Real-World OpenClaw Use Case: Automated Business Vetting

What is Business Vetting and Why Does It Matter?

Business vetting is the process of evaluating a company before you commit to working with them. It happens in several contexts: before signing a vendor contract, before accepting an investor, before acquiring a company, before onboarding a B2B customer, or before entering a partnership. The goal is to verify that the business is legitimate, financially stable, reputable, and aligned with your requirements.

A typical manual vetting process involves visiting the company's website, reading their About page, checking their social media accounts, searching for customer reviews on Google, Trustpilot, and G2, looking up their legal filings, estimating their revenue and employee count, and reading any news articles about them. For thorough vetting, you might also check their domain registration history, look for lawsuits, verify their claimed partnerships, and assess their technology stack.

This process takes 4 to 8 hours per company when done manually. If you are evaluating 10 potential vendors, that is 40 to 80 hours of research. For a small team, that is an entire week consumed by one task. The work is repetitive, error-prone, and boring. People skip sources, miss red flags, and lose focus after the third company.

This is exactly the kind of task that AI agents handle well. The research is structured, the data sources are predictable, and the output format is consistent. An OpenClaw agent can check 20 data sources in parallel, never gets tired, and produces the same quality report for the tenth company as it does for the first.

The Three-Agent Architecture

The most effective business vetting system uses three specialized agents that work in sequence. Each agent has a focused role, clear inputs, and defined outputs. This separation of concerns makes the system more reliable than a single agent trying to do everything.

This three-agent setup mirrors how human teams handle due diligence. A junior researcher gathers the data, a senior analyst evaluates it, and a manager writes the recommendation. By splitting the work across agents, each one can be optimized independently. You can upgrade the Researcher to check new data sources without touching the Analyst. You can refine the Analyst's scoring criteria without changing how reports are generated.

SOUL.md for the Business Researcher Agent

The SOUL.md file is the configuration that defines an OpenClaw agent's personality, capabilities, and boundaries. Here is a complete SOUL.md for the Researcher agent in a business vetting system:

# SOUL.md - Business Researcher Agent

## Identity
You are a Business Researcher agent. Your job is to collect
factual data about a target company from public sources.
You do not analyze, judge, or recommend. You collect and
organize raw data.

## Inputs
- Company name
- Company website URL (if known)
- Industry (optional)
- Country (optional)

## Data Sources to Check
1. **Website Analysis**
   - Homepage content and messaging
   - About page (team size, founding year, mission)
   - Pricing page (pricing model, tiers)
   - Blog (last post date, posting frequency)
   - Technology stack (via BuiltWith or Wappalyzer)
   - Domain age (via WHOIS lookup)
   - SSL certificate validity

2. **Social Media Presence**
   - LinkedIn (employee count, company page activity)
   - Twitter/X (follower count, last post date, engagement)
   - Facebook (page likes, review score)
   - YouTube (subscriber count, video frequency)

3. **Reviews and Reputation**
   - Google Reviews (rating, review count)
   - Trustpilot (rating, review count, response rate)
   - G2 (rating, review count, category ranking)
   - Glassdoor (employee rating, review themes)
   - BBB (accreditation status, complaint count)

4. **Financial Indicators**
   - Estimated revenue range (via public data)
   - Funding rounds (via Crunchbase or PitchBook)
   - Employee count trend (growing/shrinking)
   - Job postings (hiring velocity)

5. **Legal and Compliance**
   - Business registration status
   - Any public lawsuits or legal actions
   - Privacy policy presence
   - Terms of service presence

## Output Format
Return a JSON object with each data source as a key.
Include the raw value, the source URL, and a timestamp
for when the data was collected. Mark any source that
could not be accessed as "unavailable" with the reason.

## Rules
- Only use publicly available information
- Do not access any paid databases unless API keys are
  provided in the environment variables
- Do not make assumptions. If data is not found, report
  it as not found
- Do not contact the target company
- Complete all checks within 10 minutes
- Rate limit API calls to avoid being blocked

This SOUL.md gives the Researcher agent clear boundaries. It knows exactly what to check, how to format the output, and what it should not do. The structured output makes it easy for the Analyst agent to consume the data in the next step.

SOUL.md for the Business Analyst Agent

The Analyst agent receives the Researcher's JSON output and evaluates it against a scoring framework. Here is the SOUL.md:

# SOUL.md - Business Analyst Agent

## Identity
You are a Business Analyst agent. You evaluate raw company
data and produce a structured risk assessment. You are
skeptical by default and look for inconsistencies.

## Inputs
- JSON data file from the Researcher agent
- Vetting criteria (optional, defaults below)

## Scoring Categories (each scored 0-100)

### Legitimacy Score
- Domain age > 2 years: +20
- SSL certificate valid: +10
- Business registration found: +20
- Physical address listed: +10
- Team members identifiable on LinkedIn: +20
- Privacy policy and ToS present: +10
- Technology stack is professional: +10

### Stability Score
- Employee count > 10: +15
- Employee count growing (vs 6 months ago): +15
- Revenue indicators positive: +20
- Active job postings: +10
- Funding secured or profitable: +20
- Blog updated within 30 days: +10
- Social media active within 7 days: +10

### Reputation Score
- Google Reviews > 4.0: +20
- Trustpilot > 4.0: +20
- G2 > 4.0: +20
- Glassdoor > 3.5: +15
- No BBB complaints: +15
- No public lawsuits: +10

### Risk Flags
Flag any of the following:
- Domain registered < 6 months ago
- No social media presence at all
- All reviews are 5 stars with generic text
- Employee count < 5 with enterprise claims
- No pricing transparency
- Website content copied from another site
- Founder has no LinkedIn history
- Revenue claims with no supporting evidence

## Output Format
Return a JSON object with:
- overall_score (0-100, weighted average)
- category_scores (legitimacy, stability, reputation)
- risk_flags (array of flagged issues with evidence)
- confidence_level (high/medium/low based on data
  completeness)
- recommendation (proceed/caution/avoid)

## Rules
- Never fabricate data points
- If a data source was unavailable, reduce confidence
  level accordingly
- Weight recent data more heavily than old data
- Flag contradictions between data sources

Data Sources the Agent Checks

The strength of automated vetting is the breadth of sources checked. A human researcher might check 5 or 6 sources before running out of patience. An OpenClaw agent checks all of them, every time. Here is a breakdown of what each data category reveals:

The Researcher agent checks all five categories for every target company. When a source is unavailable or returns no data, the agent logs it as unavailable rather than skipping it silently. This matters because a missing data point is itself a signal. A company with no LinkedIn presence, no reviews, and no business registration is telling you something, even if no single data source returned a negative result.

How the Multi-Agent Pipeline Works

The three agents run in sequence, each one building on the output of the previous one. Here is the full pipeline from input to final report:

The total time from input to delivered report is typically 5 to 12 minutes, depending on how many data sources are accessible and how fast external APIs respond. Compare that to 4 to 8 hours of manual research. Even accounting for the time to review the automated report, you are saving 80 to 90 percent of the effort.

Results: Time Saved and Accuracy Gains

Teams that have deployed OpenClaw agents for business vetting report consistent improvements across three dimensions:

The biggest gain is not just speed. It is consistency. When a human researcher vets the first company on Monday morning, the report is thorough. By the fifth company on Friday afternoon, they are skipping sources and writing shorter summaries. An OpenClaw agent applies the same rigor to company number 50 as it does to company number 1.

The cost reduction is also significant. A senior analyst spending 6 hours on vetting at $50/hour costs $300 per company. An OpenClaw agent running Claude for the analysis step costs roughly $1.50 in API fees per company. Even with infrastructure and setup costs, the breakeven point is typically reached after vetting 5 to 10 companies.

Accuracy improves in specific ways. The agent catches data points that humans routinely miss: a domain registered 3 months ago despite claiming 5 years in business, a LinkedIn employee count that dropped 40 percent in 6 months, or a pattern of identical 5-star reviews posted within the same week. These are red flags that require cross-referencing multiple sources, which is exactly what agents excel at.

Real Scenarios Where This System Shines

Automated business vetting is not a theoretical exercise. Here are concrete scenarios where a three-agent vetting system delivers immediate value:

Setting Up Your Vetting System with CrewClaw

Getting started with automated business vetting takes less than an hour. Here is the step-by-step process:

First, grab the agent templates from CrewClaw. You need three agents, so the 5-agent bundle gives you room to add more agents later. Each template comes with the SOUL.md configuration file, a Dockerfile for containerized deployment, a docker-compose.yml for orchestrating the multi-agent pipeline, and bot scripts for connecting to Telegram or Slack.

Second, customize the SOUL.md files for your specific vetting criteria. The examples above are a strong starting point, but you will want to adjust the scoring weights based on what matters most to your business. If you are in a regulated industry, add compliance checks. If you are evaluating technology vendors, add technology stack analysis. The SOUL.md format is plain text, so editing takes minutes.

Third, configure your LLM. For the Researcher agent, a fast model works well since it is mostly collecting and organizing data. For the Analyst and Reporter agents, use a stronger model like Claude for better reasoning and writing. You can set the model per agent in the environment variables.

Fourth, deploy with Docker. Run docker-compose up and your three-agent pipeline is live. Send a company name via Telegram or Slack, and the pipeline kicks off automatically. The Researcher collects data, passes it to the Analyst, which passes the evaluation to the Reporter, which sends the final report back to your channel.

Fifth, iterate on the scoring criteria. After running 10 to 20 vettings, review the reports and adjust. Maybe your industry cares more about Glassdoor scores than Google Reviews. Maybe you want to add a check for patent filings. The system is designed to evolve as your vetting process matures.

You can also scan your own website to see how your business would score through this kind of automated analysis. It is a useful way to understand what the agents look for and how the scoring works.

Why OpenClaw is Built for This

Business vetting is a perfect use case for OpenClaw's agent architecture. The SOUL.md configuration format lets you define exactly what each agent does, what it checks, and how it scores results. The multi-agent pipeline structure mirrors the natural workflow of research, analysis, and reporting. The self-hosted deployment model means sensitive business data never leaves your infrastructure.

Other agent frameworks require writing code to define agent behavior. OpenClaw uses plain-text SOUL.md files. This means non-technical team members can read, understand, and modify the vetting criteria without touching code. Your compliance officer can update the legal check requirements. Your procurement lead can adjust the financial scoring weights. The system stays aligned with your business needs without requiring engineering resources for every change.

The containerized deployment means each agent runs in isolation. If the Researcher agent encounters a rate limit on one data source, it does not crash the Analyst or Reporter. If you want to upgrade the Analyst to use a better model, you redeploy one container without touching the others. This modularity is critical for production systems that need to be reliable.

Frequently Asked Questions