Generative AI-Driven Business Revolution: Seven Keys to Success for Producing Outstanding Results. Part 1: Generative AI-Driven Business Revolution

In September 2024, ChatGPT grew to become an app with over 200 million users in a single week, increasing the widespread embedding of generative AI in our everyday lives.
The popularizing of generative AI made explosive progress when ChatGPT arrived on the scene in 2022. Its accuracy and response speed improved with the subsequent release of GPT-3.5, and with this release, its use in business took a step further.
GPT-4o was released in May 2024, making it possible to not only generate text like before but also audio and images, as well as text based on audio and images. These advances made multimodal processing possible. Then, in September 2024, with the emergence of OpenAI’s o1, which has the ability to generate “long chains of thought” internally, it became possible to generate highly accurate answers for tasks requiring deep thinking such as mathematical problems or logical puzzles that it was previously poor at. In just over two years since the birth of ChatGPT, the accuracy of LLMs has improved significantly, with its scope of use and potential expanding dramatically.

Before the emergence of generative AI, in AI projects, it was common to use models that had already been trained. In the field of computer vision, models pre-trained on ImageNet were widely used. Meanwhile, in the field of natural language processing, pre-trained models such as Word2Vec and BERT were used for many tasks. Models were also adapted to specific tasks or domains through transfer learning. Data scientists were required to make this happen and there were also certain hurdles to using published models.
However, LLMs in recent years have become universal, high-performance models. We can now easily use these models for a wide range of tasks such as summarizing, translation, and searching. The fine tuning is no longer required and the prompt-based approach that can execute individual tasks just by submitting prompts has significantly changed how people think of AI model construction and AI usage. The lowering of the hurdle that stood in the way of making it possible to apply models to business operations without data scientists has made AI more user-friendly. Discussions now focus on how to effectively use these models in business operations to get the best output from what you put in and how to use generative AI to make business operations more efficient and generate value.

Although there are challenges such as security risks, some companies have begun examining the use of generative AI or undertaking a PoC because it offers large potential benefits. More and more companies are also starting to use generative AI in their business operations. ABeam Consulting is also engaged in projects that utilize generative AI for all kinds of companies and organizations, regardless of the type of business or industry. Last fiscal year there were quite a lot of PoCs conducted for response and inquiry operations with Retrieved-Augmented Generation (RAG), which generates answers from internal information. This fiscal year has seen further considerations given to using generative AI in marketing and system development domains.
How do companies assess the benefits of using generative AI, which has started to be deployed at some companies? In a survey on the use of generative AI by business men and women conducted by ABeam Consulting, 47.2% of those involved in the deployment of generative AI responded that they were already using it at work (see Figure 1). Also, of the respondents who are already using generative AI, 78.8% said they experienced benefits exceeding their expectations. This gives us a sense that most companies that have started using generative AI are now able to utilize it effectively (see Figure 2).
In checking the benefits of deploying generative AI for different uses, approximately 83% of the “AI system development through internal development” respondents experienced benefits exceeding their expectations. This significantly surpasses the 69.5% of those “Using existing SaaS platforms” (see Figure 3).
As reasons why the benefits of using generative AI produced results exceeding expectations (phases in which it contributed to results), “Launching generative AI projects, formulating a vision, and/or identifying use cases” received 39.2%, “Organizing the requirements that generative AI systems or services must meet” received 36.6%, and “Building verification environments, implementing technical verifications, and conducting comparative evaluations of generative AI systems or services” received 11.6% (See Figure 4).
Meanwhile, as for reasons why the benefits of using generative AI did not meet expectations (phases in which efforts were insufficient and failed to meet expectations), “Organizing the requirements that generative AI systems or services must meet” received 33.8%, “Building verification environments, implementing technical verifications, and conducting comparative evaluations of generative AI systems or services” received 26.2%, and “Launching generative AI projects, formulating a vision, and/or identifying use cases” received 17.2% (See Figure 5).
Generative AI has various options for how it can be used, such as internal development or utilizing existing SaaS services. Many of the companies that are using it experience its benefits. “Formulating a vision and/or identifying use cases” and “Organizing the requirements that generative AI systems or services must meet” in the upstream phase are especially important for experiencing the benefits of generative AI in using it for business purposes. From the fact that utilizing generative AI was more likely to offer benefits in internal development than using existing SaaS services, we can see that utilizing generative AI was more likely to provide a positive impact when used in ways aligned with each company’s vision and specific requirements.
On the other hand, those who responded that the results of deploying generative AI did not meet expectations gave “Building verification environments, implementing technical verifications, and conducting comparative evaluations of generative AI systems or services” as the reason why. From this, we can infer that the reviewing of upstream requirements and use cases, including an evaluation of its technical feasibility, was insufficient. It is important to examine the use of generative AI, including from a technical perspective, from the stage of formulating a vision or examining use cases.