AI is not a universal, but the scope and work requirements should be clear

In systems design, information systems and software engineering, the software application growth life process (SDLC), likewise described as the application development life-cycle, is a process for preparation, producing, screening, as well as deploying an information system. The systems development life process principle puts on a variety of software and hardware configurations, as a system can be composed of equipment only, software only, or a combination of both. There are normally 6 stages in this cycle: demand evaluation, style, growth and also testing, implementation, documentation, and also assessment.

Artificial Intelligence (AI) Overview is understanding, but how should I use it in the system building site? To answer these questions, we will introduce the points to know when building AI systems.

■ Requirements can not be written on the desk

The requirements are the most important process that influences the success of the information system deployment project as well as AI projects. In particular, in the AI ​​system deployment, the roles that AIs should be done or the purpose of building AI systems should not be clear, so the difficulty of requirements is higher than that of the conventional time-consuming system. In addition, it is possible to listen to the AI ​​system deployment feature that there is a conceptual demonstration and proof of concept at the steps of the requirements. Unlike projects to build a common business process, most AI system construction projects exist as many as those that can not be judged in the desk for the steps of the requirements.

■ Project range (SCOPE) Quantification and refurbishment

When initiating a project, you must clarify the AI ​​system deployment purposes. It may be a 'natural thing', but it is a difficult problem in the first time in the AI ​​system construction project. In other words, the frequent failure is to be the purpose of the introduction of AI itself. AI is only one means to achieve business purposes. 'AI introduction' Objective must be braked down (Breakdown). Also, the user (USER) can often see if the 'AI is universal'. It should not be done. Let's think that you are introducing a system that predicts demand for stores in retailers and supports product ordering. Do you think you can get any effect on AI system? In the actual project, there may be a purpose of building the following AI system.

• Improvement of marketing goods by missing

• Cost reduction in inventory disposal rate

• Reduced ambulance due to shortening of product order business hours

• Enhancements to increase the product requirements to the employee to strengthen customer's product proposal

• Impressive quality realizing customer's potential purchase desire

Some of these are the readers that feel the there is a crowd for the purpose of supporting the product orders. However, if you start a real AI system deployment project, it is not.

■ AI is not universal

To proceed with a project with constant costs and schedules, you must set up a reality purpose and scope (scope). One of the valid means to clarify the scope of the project is to quantify or specify the purpose and effectiveness of AI system construction. If the purpose of deployment in a plurality of perspectives is known, more specifications to be aimed at achievement are more likely to determine the priority of the purpose.

As soon as the previous example is set to 'Cost reduction according to the deterioration of the inventory lung rate' for project purposes, AI can obtain high prediction accuracy. On the other hand, if you educate the increase in product demand, teaching the employee to reinforce our customer's product proposal ', the interpretation (why AI) is more than the accuracy (why the AI) It is better to select a high algorithm. AI is a great influence on algorithm proposals that are the most important points to create task requirements that require analytical properties.

■ Overall Picture (Overall Picture)

When developing a new service in the AI ​​project, the development of the duration system in which the system requirements are organized and defining the system requirements. Explain how to do this requirements under the premise that there is no existing task to be systemified.

A is intended to work with the work of the system to specify the contents of the system to be developed, and finally the service planning and strategy level. First, listen to the AI ​​leader, PM, IT, and consultants.

PM: Now, I finally do it. How do I go?

IT (IT): Yes! First of all, it is the first thing to do, and then to make system functional requirements.

PM: ... . However, even if it is a new service, it's a new service. Is not it a new service?

IT charge: Um ... . surely… .

Consultants: Development of new system development requirements is different from the development of the periodic system.

In the case of developing a new system in AI projects like this conversation, it is intended for new businesses that are not in existing business. Like a periodic system, the development method of defining system requirements after organizing the work that is subject to systemification is not available. So how can I do it? The AI ​​project is also different in the operation of the work of the requirements according to its content. Specifically, the difference is the same as [Figure 2] below.

First, let's look at the flow of the Regulatory System Development Project. Examples of the Insurance Agreement of Life Insurance Company Inc., the work related to the insurance contract is required to check the insurance contract, checking the total insurance amount of the insured person in the insured person. It is a business requirement that organizes it.

When developing a new insurance product, we organize how to change the existing business. How do I systemize any business requirements during this business requirements? Or to summarize which system functions to change system requirements.

■ The same as the duration of the requirement for business reform

AI projects have two types of projects that automate projects that are currently using AI technology and projects that are currently being implemented. RPA (Robotic Process Automation) belongs to the latter. RPA thinks that there is an existing task (work) and how to automate it. In other words, basically, the flow is the same as the same as the duration system. For example, if you are building a system that reads and recognizes information that is read and recognized by an OCR (optical character-in) system that utilizes AI, the receipt processing procedure itself exists as an existing task, because this is a business requirement do. Any part of the procedure corresponds to the system requirements for automating using AI or RPA.

■ Development of new service must first define 'function' first

A new service development project using AI technology does not exist because it does not have existing tasks. So how can I do it? First, you must define what functions and systems to the system user. If you are a Web application or a smartphone application, you will create a screen (Front End) and a user interface (UI) for your system. We define the system functionality for what services to provide to the user. After being determined by what a system is provided, define what is needed by a task to operate it. When you identify your business process as a business process requirement, the process of the requirements in the new system development project is made in the order of system requirements → the business requirements of the business requirements. The order of the period is the opposite of the order.

■ AI Project Requirements for Developing New Service

If you detail the operation of the AI ​​project for new service development, you can summarize them as shown in [Table 1]. A for a project, define which part of the requirement of any part.

A task is a task that operates such as a branch office of each store, such as membership management or order management. (Table 1), the five types of tasks defined in FIG. First, it is confirmed that the service or representation of the screen function (Front End) is to be provided. Next, we define the management function that supports the operational business or its work to provide the system. The following AI functions or back-end functions, data, and related systems needed to realize the following screen (Front-end) function or management functionality, or data and related systems. Finally, it is defined as a non-functional requirement that accepts all requirements to availability or safety.

Now we find the relevance of individual work. First, the work relationship of the frontend functional requirements, business requirements, and management function requirements will be described.

In the case of a new service development project, the work requirements are not confirmed unless the function is not set for the service user.

For example, if you have the requirements that the user can apply to the service user's screen, the service user's screening, the service user screen, said the user can apply it to the service user screen. To display the campaign, you need a campaign registration task, and to enable system users to apply for campaigns, you will also check your application. Management screen is also required to do business. Campaign registration screens or applicants are referring to a list.

In this way, the requirements of the system development project must first set the front-end functional requirements during system requirements. As a result, the required operation work is organized. Ask the front-end feature requirements, business requirements, and management function requirements, and AI functional requirements and work relations for back-end function requirements. [Figure 7] Tell this.

The AI ​​feature requirements are determined to accommodate the determination of the front-end functional requirements as [Figure 8]. For example, in the front-end functional requirements, the number of recommended breads has been as requirements. In this case, two types of learning to extract three combinations and learning to extract 5 combinations for the requirements of AI are needed. The back-end functional requirement also accepts the determination of the front-end functional requirements.

For example, in the front-end functional requirement, it is a 'payment method as a' payment method. ' Because it is necessary to send billing data to the Payment agency that agencies the account transfer (Payment), since it is necessary to send the billing data to the BACK END side, necessary. The back-end feature requirements are also influenced by business requirements.

It is cumbersome to print a daily basis when performing a job attaching a recipient seal (seal) in a box for bread shipping. It is possible to expect that a back-end feature that collects the recipient seal of the member to be shipped is required. In this way, the back-end function requirement accepts the front-end function, business requirements, and management function requirements.

■ Non-functional requirements are affected by various requirements

Finally, it is a non-functional requirement. This is affected by front-end functionality, business requirements, and management function requirements, and back-end requirements.

For example, the performance requirements required in this system are largely different as campaign display and application functions are attached with attaching (attaching). If you do not attach campaigns and application capabilities, you have only accessible (access) to check the bread on your smartphone. When campaigns and applications are enabled, depending on the campaign period, many system users access the same time.

Security (security) requirements are also affected. If you implement the campaign's winning service in your business requirements or management functionality, high security is required because you are dealing with your personal information in your stores. If you need to send your account transfer request to the external storage agency with back-end functionality requirements, security requirements also arise.

What about availability requirements? Availability requirements are the requirements of what the system does not want to stop (to secure a high uprising rate). If you are a system that can be confirmed by bread or a campaign application, 'Never system is not required, such as' should never be down'. As such, non-functional requirements need to be determined based on various requirements.

This time I found the overall picture of AI project requirements. The next page deals with how to create a front-end functional requirement.

• references

• Textbooks of DX projects that actually use in system development in planning 〮 (Weekly BP, March 2020)

2.Nikkei Systems (2018.10 Monthly Referred Articles

■ I planted ...

Since 1976, we started computer work as a SW developer in KEPCO. He worked as an SI business, such as SI business, while performing information business planning, development and operation for 30 years.

In Korea, the SW business history by the FP (functional number) law, contributed to the advancement of SW business due to the standard research and dissemination of the standard. He spent the SEC policy advisory committee, the SW Business Confection Committee, the Association of Information and Communication Technology Association, KAIST Computer Department, Professor, and SW Policy Institute Invitation Research Institute. At the Graduate School of Soongsil University, a Ph.D. was held as a Multi-regression analyst using FP. Currently, we established a planning technical editorial office to establish SW design, estimate and emotion. In particular, it is active as an SW business requirement for preventing SW business disputes and a basic design expert.

• This column contents may be different from the editing direction.