# Industry Funnel vs Industry Research Filtering Approaches in ai-berkshire

> Understand industry funnel vs industry research filtering in ai-berkshire. Learn how to narrow market universes or construct industry panoramas for AI-driven analysis.

- Repository: [Xbt Lin/ai-berkshire](https://github.com/xbtlin/ai-berkshire)
- Tags: deep-dive
- Published: 2026-07-26

---

**Industry funnel narrows a full-market universe down to 3-5 hand-picked companies through layered quantitative screens, while industry research constructs a complete industry-chain panorama evaluating every listed player across segments before any narrowing occurs.**

The `xbtlin/ai-berkshire` repository provides two distinct systematic methodologies for equity analysis. Both approaches guide investors through structured sector evaluation, but they follow opposing filtering philosophies—one company-centric and selective, the other chain-centric and comprehensive.

## Core Distinction: Narrowing Versus Mapping

The fundamental difference lies in the direction of analysis. **Industry funnel** operates as a **subtractive filter**, progressively eliminating candidates through hard quantitative screens until only the strongest remain. **Industry research** operates as an **additive mapper**, first constructing the complete competitive landscape across upstream, midstream, and downstream segments, then classifying every participant.

According to the source code in [`skills/industry-funnel.md`](https://github.com/xbtlin/ai-berkshire/blob/main/skills/industry-funnel.md) and [`skills/industry-research.md`](https://github.com/xbtlin/ai-berkshire/blob/main/skills/industry-research.md), these approaches are designed to be complementary: research defines the investment arena, while the funnel picks the specific players.

## The Industry Funnel Approach

### Goal and Philosophy

The industry funnel targets **company-level selection**. Its primary objective is to digest a full-market universe of active stocks and output a concise shortlist of 3-5 targets suitable for deep investment analysis and portfolio construction.

### Four-Stage Filtering Mechanics

As implemented in `skills/industry-funnel.md#L13-L33`, the funnel executes a rigid four-stage progression:

1. **Full-market scan** – Screen active stocks by momentum and market-cap to generate 30-60 candidate firms.
2. **Hard-value criteria** – Apply five quantitative filters (PE, ROE, cash-flow, debt-ratio, and **moat** (护城河) assessment) to reduce the list to ≤10 firms.
3. **Granular analysis** – Produce 300-500 word write-ups for remaining candidates, maintaining ≤10 firms.
4. **Four-master deep dive** – Conduct qualitative analysis to finalize a 3-company shortlist.

Each layer documents elimination rationale, creating an audit trail from broad universe to final selection.

### Output and Deliverables

The approach generates a final report limited to **3-5 "target" stocks** complete with investment theses, position sizing, and risk notes. This output suits analysts needing a concise shortlist for immediate portfolio construction.

## The Industry Research Approach

### Goal and Philosophy

Industry research pursues **industry-chain mapping** across A-share, HK, US, and international markets. Rather than filtering down, it builds a **complete panorama** to produce a sector-level investment portfolio covering all relevant sub-segments.

### Chain-Centric Analysis Flow

The methodology in `skills/industry-research.md#L16-L30` follows a distinct four-step workflow:

1. **Logic-chain construction** – Map trend → demand → bottleneck → sector relationships.
2. **Full-chain scan** – Evaluate all listed companies across global markets.
3. **Tiered classification** – Sort firms into **Tier 1-4** categories based on market-cap, purity of exposure, and segment leadership.
4. **Four-master analysis** – Apply deep qualitative assessment to Tier 1-2 firms per segment.

### Output and Deliverables

This approach yields a **comprehensive industry-chain report** containing segment tables, tiered company listings, and overall portfolio allocation guidelines (core, satellite, option, and ETF substitutes). It serves analysts requiring structural understanding of entire sectors before security selection.

## Practical Code Examples

Execute these analyses through the repository's CLI or bot framework. The following markdown commands trigger the respective workflows:

```markdown

# Run industry funnel for AI computing sector

/industry-funnel AI算力

```

```markdown

# Run full industry research on innovative drugs

/industry-research 创新药

```

For programmatic execution using the internal CLI wrapper, use Python's subprocess module:

```python
import subprocess

# Execute funnel analysis for robotics sector

subprocess.run(["python3", "scripts/run_skill.py", 
                "industry-funnel", "--arg", "机器人"])

# Execute research analysis for nuclear power sector  

subprocess.run(["python3", "scripts/run_skill.py", 
                "industry-research", "--arg", "核电"])

```

Both commands generate timestamped markdown reports under the `reports/` directory (e.g., `reports/机器人-funnel-20260707.md`). Before publication, [`tools/report_audit.py`](https://github.com/xbtlin/ai-berkshire/blob/main/tools/report_audit.py) provides the data-verification pipeline used by both approaches to validate findings.

## When to Use Each Approach

| Aspect | Industry Funnel | Industry Research |
|--------|----------------|-------------------|
| **Primary Goal** | Narrow to 3-5 high-conviction stocks | Build complete sector panorama |
| **Starting Point** | Full-market universe | Industry-chain logic construction |
| **Key Filter** | 5 hard-value criteria + moat assessment | Tier 1-4 classification by exposure purity |
| **Typical Output** | Target shortlist with position sizing | Segment tables with portfolio allocation |
| **Source Files** | [`skills/industry-funnel.md`](https://github.com/xbtlin/ai-berkshire/blob/main/skills/industry-funnel.md), [`codex-prompts/industry-funnel.md`](https://github.com/xbtlin/ai-berkshire/blob/main/codex-prompts/industry-funnel.md) | [`skills/industry-research.md`](https://github.com/xbtlin/ai-berkshire/blob/main/skills/industry-research.md), [`codex-prompts/industry-research.md`](https://github.com/xbtlin/ai-berkshire/blob/main/codex-prompts/industry-research.md) |

Use **industry funnel** when you need a concise shortlist for immediate portfolio construction. Use **industry research** when you must understand structural drivers across upstream, midstream, and downstream segments before allocating capital.

## Summary

- **Industry funnel** applies a company-centric, multi-stage filter (`skills/industry-funnel.md#L13-L33`) that trims the universe to 3-5 stocks through quantitative screens and four-master analysis.
- **Industry research** employs a chain-centric, comprehensive mapping (`skills/industry-research.md#L16-L30`) that evaluates entire sectors and classifies firms into Tier 1-4 categories before selection.
- Both approaches utilize [`tools/report_audit.py`](https://github.com/xbtlin/ai-berkshire/blob/main/tools/report_audit.py) for data verification and generate timestamped reports in the `reports/` directory.
- The methodologies are complementary: run research first to define the arena, then apply the funnel to pick specific players.

## Frequently Asked Questions

### Should I run industry research before using the industry funnel?

While both can operate independently, they are designed to work sequentially. Running **industry research** first ensures your subsequent **industry funnel** analysis covers all relevant sub-segments within a sector, preventing omission of niche but critical players in the upstream or downstream chains.

### What are the five hard-value criteria used in the industry funnel?

According to [`skills/industry-funnel.md`](https://github.com/xbtlin/ai-berkshire/blob/main/skills/industry-funnel.md), the five quantitative screens are: **PE ratio**, **ROE**, **cash-flow metrics**, **debt-ratio**, and **moat** (护城河) assessment. These filters reduce the initial 30-60 candidates to 10 or fewer firms before qualitative analysis begins.

### How does industry research classify companies into tiers?

The **industry research** approach in [`skills/industry-research.md`](https://github.com/xbtlin/ai-berkshire/blob/main/skills/industry-research.md) uses a **Tier 1-4 classification** system based on three factors: market capitalization, purity of segment exposure, and leadership position within the specific upstream, midstream, or downstream segment. Tier 1-2 firms receive full four-master analysis.

### Which source file contains the report verification tools?

Both approaches rely on [`tools/report_audit.py`](https://github.com/xbtlin/ai-berkshire/blob/main/tools/report_audit.py) for the data-verification pipeline. Additionally, [`codex-prompts/industry-funnel.md`](https://github.com/xbtlin/ai-berkshire/blob/main/codex-prompts/industry-funnel.md) and [`codex-prompts/industry-research.md`](https://github.com/xbtlin/ai-berkshire/blob/main/codex-prompts/industry-research.md) provide auto-generated prompt wrappers that ensure Codex compatibility when generating analysis reports.