Woter AI detection.Hurry - ends Jun 29th

New Year Sales :up to 80% OFF

AI Humanize AI Translator Bypass AI AI Rewriter AI Detector

PRICING

TRY FOR FREE

How to build an AI Agent to quickly analyze industry prosperity

Written by

Iris Vance

Updated on:June-13th-2025

// Report content //

How to build an AI Agent to quickly analyze industry prosperity

How to solve the pain points of the research on the prosperity of the artificial industry through AI Agent?

The current iteration speed of AI technology is relatively fast. Terminal products such as KIMI and Tencent Yuanbao have gradually become important tools for investment research personnel. The main reason may be that their fast information search capabilities, long-context reading comprehension and induction capabilities, and rigorous logical reasoning capabilities meet the needs of financial investment research.

We believe that current AI applications are more suitable for text-based collation and analysis, such as collecting recent data and opinions of a certain industry, or doing preliminary data collection and financial analysis of a single company. In strategic analysis, the scenario that is more suitable for AI replacement may be to analyze and compare the prosperity of different industry-specific time periods, such as the prosperity of the current month or the current quarter.

The original industry prosperity analysis requires organizing qualitative and quantitative (monthly, quarterly, and annual frequency data) data frameworks of different industries, updating these data and information at specific points in time, and conducting prosperity analysis on a single industry, and then conducting quantitative/qualitative comparisons of prosperity across different industries. The pain points of the overall research may be:

(1) The workload is large , and each update requires manual collection and update of a large amount of data; (2) Adding new sub-industries requires a large time cost . Since the corresponding framework needs to be sorted out when establishing a single industry analysis framework and database, it takes relatively more time to re-establish the framework when adding new sub-industries or popular concepts/tracks after the industry classification framework is determined. For example, after determining the analysis framework of automotive parts, the track of humanoid robots is derived, and the relevant analysis framework and database need to be rebuilt.

In view of the above pain points, we give full play to the characteristics of AI big models and build an AI workflow based on the COZE platform to quickly analyze the prosperity of specific industries. AI workflow is an automation framework that embeds artificial intelligence technology into business processes, processes tasks through preset steps and rules, and achieves efficiency improvement and cost optimization.

Why use AI Agent solutions with preset steps nowadays?

Due to the limitation of input/output tokens, it is difficult for KIMI, Yuanbao and other platforms to give comprehensive answers to complex and comprehensive questions such as economic conditions in a single Q&A session, and the consistency of multiple answers is weak. On the one hand, AI Agents that independently plan tasks like Manus have weak controllability and poor applicability for investment research tasks that require high precision. On the other hand, the cost of completing a single task is also higher than that of AI Agents with preset steps. The main reason is that it is difficult to assign large models with different attributes according to tasks of different difficulty levels.

The main construction idea is to first build a large model with high performance to plan the analysis framework for a specific industry, and then sort out the corresponding subdivided items through two parallel loop components. After the sorting is completed, all information and data are comprehensively analyzed through a large model with long context, and scores are given to each item, finally obtaining a comprehensive score for the industry.

The AI Agent method can be used to analyze the prosperity of a specific industry in a specific time period more quickly and in a customized manner. The industry name and time period have a higher degree of freedom and do not require excessive labor costs. It complements the traditional prosperity analysis method. According to the backtest results, the prosperity score of the AI Agent also has a certain correlation with the rise and fall of the industry index in the corresponding time period. In addition to supplementing the information, it can also provide a reference for industry comparison and industry timing.

Preset steps AI Agent construction process

The construction methods on different AI workflow platforms may be slightly different, but we believe that the construction ideas are generally similar. Therefore, below we mainly show the process and main details of building AI Agent on the Kouzi platform for reference by investors.

First, create an agent in the workspace, then create a workflow, and set up the relevant segmented agents on the business logic page.

Initial node: The initial node is the content/parameters that the user/user needs to input. In this workflow, the business climate analysis requires the user to input two key parameters: (1) The specific industry to be analyzed: such as photovoltaics, new energy vehicles, real estate, infrastructure and other sub-industries. Users can refer to the first, second, and third-level classifications of industries such as Shenwan and other industries to ask questions, or ask questions about customized industry names, such as brain-computer interface, humanoid robot, robot reducer, etc. The logic behind this is that the AI Agent in the subsequent process will understand the relevant industry names and generate keywords for search on the Internet. (2) The time interval that needs to be analyzed, such as the first quarter of 2025, the second half of 2024, February 2025, etc. The time interval can be set freely, but it is also recommended to enter the more commonly used monthly and quarterly intervals. If the time interval entered is relatively high frequency or relatively close to the current time point (for example, in May 2025, analyzing the prosperity of the XX industry in May 2025), there may be a problem of lack of network information and data, so the comprehensive score result is also difficult to meet the comprehensive and objective standards. However, if the user wants to use this AI Agent to search for the latest industry-related information, he can also search for a relatively recent time interval. The subsequent AI analysis report will list the relevant news and data within the interval. (3) The number of factors that need to be analyzed. The subsequent AI Agent will split the framework of the corresponding direction according to the number given by the user. The more the number given by the user, the more analysis angles and data collection directions the AI splits. However, from the perspective of cost, the larger the number, the more tokens and time will be used. Therefore, we believe that 5-10 split factors may be more reasonable. If more precise analysis is required, the number of analysis factors can be increased to more than 10.

Big Model 1 - Proposing an Industry Analysis Framework: After the initial node, we introduce the first big model to propose an overall industry analysis framework. After proposing the industry analysis framework, each angle is output in the form of an array to the big model used to search for information later. We provide the corresponding prompt word examples for investors to use as a reference:

System prompt words:

# Role

You are a professional and rigorous financial analyst who provides users with a fundamental data analysis framework for analyzing the investment value of an industry or a stock in {{input2}}.

Based on the relevant information, you analyze which sub-segments can be used to describe the {{input}} industry's {{input2}} prosperity and operating conditions, and make an analysis list of relevant directions, with a maximum of {{input3}} items. Later, you will conduct {{input2}} fundamental analysis of the {{input}} industry based on the data searched in these directions.

### Skills: Sort out the fundamental data analysis framework

Step 1: When a user requests an {{input2}}} investment value analysis for a certain industry, first clarify the name of the industry.

Step 2: If you want to analyze the fundamental prosperity of {{input}}} industry {{input2}}}, what dimensions can the fundamental analysis be divided into?

Step 3: Sort the dimensions by importance and analyze which dimensions are most representative, which can accurately analyze the operating conditions of the industry. Output the top {{input3}} dimensions with the highest importance.

## limit:

- Only answer questions related to the industry or stock investment value fundamental data analysis framework in the field of financial investment, and refuse to answer irrelevant topics.

- The output must be organized in a quantitative and structured manner as required above and cannot deviate from the framework requirements.

- The language expression should be rigorous and professional, and avoid ambiguous expressions.

User prompt words:

Based on the relevant information, you analyze which fundamental analysis dimensions can be used to describe the {{input2}} prosperity and operating conditions of the {{input}} industry, and make a list of relevant data, with a maximum of {{inpu3}} items.

Note: Input is the industry name that the user needs to enter at the initial node, such as photovoltaic equipment; input2 is the time node that the user needs to enter at the initial node, such as the first quarter of 25; inpu3 is the number of factors that need to be analyzed, such as 3.

Loop - Automatically collect corresponding data for each dimension: After completing the design of the analysis framework in the big model 1, we set up loop nodes, which is to automatically collect data on the network corresponding to the segmented analysis framework items in sequence.

The big model 2 in the cycle - collects data from the characteristic time interval of the analysis item: the first big model is constructed in the cycle interval. The big model is used to collect data from the characteristic time interval of the specific analysis item. For example, the initial node inputs the parameter "photovoltaic, Q1 25, 3". In the big model 1, the fundamental analysis of photovoltaics is divided into three sub-analysis frameworks: "photovoltaic product price trend, photovoltaic capacity change, photovoltaic related policies". Then the big model 2 will collect network data according to these three sub-analysis frameworks in turn. The first one collected corresponds to the "photovoltaic product price trend" item, and collects relevant information and news on the Internet. The corresponding prompt words are as follows:

System prompt words:

# Role

You are a professional and experienced financial analyst with profound financial knowledge and keen market insight, and you can accurately search and analyze relevant data of a certain industry for users.

## Skill

### Skill 1: Search industry data

1. Use the tool to search for relevant fundamental data of the industry {{time}}

2. Organize and analyze the searched data, sort out and summarize the relevant data, and present them as completely and structured as possible.

## limit:

- Only answer questions related to search and analysis of industry {{time}} related data, and refuse to answer irrelevant topics.

- The output data and analysis content must be accurate, clear and organized according to the given format.

- Data sources must be searched and obtained through reliable tools.

User prompt words:

You need to search for data in the {{time}} interval of {{input}} in {{topic}} industry on the Internet. If you cannot find directly corresponding data, please find relevant data based on the information to infer the changes in the target data, and add () to the output results of each sentence (abbreviation of the data source, hyperlink URL). Pay attention to the timeliness of the data during analysis.

Industry: {{topic}}

List of all data: {{all}}

Searched data results: {{article}}

The data you need to find now: {{input}}

Note: topic is the industry name that the user needs to enter at the initial node, such as photovoltaics; all corresponds to the array of all analysis items output in the large model 1, article is the result of the previous cycle, input is the analysis item that needs to be collected in this cycle, and time is the time interval that needs to be collected, such as the first quarter of 25.

Big Model 3 in the cycle - Collect historical trends of data related to the analysis item: To understand the prosperity of an industry or a segment type, it is necessary to compare the current value of the segment data with the corresponding historical value. Therefore, in addition to collecting data for the current time interval, historical data of the corresponding data caliber must also be collected. In the data collection part, we split the characteristic time interval data and historical data into two big model processes. Big Model 3 is mainly used to collect historical values of data in a specific time interval of Big Model 2. After the historical value and current value of a segment item are collected, it is convenient to hand it over to the subsequent big model for overall analysis. Examples of prompt words corresponding to Big Model 3 are:

System prompt words:

# Role

You are a professional and experienced financial analyst with profound financial knowledge and keen market insight, and you can accurately search and analyze relevant historical data of an industry for users.

## Skill

### Skill 1: Search industry data

1. Use the tool to search for historical data related to the industry. The list of data to be found refers to the list in {{input}}.

2. Organize and analyze the searched data and present it to users in a clear and understandable way.

## limit:

- Only answer questions related to searching and analyzing industry-related data, and refuse to answer irrelevant topics.

- The output data and analysis content must be accurate, clear and organized according to the given format.

- Data sources must be searched and obtained through reliable tools.

User prompt words:

Below is the relevant data of {{topic}} industry {{input2}}. Please search for relevant historical data of similar caliber based on the data below to facilitate longitudinal historical analysis. The searched historical data is retained as completely as possible, but you need to do some preliminary sorting.

Note: input refers to all analysis items given by the large model 1, input2 refers to the subdivided items searched in this round of loop; topic refers to the industry name input by the initial node.

In terms of setting up large models in the loop, we recommend using large models that are faster and less expensive. The main reason is that the loop mainly uses a "quasi-series" workflow, where the next task is started only after the previous task is completed. Therefore, using a high-speed model can shorten the working time of the overall workflow. Large models 2 and 3 are mainly responsible for extracting effective data from the search engine, and do not require high model reasoning capabilities.

Other nodes: text processing - summarizing the data collected by large model 2 and large model 3, variable setting: packaging the data collected in this cycle and transferring it to the next cycle, and finally collecting all data segment items in the final cycle to obtain the summarized data collection results.

Since the news and data that can be collected by the search engine corresponding to a single model are limited and have a certain degree of randomness, we are replicating another cycle process as a supplement. That is to say, the analysis items given by the big model 1 will be handed over to the big models in the two cycles for collection and analysis at the same time, and the final data reports will be summarized into the big model used for subsequent data analysis and industry scoring.

Big Model 4 - Analyze the industry's prosperity based on data and give a score: The data collection of each segment item has been completed in the cycle. We set up Big Model 4 to complete the overall analysis. There are two main analysis tasks. First, analyze the prosperity of a certain segment item based on the comparison of current data with historical data, and give a score for each segment item (1-5 points, 1 point is the lowest, the data of this segment item is at the lowest position in history, 5 points is the highest, the data of this segment item is at the best position in history); then, according to the scores of different segment items, comprehensively analyze the prosperity of the industry and give a comprehensive score.

This large model requires higher comprehensive reasoning ability, so it is recommended to choose a large model with stronger analytical reasoning ability.

System prompt words:

# Role

You are a professional and experienced financial analyst who can accurately analyze the industry with your profound professional knowledge and rich experience. You need to use rigorous language, quantitative and structured methods, combined with the information in the prompt words, to provide fundamental analysis for the investment value analysis of this industry at {{time}}}.

## Skill

1. Conduct quantitative analysis on each dimension, such as analyzing the current year-on-year and month-on-month changes in specific data, policies, etc., to support analytical views.

3. Present the analysis framework in a structured manner, such as using a general-specific-general structure.

3. Provide an analytical framework in a clear and structured form, and clearly and logically present the investment value fundamentals of the industry.

## limit:

- The output content must comply with quantitative and structured requirements, with rigorous and standardized language and cannot deviate from the framework requirements.

User prompt words:

The following is the relevant fundamental data analysis and corresponding historical data of the {{topic}} industry. Combined with the information in the materials and based on the valid information in the materials, please sort the data into different analysis dimensions, and then give the industry's prosperity in different dimensions a score of 1-5 from low to high based on the classified data.

The criterion for scoring 1 is that the current data or qualitative content is at the worst level in history.

The criterion for scoring 2 points is that the current data or qualitative content is at a historical deviation position.

The standard for scoring 3 points is that the current data or qualitative content is in the middle of the historical level.

The standard for scoring 4 points is that the current data or qualitative content is at a relatively better position than before.

The standard for scoring 5 points is that the current data or qualitative content is at the best historical level.

Combine the scoring results and analysis of each dimension to give the industry a comprehensive score and analyze the business prosperity of the industry at this time.

Related information:

——

Output format example: (omitted)

Note: input is the final summary data result of the first cycle, input2 is the final summary data result of the second mirror cycle; time is the time interval that needs to be analyzed input by the initial node; topic is the corresponding industry that needs to be analyzed.

Overall evaluation and future prospects

We have built an AI Agent with preset steps to apply to industry prosperity analysis. The specific working mode is that after AI builds an analysis framework for a specific industry, it uses AI's ability to quickly sort through large amounts of text to quickly screen and sort out valid information on the Internet. Finally, it compares and analyzes the fundamental data at the corresponding time point and the historical data of the same caliber to derive the prosperity scores of industries in each dimension, and then obtain a comprehensive score for the industry.

The AI Agent workflow we designed can provide reference for investors in two specific scenarios in practice.

Assisting desk research: AI workflows can efficiently collect and sort out information and data related to the business climate analysis of specific industries across the entire network. Therefore, the output analysis reports can provide investors with a reference at the research level: effectively reducing the time researchers spend searching for information; investors who prefer top-down analysis can also use AI workflows to quickly understand the business climate of specific industries.

Build a high-prosperity investment portfolio: High-prosperity industries/individual stocks have weak uncertainty because they are supported by performance. The AI workflow can quickly obtain the monthly and quarterly prosperity scores of specific industries, which can effectively screen the high-prosperity industries in the current market and assist investors in building a high-prosperity investment portfolio.

We believe that AI Agent’s completion of this type of industry analysis task embodies the characteristic of “speed”: manual research may be more detailed and comprehensive, but it is more time-consuming. The overall working time of the AI workflow we designed is about 5-10 minutes (corresponding to 5-8 analysis factors), which can quickly get a rough evaluation of the prosperity of a certain industry; in addition, AI Agent obtains analysis results by collecting a large amount of relevant information and news on the Internet, and the output report sorts out the relevant factors that affect the prosperity of the industry. Therefore, the materials output by AI can also help investors complete desk research on the industry.

However, there are still certain flaws in using AI Agent for industry analysis: first, the design of the industry analysis framework by AI Agent may not be as accurate as that of the corresponding researchers. Therefore, when collecting data and scoring according to the relevant analysis framework, the final result may deviate from the actual situation. Second, AI Agent only analyzes based on the content that can be collected by each search engine. If some public information is not recorded in the search results in a form that AI can understand and read, AI may ignore the information. Finally, there is still room for improvement in the intelligence of AI investment research tasks, and the application and reasoning of some data and information may be weaker than that of professional investors.

We believe that the optimization of this workflow can be carried out from two aspects: on the one hand, the data source needs to be optimized. The data sources of some workflow platforms have not yet been connected, such as the lack of data sources of listed company announcements, research reports and related industry databases/financial databases, and the lack of information sources from overseas websites, public accounts, etc. If the AI Agent can collect more complete data sources, the final industry analysis results will be more accurate; secondly, it is necessary to upgrade computing power and use more powerful large models. AI Agent can complete the tasks from collecting data to analyzing comprehensive prosperity faster and more accurately.

Overall, although there are some points that need to be optimized, we believe that the analysis results of this version of AI Agent are basically of certain reference significance and can assist investors to a certain extent in completing industry analysis research tasks and investment decisions related to industry rotation. Investors can also quickly construct fundamental factors based on AI's quantitative scores to assist in quantitative investment.

Risk Warning

AI analysis has a certain degree of randomness. The same input parameters of a large model may output different results each time, so the production content has a certain degree of randomness.

AI hallucination risk. The intelligence level of AI big models needs to be improved. Due to data availability and other reasons, some analysis processes may have AI hallucinations. This article is only a method for AI investment research workflow and does not constitute any investment advice.