Insight Paper May 23, 2024
GenAI: Considerations for Implementation
How Trexin can help companies implement GenAI to realize business value.
Generative Artificial Intelligence (‘GenAI’) entices businesses across industries with promises of saving time and money while enhancing their existing processes to become more effective and efficient. New GenAI models and tools are developed and released regularly, each one bringing with it their own potential uses and benefits. The excitement currently surrounding GenAI is encouraging companies to dive into how they could implement it, and how to do so quickly.
Trexin can help companies implement GenAI based on our three-pronged approach that balances receiving the benefits of AI with mitigating the risks. We focus heavily on the use case, the data, and governance. Each of these pieces requires dedicated planning and attention before implementing GenAI to ensure the implementation is not only successful but employs responsible business practices and is safe for both the company and the user(s).
Use Case
The first step in developing an implementation plan for a GenAI is establishing a compelling use case. With all the buzz surrounding GenAI in the news, it can be enticing to skip over the use case development and simply start using it wherever possible to ensure that your business is staying current. However, this can lead to negative outcomes. A fully fleshed out use case is vital in many ways. First, it ensures that where in the business process GenAI is implemented makes the most sense to bring the most value. It then allows for the team to choose the best GenAI model for the job – preventing rework down the road. By setting a defined scope and goal, the product can be measured against the use case to determine success, allowing for retraining and fine-tuning as needed. Companies with low data or AI maturity should begin with use cases that deliver the most value with the lowest level of complexity. Trexin can help companies take stock of their current state business processes to analyze where GenAI can be used to provide the most business value.
Data
After a use case is created, reviewed, and approved, it’s imperative to take a close look at the data that will be used. The output of GenAI is only as good as the data being input. If the data is inaccurate, out of date, or otherwise flawed, GenAI will provide bad output. This could not only lead to the company using incorrect analytics for other deliverables but could ultimately end with legal action being taken due to false information being presented to clients, stakeholders, or external sources. Training data also needs to be carefully selected to ensure the best possible output. Data must be in the correct format for the tool to ingest, which can vary depending on the tool. Trexin can help clean up data/reformat for the tool, review for accuracy, and provide new processes to ensure data clarity in the future. Once the data is confirmed to be accurate and ready for use, the chosen GenAI needs to be trained using carefully curated data to provide the desired outcome. This process can only be successful with high quality data, once again proving the importance of data quality to the entire implementation process.
Governance
From establishing a use case to fine tuning GenAI’s output, governance underlies every step of the implementation process. Governance is a broad term that covers many aspects of mitigating potential business risk, including considering legal, ethical, and security related risks. For legal risks, such as copyright or discrimination, it is vital that companies adhere to any state and federal rules and regulations regarding artificial intelligence. When considering ethical risks, Trexin often looks at the use case. Is this an area that impacts current workers’ jobs? What will happen to them? Does it aid and optimize work, or does it replace people? Asking these questions helps to ensure that the company is protecting its employees while using GenAI to make their jobs smoother. For security related risks, Trexin focuses on the protection of company and client data. When using any tool, there are risks of malicious actors, data leaks, and cyber-attack attempts. Especially when using company data to train an AI, it is of the utmost importance to ensure that there are policies and procedures in place being actively enforced to protect the data, company, and people.
A strong governance framework will also include a process for monitoring model output. Just as human processes are reviewed for accuracy and quality assurance, GenAI output must always be reviewed by human eyes. Model inaccuracies resulting from data drift, hallucination, and bias could lead to reputational, legal, or operational consequences if incorrect output is used for business decisions or delivered to clients. Data drift occurs when new data fed to the model differs from the training data and causes the model to deliver incorrect results. Hallucination can be caused by incorrect or incomplete training data, and results in the model providing incorrect output. Finally, GenAI tools have been shown to have biases based on the data with which they are trained. By looking for, acknowledging, and actively working to counter these biases, companies can ensure their implementation is most effective and appropriate. Monitoring your model is a key piece of the governance framework that protects your company, employees, and clients from the consequences of inaccurate GenAI output.
Trexin can help companies establish, update, and enhance their governance framework to tailor it to their specific business concerns and ensure the protection of the company throughout the implementation and usage of GenAI.
Contact a Trexin Advisor today!
