. Efficiency in the data collection stage leads to practical benefits, such as increasing the speed of recruitment decisions and strengthening competitiveness in securing top talent.
One day's worth of manual labor disappeared from the entire process of writing audit reports for 300 entities. An entity refers to an individual legal corporation that is the subject of legal rights and obligations. The core is the automation of the process of gathering scattered data from hundreds of targets and converting it into a designated report format. In report writing tasks dealing with large datasets, AI replaced the simple transcription work performed by humans, reducing the possibility of data omission.
The task of renaming and uploading 1,100 files was reduced from several days to a few minutes. This is the result of automating simple management tasks, such as applying naming rules to the file system in bulk and deploying them to designated servers. This corresponds to the efficiency of the engineering domain, where mass processing according to set rules is required rather than high-level intelligent reasoning. The full automation of simple repetitive tasks reduces the fatigue of manual labor felt by practitioners and increases their immersion in core tasks.
These quantitative achievements are concentrated in operational support tasks rather than legal reasoning, which is the core of legal advice. AI did not directly establish complex legal logic but operated as an auxiliary tool to quickly prepare the basic data necessary for experts to make judgments. Practically, a strategy was taken to increase the hourly production value of professional personnel by removing time from the data preprocessing stage. However, since the structure is one where humans still bear responsibility for the final output, an internal process for verifying AI outputs must be a prerequisite.
Prerequisites for AI Adoption in Professional Services: Data Residency and Accountability
Gilbert + Tobin, a leading Australian law firm, established an Australian data residency environment (where data is physically stored on servers within Australia) while adopting the OpenAI environment. Through this, they met the internal security requirements and customer expectations of an organization handling sensitive client information and business data. Management control was strengthened by reviewing role-based access control (a security method that limits system access range according to user permissions) and data processing requirements in advance.
For professional legal workflows, separate approved platforms such as Harvey (a legal-specific AI platform) are used to handle operational support tasks. A structure is maintained where humans directly set task limitations for all outputs, verify the results, and perform professional judgment and final approval. Instead of directly applying generative AI to the core judgment process of legal advice itself, the focus is on raising the standard of operational support tasks that support it.
As explained, "Codex can leap AI from a simple assistant to an executor. It directly performs the steps of the operational workflow, and our people are responsible for reviewing and approving the results," a system operates where the system handles actual execution and humans bear the responsibility. The future goal is to build an interconnected environment where users can access organizational context and derive results without manual movement between systems.
Image source: OpenAI
To expand the adoption of artificial intelligence in professional environments, a strict governance system and the securing of data sovereignty must come first. The case of Gilbert + Tobin shows a practical standard for increasing operational efficiency based on thorough security controls and clear limits of human responsibility.
