The popularity of Chat GPT has sparked a surge of interest in AI and GPT, with many people fearing the impact it may have on their careers and the future of humanity. GPT, short for Generative Pre-trained Transformer, is an algorithm mainly used for natural language tasks such as understanding, generation, and dialogue. GPT utilizes algorithms and computing power to improve hit rates and reduce the time required to get an answer, much like the analogy of monkeys randomly typing on a keyboard until one produces a Shakespeare-level work. By using a pre-trained neural network model based on statistics and probability, the hit rate and time required to get an answer are improved.
A neural network can be thought of as a fishing net composed of nodes and connecting lines between nodes. These connecting lines represent parameters, and the more layers and parameters a neural network has, the more accurate its results. GPT-3, for instance, has 96 layers and 175 billion parameters. For pre-training, a function with hundreds of billions of parameters is trained using supervised or unsupervised methods and fixed input and output data. Afterward, the trained model can generate a close answer to a given question in a short time. As a complex calculation, the process and result are difficult to predict, raising ethical, value, and regulatory issues.
GPT technology's current successful experiments are primarily in natural language processing, which will initially impact search, question answering, and content generation fields. When it comes to the SaaS field, the impact of large models like GPT will likely start with SaaS combined with AI, followed by AI-based SaaS. However, with the development of large models, SaaS-based AI may emerge in some scenarios, with part of SaaS hidden behind as microservices. While GPT technology currently does not affect the database layer, it does impact interaction, input, and output dimensions. The first step towards combining SaaS with GPT technology will bring about changes to the form of SaaS.
I. Interaction Method
Conversational entrances will become a vital mode of interaction.
Will conversational entrances completely replace the traditional graphical delivery method? I don't think so. Conversational entrances will be an essential addition, but certain functions are difficult to replace with conversational interaction, including:
- Functions that require high-frequency operations and data viewed often. These are typically displayed on the homepage or in important areas, and interaction through dialogue may reduce efficiency.
- Functions that require complex descriptions and logic, as direct display of results on the graphical interface or clicking buttons are more efficient.
However, conversational interaction can easily replace certain scenarios, including:
1. Low-frequency data viewing or long-tail functions requiring low-frequency operations
As SaaS systems tend to have complex business structures, about 90% of the demand is for low-frequency long-tail functions. As the information and function displays are relatively complex, their weight may be weakened to improve system usability, making them inconvenient to operate. Conversational interaction with these functions can significantly improve the product's usability.
2. Personalized data statistics query functions
Standard products often lack support for personalized statistical query functions. Conversational interaction patterns can be used to support such functions, requiring a large model between the display and logic layers to interpret natural language commands and call and execute corresponding actions.
3. Voice input-friendly scenarios
Natural language input may be more convenient and natural for some data, depending on the scene and the quality of the crowd.
In the future, conversational entrances will likely be present on the homepage of all SaaS software, and in some industries, they may serve as the primary entrance for software scenarios.
To enable natural language conversational interaction, a large model interpretation layer for natural language is necessary between the presentation and logic layers to execute database operation commands.
II. About the Data Entry Section
For SaaS software, data entry is a critical function that involves manual input or import of structured data such as personnel, salary, customer, visitor, and order information. However, this kind of information input is a repetitive and cumbersome task, which is also one of the reasons why people find SaaS products cumbersome. With the development of GPT technology, data entry can be automated and moved to the front. This means that the use of natural language speech, text, and pictures will increase significantly. But, there is a big challenge in turning unstructured data in natural language into structured data. Even with GPT technology, this problem cannot be solved entirely, and it takes a long time to achieve full automation in many scenarios.
For instance, in a sales or purchase order scenario, when two people chat, a lot of order-related information comes out during the conversation, which needs to be identified by GPT and then inputted into the system. However, this input is not efficient, as the structured information outputted by GPT needs to be confirmed or corrected by humans. Though a few vertical scenarios, such as obtaining some sales chat clues or medical record information, may achieve automatic structured natural language data in the long run, most fields and scenarios require human input. Therefore, AI-based SaaS is more feasible in the present scenario.
III. New SaaS Opportunities
The advancement of GPT technology will significantly impact various natural language and image SaaS software, such as communication, conferencing, content creation, and visual design.
Currently, SaaS software dealing with natural language, images, conferences, email, Office, and picture creation are the first to benefit from GPT technology. The application of GPT technology in content generation scenarios allows for more efficient content creation, making the software more productive. Therefore, these are the first SaaS software to apply GPT technology.
Over time, language, text, and picture-based software and companies will merge with data software systems such as CRM, ERP, and HR, resulting in more robust content output. This integration will blur the boundaries between SaaS products. As GPT is efficient in processing language and images, these fields are beyond the reach of traditional SaaS. This expansion will generate numerous new SaaS tools based on specific scenarios, particularly in some lower-level markets.
IV. Impact on SaaS R&D
The development of GPT technology can significantly enhance the efficiency of research and development, as a vast amount of code can be automatically generated with minor modifications. One example of such an opportunity is the PaaS platform.
It is highly probable that new tools for natural language programming development will emerge, such as qqbot.dev and similar tools. Early companies have been exploring this area, and traditional development platforms can use GPT technology to become more intelligent.
For some relatively focused vertical scenes, there is potential for the creation of vertical PaaS platforms developed through natural language, such as a development platform similar to WeChat Mini Program. These platforms target relatively simple and pure scenes and have the possibility of developing through natural language.
There are opportunities for the existence of standard SaaS in fields such as CRM, HR, Finance, ERP, and others. Additionally, there are opportunities for PaaS platforms based on GPT technology to supplement standard SaaS in these vertical fields. Standard SaaS companies can combine this type of PaaS platform to complete deliveries more efficiently.
V. Effects on the Role of Various Departments in SaaS
- The development of GPT technology can greatly empower sales staff in terms of communication with customers through conferences, emails, articles, PPT, video scripts, and other related software tools.
- GPT technology can also automate a large number of customer service and after-sales processes, benefiting the customer success department.
- For the R&D department, GPT technology greatly improves product development efficiency, automating the generation of much of the code. This may reduce the number of primary and intermediate R&D people needed, while increasing the demand for architects and engineers with technical solutions.
- GPT technology can also improve efficiency in document writing for product departments, but demand research and design work cannot be replaced.
With reduced research and development costs and increased efficiency, software development will expand, creating a need for more product managers who can understand customer needs and describe them clearly.
For software requirements in simple scenarios, a product manager can use GPT PaaS software to undertake all the work of requirements research, design, development, and testing.
In the long run, if GPT technology cannot lead to the subversion of an enterprise database, SaaS will continue to exist but will have far-reaching impacts on the industry, including:
- The use of GPT technology is expected to solve the problem of SaaS software being generally difficult to use, improving user experience and accelerating the informatization speed of the sinking market.
- GPT technology is expected to greatly reduce the cost of developing products and delivering services for SaaS companies.
- GPT technology is also expected to improve delivery efficiency and maintenance cost for some personalized needs.
A huge wave is coming, and we all need to actively embrace it. In the long run, the people who will succeed in the SaaS industry are not only those who are straightforward and capable of execution, but also those who are patient and steady.