Deep Thinking: How to Use AI to Improve the Efficiency of "Product Research"?

Now, combined with AI tools, we can spend relatively little time to efficiently write a relatively in-depth product research analysis, how to do it?

Written in front of the words

Last time, Sanbai spent nearly 2 weeks writing a product analysis of "In-depth Analysis of National Office Software WPS", and received a lot of praise and feedback from fans of the official account, and many fans also consulted me about how to systematically do product research related topics.

1. Abstract of the article

In this article, based on my past practice and experience, we will focus on sharing with you how I quickly built an analytical framework for product research, and comprehensively used various AI creation tools, AI search tools, At the same time, I will also share how Sanbai summarized the experience and methods and productized the entire process through self-developed AI efficiency tools to further accelerate the efficiency of research output;

2. Who is it suitable for?

1. Product managers and market researchers, applied to competitive product research and competitive product analysis scenarios;
2. Knowledge paid content creators, applied to the output of high-quality product research and analysis content;
3. Job seekers, including office white-collar workers and college students, should be used to prepare for interviews before applying for a job;
4. Financial researchers, product research and analysis in financial scenarios;

3. What will you gain?

1. A set of standard processes and ideas for output product research and analysis;
2. A summary of several AI tools and data retrieval platforms that can effectively improve research efficiency;
3. Sanbai's thinking process and practice of productizing the entire research workflow through AI;

1. The process of outputting a product research report

First of all, let's summarize the process of writing a product research report, which includes the following 6 steps:

1. Clarify the research product and research objectives;
2. Construct the ideological framework and research outline of product research;
3. Retrieving research materials and establishing a knowledge base for product research;
4. Read the data and organize the information, and output the preliminary research report
5. Drill carefully and analyze, verify information, remove the false and retain the true;
6. Integrate personal original and first-hand research to form an in-depth analysis report;

Next, I will try to take the research topic of "ChatGPT Product Research and Analysis" as an example to discuss how to do each of the above links and what efficiency tools can be used.

Because the first step is often a precondition for determining the investigational product and research objectives, let's start directly with the second step.

2. Construct the ideological framework and research outline of product research

Tool Recommendation: AI Quick Researcher (www.kuaiyanai.com)

1. Summary of the product research outline framework

Sanbai combined with the past personal work experience, as well as the reference to learn most of the product managers of the goose factory to write the ideas and habits of competitive product research and analysis, summarized and output the most common twelve research topics in the product research analysis report, and built it into a general product research framework, as shown in the following brain diagram, for most of the primary product researchers, basically through the following framework to quickly establish a product research ideas, Sanbai is currently for all product research ideas, but also basically refer to this framework, share it for everyone's reference and discussion。

2. How to quickly output a research outline framework for a new product?

Pain points of personalized outline output:

Even though there is already a ready-made research framework above, Sanbai found that every time a new product is researched, it takes about 3~4 hours to apply the above research framework to output a personalized research outline (instead of a generalized general framework), which is quite a headache, and the efficiency is not necessarily improved;

Make your own AI-generated product research outline product:

In view of this, Sanbai himself thought of a way to productize the above research framework, and precipitated the rules and knowledge of the entire outline writing, and then combined with the technical capabilities of the AI large model, he developed a tool product "AI Quick Research Man (www.kuaiyanai.com)" to quickly generate product research outlines, and after nearly a month of continuous polishing and tuning, the current Fast Research Man can basically be in 3~In about 5 minutes, I quickly wrote a research outline that was basically the same as what I wrote, which greatly improved my efficiency and shortened the creation time of at least 3~4 hours for me to write an average product analysis;

Try a specific example:

Next, we take the research topic of "ChatGPT Product Research and Analysis" as an example, and try to use AI Quick Research to generate a research outline that benchmarks the above research framework:

The following example picture shows the content of the outline actually generated by AI, you can see that the outline is basically output according to the framework I organized, and the outline is also refined to the third level of outline, and the content of the outline is not suitable for a general generalized content, but around the personalized content of the ChatGPT product;

3. Retrieving research materials and establishing a knowledge base for product research

Tool recommendation: Kimi, Secret Tower Search, Everyone is a Product Manager, Research Report Platform, Data Platform, etc.;

Avoid reinventing the wheel:

After establishing a good research outline, the next step is to output specific content around the outline, usually there is more content in the research outline is actually already a topic with research results, so first of all, we do not need to repeat the wheel, but more to play the role of knowledge sorting and summarization, the preliminary research information is sorted out, this process first we need to collect as much research data as possible around the research theme, including research reports, product analysis, user research reports, etc.

Search tools for common references:

Finding a high-quality reference usually consumes a lot of time and energy from users, so Sanbai combines his own habits and experience to organize a platform collection for quick retrieval of reference materials, which is convenient for everyone to quickly collect reference materials, as follows:

1. AI search tools: Recently, Sanbai used more search tools are kimi, secret tower search, Tiangong and other search tools, in terms of reducing advertising and direct search results, these products are indeed better than traditional search engines such as Baidu and Google, but there are also problems such as incomplete search results and low timeliness;
2. Product manager community: The second is obtained from the product manager community, I have to say that everyone has contributed more product research and analysis in the product manager community, and you can search for the corresponding product research materials in it;
3. Research report platforms, brokers, securities companies, and consulting institutions: These public research report retrieval platforms usually provide a lot of research reports, some of which are paid and some are free, and you can also try them on these platforms;
4. Data statistics platform: You can obtain the data published by the National Bureau of Statistics on platforms such as the National Statistics Network, which will generally have a higher credibility, and then in the third-party data service platform, individuals recognize the data on Questmobile, which is relatively high in terms of accuracy;
5. Company financial report platform: You can directly download the financial report of the designated company on the snowball, flush and other platforms, generally if the company has been listed, I will basically give priority to the company's financial report, because the financial report information is a very high-quality reference.

A detailed list of the various types of search tools is as follows:

Fourth, the data reading and information collation, the output of preliminary research reports

Tool Recommendation: AI Quick Researcher (www.kuaiyanai.com)

1. It takes a long time to read a lot of reference materials

In the previous step, we may be able to retrieve more than a dozen relevant reference materials, but it is not easy to read each material in detail and collect and summarize the information within the scope of the research outline, and the following problems are usually encountered in this link:

1. You don't know which report will be in which report a particular research outline will be, which means you need to go through all the documents;
2. Reading a large number of documents itself will consume a lot of your time, and after reading, you also need to take notes to organize the information, which also takes a lot of effort here;
3. Based on your notes, a basic research report is compiled and re-written.

Taking the creation of the product analysis of "In-depth Analysis of National Office Software WPS" as an example, basically in this link, Sanbai spent nearly a week or so, and a lot of time was basically spent here, so it was very time-consuming to output a relatively high-quality analysis report, which is also the main reason why Sanbai's iteration speed of each content output is very slow.

2. Try to solve it with AI tools like ChatGPT and Kimi

In order to improve the efficiency of this link, at the beginning, Sanbai tried to use ChatGPT, kimi and other products to ask questions one by one by uploading more than one document, and then asking questions one by one through multiple conversations and Q&A, so that AI could help me extract the information in the references, but the results were not satisfactory, and the main problems were as follows:

1. Products such as ChatGPT and Kimi do not support batch Q&A, and do not support answering many questions at once, once there are too many questions, it will lead to a decrease in the quality of the generated content, or you can't get the desired results, which means that if I follow the outline of the example to ask one sub-question at a time, it will take at least hundreds of conversations and Q&A to organize the entire research outline, which is very inefficient;
2. Second, even though I did have hundreds of Q&A conversations, I still needed to integrate the fragmented results into information before I could produce a first draft of the study, which didn't save much time and was still a laborious manual work.

Therefore, using general AI products on the market cannot solve the problems mentioned above.

3. Let the AI help you read the reference materials and organize them to generate a research report

After an uncomfortable period of time, Sanbai also decided to continue the previous process of AI generating research outlines, further supporting the completion of batch dialogue reading and information sorting through AI, and automating the processing of things performed manually through AI, because in essence, I only need to find a way to let the model complete hundreds of dialogue questions and answers in batches at one time, and sort out the results of dialogue questions and answers through the model, I can complete the above manual operations, and it seems that it is not impossible to achieve, so I pulled my R&D team to evaluate and start to implement;

After more than 2 months of polishing and tuning, I have overcome various abnormal problems in the middle, and finally realized the ability to let AI help me extract relevant information from reference materials in batches according to the research outline, and organize and output them into documents in a natural language way; 。

4. Try a specific case

Next, let's also take "ChatGPT product research and analysis" as an example to see how effective AI is in generating research reports based on reference documents.

First of all, we will clarify the scope of the research based on the research outline that has been established before, and then start to upload the reference materials related to the product, which means that I will let the model refer to and learn these documents to obtain a certain amount of knowledge;

Then, click "Intelligent Generate Report" to start creation based on references, after about 3~5 minutes, the large model can generate a research report of tens of thousands of words, referring to the following example, AI generated a research report with a word count of up to 96,000 words, so far a preliminary research report is basically completed;

We can test the quality of its generation? Extract some of the fragments, such as the product introduction of ChatGPT, including the type of product, online time, company valuation and other information, as well as the technical basis, and its description is basically accurate and true, because there are relatively complete reference materials, so the large model can create relatively accurate, real, high-quality content;

Of course, for some topics that are not covered by reference documents or the model has not yet been learned, the large model will try to create its own content, but sometimes the content it creates is not necessarily true and can only be used as a reference, but we believe that with the continuous update of the model's knowledge base and data corpus, and the continuous enhancement of the model's capabilities, this situation will be greatly reduced in the future.

Fifth, drill and analyze, verify information, remove the false and retain the true

Next, of course, we can't fully trust the authenticity of AI-generated content, because of the hallucination phenomenon of large models, sometimes what it writes may be nonsense, so we have to try to find a way to identify it as much as possible, based on this problem, I thought of the following methods:

1. Support AI-generated content reference source traceability

For each fragment generated by AI, we clearly mark the reference source of the generated content, mark each paragraph with a reference source with a red index mark, click the index mark, you can view the summary of the original fragment and locate which document is specifically referenced, and even I will support locating the page in the future;

2. Distinguish between content created by the model based on reference materials and content created by the model itself

Paragraphs with index marks mean that the content of the paragraph is created by the model based on your real reference materials, which may be higher in authenticity and accuracy, while those without indexes are the content created by the model itself, and we have no way to guarantee that the content created by the model itself must be authentic and reliable, but we can first find a way for users to distinguish which ones are written by the model itself.

3. Drill down and dig deeper

At this time, you can consider using kimi and chatgpt to do fragmented dialogue Q&A, this process, you don't need to use cumbersome use, you only need to spot check and confirm a few key questions, and the amount of conversation is far lower than the amount of conversation in the previous implementation of a complete outline content.

6. Integrate personal original and first-hand research to form an in-depth analysis report

The content generated earlier, basically can only be said to have completed a basic research quickly, although it has been possible to input a relatively good research result, but the problem is that most of it is ready-made or other people's research results; lack of personal first-hand research output and original content, so the next thing to do, is to supplement personal thinking and opinions, this part needs to rely on personal professionalism and vision, at present, this part needs to start by yourself, AI has no way to replace you, but only by supplementing your own first-hand research and original views, can you form an attitude, opinionated, and in-depth analysis reports.

At this point, after the first 6 steps, a relatively mature and in-depth product research report has been basically completed, and then let me take a look at the "ChatGPT Product Research Analysis" I output according to the above steps The effect of the original content of the case can be found in the document link below (click to view it directly), because the personal original content is relatively small, and the current output of this version of the content is relatively rough, but basically it has been able to give yourself enough information, which is basically enough for a person who has a preliminary understanding of the ChatGPT product;

ChatGPT Product In-Depth Analysis Report: https://kdocs.cn/l/cd5nakvQASsq

7. How much time has the combination of AI tools saved me?

Sanbai spent nearly half a year or so, polishing himself an AI tool to improve the efficiency of his creation and research, before the output of an analysis, it took 2 weeks ~ 1 month, now basically, with the support of the tool, I can basically quickly form a preliminary study of tens of thousands of words within 1~2 hours, and it is expected to output a relatively in-depth product analysis within 2~3 days, which not only improves my output efficiency by 7~10 times, but also allows me to output more content;

The results of these analyses also help me to provide high-quality research results and services to my clients and users.

Of course, Sanbai always keep in mind that AI itself is an efficiency tool, I don't agree with using AI as a lazy tool, outputting some shoddy content to pollute the environment of knowledge payment, content creators should maintain the original intention of outputting high-quality content, and treat AI only as an efficiency tool, rather than replacing their own body and brain, personal original views and output, is still the most important, in the future, Sanbai will not use AI tools to do batch washing and production of hydrology.

OK, that's all for this sharing, and in the next article, I will share some of my personal methodologies and experiences around how to do industry research and company research, thank my users for reading this.

Author: Sanbai has something to say, public account: Sanbai has something to say

This article was originally published by @三白有话说 on Everyone is a Product Manager. Reproduction without the permission of the author is prohibited.

The title image is from Unsplash and is licensed under CC0.

The views in this article only represent the author's own, everyone is a product manager, and the platform only provides information storage space services.