When the enterprise repairs the “highway” (data platform), the “car” (data) on the road is getting more and more chaotic – the business personnel have become tired “traffic police”. This article proposes to use a conversational AI product, Digital Intelligence Staff, to equip each employee with an “intelligent co-pilot” 24×7 hours a day: natural language for question-and-answer, RAG real-time integration of business knowledge, one-click attribution, and push insights.
Digital transformation is a key step from the “infrastructure” stage to the “value realization” stage.
Just like we have repaired the “highway” (data platform), but now there are too many “cars” (data) on the road, and business personnel have become tired “traffic police” rather than calm “drivers”. Our goal is to equip every business person with an AI “intelligent navigation” and “driving assistant”.
Below, it is convenient to share my ideas from the pain points of the product, product construction ideas and product value.
Part 1: Business pain point analysis
As a digital service customer service product for enterprises (we can call it”Digital intelligence advisorWe start by deeply understanding our “customers”, business people, and the core pain points they encounter when using existing data platforms:
1) Cognitive overload (information overload, low signal-to-noise ratio) Pain point description:Business personnel are drowned in massive data reports, indicators, and dimensions. They need to spend a lot of time “finding” and “right” data instead of “using” it. Faced with a complex BI report, they often don’t know where to start and which indicator fluctuations are the most worth paying attention to.
Business scenarios:When he finds that the overall sales are not up to standard, he needs to check which region, which product line, and which channel are problematic one by one, which process is time-consuming and labor-intensive, and it is easy to miss key information.
2) Description: Pain points with high analysis threshold (complex tools, experience dependence):In-depth data analysis often requires mastering SQL, Python, or complex BI tool operations. Business personnel often do not have these specialized skills. More importantly, effective analysis relies on “business feel” or “expert experience”, and this tacit knowledge is difficult to replicate and pass on, resulting in the solidification of analytical skills in a few experts.
B-end product manager’s ability model and learning improvement
The first challenge faced by B-end product managers is how to correctly analyze and diagnose business problems. This is also the most difficult part, product design knowledge is basically not helpful for this part of the work, if you want to do a good job in business analysis and diagnosis, you must have a solid …
View details >
Business scenarios:Xiao Li of the marketing department wanted to analyze the effect of the “Double Eleven” promotional activities, she not only needed to pull the sales, traffic, and conversion rate data during the event, but also needed to compare it with the data of the same period and the inactive period in history, and even combined with user portraits, channel sources, coupon usage, etc. for multi-dimensional drilling. This process is extremely high for non-data analysts.
3) Insight into the pain points of lag (passive response, missed opportunities) Description:The existing data presentation methods are mostly “passive”. Business personnel must take the initiative to inquire and find problems. When problems are discovered, they often have been happening for some time and may have missed the best window of intervention.
Business scenarios:The user activity of a product has declined for three consecutive days, and this trend was only discovered when the product manager made a weekly report on Friday, and the problem may be a small bug that went live on Monday, which has affected the user experience for nearly a week.
4) Difficulty in attribution and inability to make decisions (knowing what it is, not knowing why) Pain point description:Even if the salesperson discovers the “phenomenon” of “sales drop by 10%” through the report, what is the “cause” that leads to this phenomenon? Is it a competitive event? Is it a channel problem? Is it the weather? Or is the quality of marketing materials declining? Existing tools cannot correlate business logic with data to provide convincing attribution explanations, leading to subsequent decisions that can be slapped on the head.
Business scenarios:Management saw a decline in profits on the financial report, and the data platform showed that it was caused by the rise in the cost of product line A. But why are costs rising? Is it the increase in the purchase price of raw materials, or the increase in energy consumption of the production line? This requires a combination of business knowledge from multiple links such as supply chain and production to explain.
Summary:The core contradiction is thatThere is a huge gap between massive, standardized data and personalized and scenario-based business decision-making needs。 Our products are designed to fill this gap.
Part 2: Product construction ideas and documentation
Based on the above pain points, we designed a product called “AIBusinessCopilot“.
1. Product construction ideas
Positioning:From Data Tools to Decision Partners. It does not replace BI reports, but provides a conversational, intelligent, and business-based analysis portal on top of BI.
Core Technology:large language models (LLMs) are used as the “brain” responsible for understanding, reasoning, and generation; RAG (Retrieval-Augmented Generation) is used as the “outer brain” to provide accurate and real-time private domain data and business knowledge.
Construction path:
- Step 1 (Connect):Open up the underlying data platform to realize the query and call of global data.
- Step 2 (Knowledge):Build an enterprise knowledge base to vectorize unstructured business experience (such as analysis reports, SOPs, market insights, meeting minutes) to provide “ammunition” for RAG.
- Step 3 (Intelligence):Develop core AI capabilities, including natural language query, intelligent attribution, trend prediction, report generation, and more.
- Step 4 (Scenario):Create templated, one-click analysis applications for high-frequency, high-value business scenarios (such as sales review and marketing campaign analysis).
“AIBusinessCopilot” Product Requirements Document (PRD) V1.0
2. Product vision and goals
Product Vision:Let every business person have a 24×7 online AI data analysis expert who understands business, transforming data analysis from a heavy task into a simple conversation, truly realizing data-driven decision-making and empowering business efficiency.
Product Objectives:
- Business Objectives:reduce the average time it takes for business personnel to find the cause by more than 50%; Increase the adoption rate of data-driven decision-making by 30%.
- User Objectives:80% of daily data query and analysis needs are completed through “digital intelligence staff”, and the user satisfaction (NPS) reaches +50.
- Strategic Objectives:Precipitate and scale the data analysis capabilities and business knowledge of enterprises, and build a core moat for digital transformation.
3. Target users and usage scenarios
4. Core Feature Design
4.1 Natural language interactive query (ConversationalQuery)
Function description:Provides a chatbot-like interface where users can ask questions in natural language (Chinese). The system can understand complex, colloquial query intent, including multiple rounds of conversation, referential elimination, intent questioning, etc.
Key points of implementation:
- NL2SQL/NL2API：Turn natural language into precise query statements (SQL) to your data warehouse or calls to APIs.
- Intent recognition and slot filling:Accurately identify the core intent of a user query (e.g., “query”, “comparison”, “attribution”) and key entities (e.g., time, region, product).
- Contextual Memory:Maintain contextual understanding during multiple rounds of conversation, for example, users first ask “check last week’s sales” and then “What about Beijing?” The system should be understood as “check the sales in Beijing last week”.
4.2 RAG-Powered Contextual Analysis Function Description:When AI conducts data analysis, it can automatically retrieve and integrate the internal business knowledge base of the enterprise, providing “business common sense” and “historical experience” to the data, making the analysis results more in-depth and credible.
Key points of implementation:
1) Enterprise knowledge base construction:
- Data Sources:Unstructured and semi-structured documents such as historical analysis reports, business SOPs, market research reports, excellent review PPTs, industry information, financial announcements, and product documents.
- Technology:The above documents are regularly extracted and cleaned by ETL tools, and they are vectorized and stored in the vector database using the embedding model.
2) Search and generation:
- When a user asks a question (e.g., “Analyze the reasons for the decline in sales”), the system not only queries the sales data, but also retrieves historical reports and SOPs related to “Sales Decline Analysis” in the vector database.
- LLMs integrate real-time data with retrieved knowledge fragments to generate answers that blend “data facts” and “business experience.” For example: “Sales in East China fell by 15%, mainly due to Nanjing contributing 80% of the decline.”Based on our past experience, the Nanjing market is highly sensitive to the promotion of competing product A. Retrieved that competitor A was conducting a buy-one-get-one-free activity in Nanjing last week (Source: Marketing Department Competitive Product Monitoring Weekly Report), which is likely to be the main reason. ”
4.3 IntelligentRootCauseAnalysis & Insights
Function description:It can automatically drill down and correlation analysis on the indicators that users are concerned about, explore possible drivers, and present them to users in a clear logical chain.
Key points of implementation:
- Attribution tree analysis:Automatically break down core metrics (e.g., “profit”) into drivers (profit = revenue – cost; Revenue = sales x unit price… ), drill down layer by layer to locate the leaf node with the greatest contribution.
- Correlation vs. causal inference:Combined with statistical models (such as correlation analysis, Granger causality test, etc.), the potential correlation between indicators is discovered, and interpreted by LLM combined with business knowledge.
4.4 Multi-modalReportGeneration&Push
Function description:According to user instructions, it can generate comprehensive analysis reports including data charts, text conclusions, and insight summaries with one click, and support exporting to PPT, PDF, Word and other formats. At the same time, monitoring tasks can be configured to proactively push abnormal fluctuations and opportunity insights.
Key points of implementation:
- Automatic Chart Generation: Intelligently recommends and generates the most suitable charts (such as line charts, bar charts, scatter charts) based on data types and analysis purposes.
- Report Templates:Built-in a variety of commonly used analysis report templates (such as monthly business analysis and activity review report) can be applied by users with one click.
- Proactive push:Based on preset KPI thresholds or abnormal patterns discovered by AI, early warnings and briefings are sent to relevant responsible persons through channels such as DingTalk, Feishu, and Enterprise WeChat.
4.5 Explainability & Traceability
Function description: In order to build user trust, all analysis conclusions must be traceable. Users can view the data sources, calculation logic, and business knowledge documents referenced by the conclusion.Key points of implementation:
- Data traceability:Each chart and data point is clickable to display its source data table, fields, and filters.
- Logical traceability:Show the Chain-of-Thought that AI generated the conclusion, as well as the link to the original RAG knowledge fragment cited.
5. Brief description of technical architecture
Access layer:Support various clients such as web, mobile, and IM tools (DingTalk/Feishu).
Application layer:The back-end service of “Digital Intelligence Staff” includes modules such as user management, dialogue management, and task scheduling.
AI Capability Layer:
- LLM Engine:Core large language models can be privatized with open-source models (such as Llama) or call commercial APIs (such as GPT-4ERNIEBot).
- RAG Module:Includes document processing pipeline, vector database (VectorDB), and retriever (Retriever).
- Data Analysis Engine:Perform tasks such as SQL queries, statistical model calculations, and machine learning predictions.
Data and Knowledge Layer:
- Structured Data:Enterprise data warehouse/data lake.
- Unstructured Knowledge Base:A vector database that stores business documents and their vector indexes.
6. Roadmap
Phase 1 (MVP, 3-6 months):
- Focus on core scenarios: Select 1-2 business units (such as sales) as pilots.
- Core functions: Implement natural language-based single-round/simple multi-round queries, integrate RAG, and perform attribution analysis on core KPIs.
- Knowledge Base: Manually import the first batch of high-quality analysis reports and SOPs.
Phase 2 (Functional Improvement and Promotion, 6-9 months):
- Expansion scenarios: Promote to more departments such as markets and products.
- Feature enhancements: Develop one-click generation of multimodal reports and active insight push functions.
- Knowledge base automation: Establish a mechanism for automatic knowledge base updates.
Stage 3 (platformization and deepening, 9-12 months):
- Open capabilities: Provide API interfaces so that other business systems can also call the analysis capabilities of the “Digital Intelligence Staff”.
- Personalization: Allows users to customize the metrics they follow, report templates, and knowledge feeds.
- Prediction and Advice: From “Explaining the Past” to “Predicting the Future”, providing forward-looking decision-making suggestions.
Part 3: Product value refinement
The core value of the “Digital Intelligence Staff” product is to promote the digital transformation of enterprises fromHe gave the fish to the people“(Provide data reports) to “Teach people to fish, and pair them with AI fishermen“A fundamental shift in “providing analytical capabilities and decision-making partners.”
1. Value to business personnel: empowerment and burden reduction
- Lowering the Barrier to Analysis:“Inclusive” complex data analysis capabilities, so that ordinary business personnel who do not understand SQL or Python can become data analysis masters, and realize that “everyone is a data analyst”.
- Increased productivity:Free business personnel from tedious, repetitive data retrieval and organization, allowing them to focus on business strategy thinking and execution, and multiply their work efficiency.
- Improve decision-making quality:It no longer provides cold data, but integrates business knowledge, in-depth and warm insights and suggestions to help business personnel make more accurate and timely decisions.
2. Value to the enterprise: efficiency increase and precipitation
- Accelerating the Decision Closure:Dramatically shorten the cycle from “data generation” to “insight discovery” to “business action”, allowing enterprises to respond faster to market changes and seize fleeting business opportunities.
- Precipitate tacit knowledge:The analysis ideas and business experience of experts are solidified in the form of RAG knowledge base, and turned into digital assets that can be reused and inherited by enterprises, breaking down knowledge barriers and improving the “data IQ” of the organization as a whole.
- Maximize the value of your data assets:Completely revitalize the massive data dormant in the data platform, transform it into a powerful driver of business growth, and ensure that enterprises get real returns on their huge investments in digital transformation.
Ultimately, this product will become a key symbol of digital transformation success. It is not only a tool, but also one of the core competitiveness of enterprises in the digital age, and the “intelligent center” of the enterprise brain, ensuring that enterprises can always see, think clearly and act quickly in the complex and changeable market environment.
