CrewAI for Local Business: Build a Team of AI Agents

What Is CrewAI?

• A role: "Senior Review Analyst" or "Instagram Content Creator"
• A goal: what this agent is trying to achieve
• A backstory: persona context that shapes tone and judgment
• Tools: functions the agent can call (search the web, read files, call APIs)

Why Multi-Agent Beats Single-Agent for Local Business

CrewAI Core Concepts

Installation and Setup

pip install crewai crewai-tools langchain-anthropic python-dotenv

# .env — never commit this file to Git
ANTHROPIC_API_KEY=YOUR_API_KEY
# Or if using OpenAI:
# OPENAI_API_KEY=sk-YOUR-KEY

import crewai
print(crewai.__version__)  # Should print 0.80.0 or later

The Local Business Marketing Crew: Full Example

1. Review Monitor Agent — reads new Google reviews, flags any with 3 stars or below
2. Response Writer Agent — drafts professional, empathetic replies to flagged reviews
3. Social Media Agent — writes 5 Instagram captions based on the week's top services
4. Weekly Report Agent — compiles all outputs into a Monday morning email digest

Step 1: Define Your Tools

# tools.py
from crewai.tools import tool
import json
from datetime import datetime, timedelta

@tool("Read Recent Google Reviews")
def read_recent_reviews() -> str:
    """Reads Google reviews from the past 7 days and returns them as JSON."""
    # In production: call Google My Business API
    # For demo: return mock reviews
    reviews = [
        {
            "id": "rev_001",
            "author": "Sarah M.",
            "rating": 5,
            "text": "Amazing blowout! My hair has never looked better. Booked again already.",
            "date": "2026-06-12"
        },
        {
            "id": "rev_002",
            "author": "James T.",
            "rating": 2,
            "text": "Waited 25 minutes past my appointment time with no apology. The cut was fine but the wait ruined my morning.",
            "date": "2026-06-11"
        },
        {
            "id": "rev_003",
            "author": "Priya K.",
            "rating": 4,
            "text": "Great colour work, really happy with the result. Parking can be tricky though.",
            "date": "2026-06-10"
        },
        {
            "id": "rev_004",
            "author": "Mike D.",
            "rating": 1,
            "text": "Terrible experience. Wrong colour applied, had to come back next day. Still not fully fixed.",
            "date": "2026-06-09"
        }
    ]
    return json.dumps(reviews, indent=2)

@tool("Read Weekly Bookings")
def read_weekly_bookings() -> str:
    """Returns this week's most popular services from the booking system."""
    # In production: query your booking software's API (Acuity, Fresha, etc.)
    bookings = {
        "week": "June 9–13, 2026",
        "top_services": [
            {"service": "Full Balayage", "count": 14},
            {"service": "Keratin Treatment", "count": 9},
            {"service": "Precision Cut & Blow-dry", "count": 22},
            {"service": "Brow Tint + Shape", "count": 11},
            {"service": "Wedding Trial Hair", "count": 3}
        ],
        "total_appointments": 59,
        "new_clients": 8
    }
    return json.dumps(bookings, indent=2)

@tool("Save Report to File")
def save_report(content: str) -> str:
    """Saves the weekly digest to a markdown file."""
    filename = f"weekly_digest_{datetime.now().strftime('%Y_%m_%d')}.md"
    with open(filename, "w") as f:
        f.write(content)
    return f"Report saved to {filename}"

Step 2: Define the Agents

# agents.py
from crewai import Agent
from langchain_anthropic import ChatAnthropic
from tools import read_recent_reviews, read_weekly_bookings, save_report

# LLM — using Claude Sonnet for cost/quality balance
llm = ChatAnthropic(
    model="claude-sonnet-4-5",
    api_key="YOUR_API_KEY",  # Load from env in production
    temperature=0.7
)

review_monitor = Agent(
    role="Google Reviews Monitor",
    goal="Identify all reviews with a rating of 3 stars or below and summarise the key complaints.",
    backstory=(
        "You are a reputation management specialist for local service businesses. "
        "You have 10 years of experience identifying patterns in negative customer feedback "
        "and distilling them into actionable insights for business owners."
    ),
    tools=[read_recent_reviews],
    llm=llm,
    verbose=True
)

response_writer = Agent(
    role="Customer Response Specialist",
    goal=(
        "Write professional, warm, and empathetic replies to negative Google reviews. "
        "Each reply should acknowledge the issue, apologise sincerely, and invite the customer back."
    ),
    backstory=(
        "You are a customer service expert who has helped hundreds of salons, cafés, and small businesses "
        "recover from negative reviews. Your replies are always specific, never generic, "
        "and consistently increase the chance of the reviewer returning."
    ),
    tools=[],
    llm=llm,
    verbose=True
)

social_media_agent = Agent(
    role="Instagram Content Creator",
    goal=(
        "Write 5 Instagram captions for this week's most popular services. "
        "Each caption must be engaging, include a call to action, and be under 150 words."
    ),
    backstory=(
        "You are a social media strategist specialising in local beauty and wellness businesses. "
        "You know how to write captions that drive bookings — not just likes. "
        "Your tone is warm, aspirational, and always authentic."
    ),
    tools=[read_weekly_bookings],
    llm=llm,
    verbose=True
)

report_agent = Agent(
    role="Weekly Marketing Digest Compiler",
    goal=(
        "Compile all outputs — flagged reviews, draft responses, and social media captions — "
        "into a clean Monday morning email digest the business owner can act on immediately."
    ),
    backstory=(
        "You are a marketing operations specialist. You turn raw AI outputs into polished, "
        "structured reports that busy business owners can read in under 5 minutes and act on immediately. "
        "You prioritise clarity, action items, and brevity."
    ),
    tools=[save_report],
    llm=llm,
    verbose=True
)

Step 3: Define the Tasks

# tasks.py
from crewai import Task
from agents import review_monitor, response_writer, social_media_agent, report_agent

task_monitor_reviews = Task(
    description=(
        "Use the Read Recent Google Reviews tool to fetch this week's reviews. "
        "Identify every review with a rating of 3 stars or below. "
        "For each flagged review, extract: the author name, rating, date, and a one-sentence summary of the core complaint."
    ),
    expected_output=(
        "A structured list of flagged reviews with author, rating, date, and complaint summary. "
        "Example format: '⚠️ James T. (2★, June 11) — Complaint: waited 25 min past appointment, no acknowledgment.'"
    ),
    agent=review_monitor
)

task_write_responses = Task(
    description=(
        "Using the list of flagged reviews from the previous task, write a professional Google review reply for each one. "
        "Each reply must: (1) address the reviewer by first name, (2) specifically acknowledge their complaint, "
        "(3) apologise sincerely without being defensive, (4) offer a resolution or invitation back, "
        "(5) be signed 'Warm regards, [Your Name] and the Team'. Keep each reply under 120 words."
    ),
    expected_output=(
        "One polished reply per flagged review, clearly labelled by reviewer name. "
        "Ready to copy-paste directly into Google Business Profile."
    ),
    agent=response_writer,
    context=[task_monitor_reviews]
)

task_social_captions = Task(
    description=(
        "Use the Read Weekly Bookings tool to see this week's most popular services. "
        "Write 5 Instagram captions — one per top service. Each caption must: "
        "(1) highlight the benefit to the customer (not just the feature), "
        "(2) include a relevant emoji or two, "
        "(3) end with a call to action (e.g. 'Book your slot — link in bio'), "
        "(4) include 5–8 relevant hashtags. Max 150 words per caption."
    ),
    expected_output=(
        "5 complete Instagram captions, numbered 1–5, each ready to post. "
        "Include hashtags at the end of each caption."
    ),
    agent=social_media_agent
)

task_compile_digest = Task(
    description=(
        "Compile the outputs from all previous tasks into a clean Monday morning email digest. "
        "Structure it as follows:\n"
        "## 📊 Monday Marketing Digest — [Date]\n"
        "### 🔴 Reviews Requiring Attention ([count] reviews)\n"
        "[Flagged reviews list]\n"
        "### ✉️ Draft Responses (ready to post)\n"
        "[Review responses]\n"
        "### 📸 This Week's Instagram Captions\n"
        "[5 captions]\n"
        "### ✅ This Week's Action Checklist\n"
        "[3–5 bullet points the owner should do today]\n"
        "Save this digest to a file using the Save Report to File tool."
    ),
    expected_output=(
        "A complete, formatted markdown digest saved to file and also returned as output. "
        "The action checklist must be specific and prioritised."
    ),
    agent=report_agent,
    context=[task_monitor_reviews, task_write_responses, task_social_captions]
)

Step 4: Assemble and Run the Crew

# main.py
from crewai import Crew, Process
from agents import review_monitor, response_writer, social_media_agent, report_agent
from tasks import (
    task_monitor_reviews,
    task_write_responses,
    task_social_captions,
    task_compile_digest
)
from dotenv import load_dotenv

load_dotenv()

marketing_crew = Crew(
    agents=[review_monitor, response_writer, social_media_agent, report_agent],
    tasks=[task_monitor_reviews, task_write_responses, task_social_captions, task_compile_digest],
    process=Process.sequential,
    verbose=True
)

if __name__ == "__main__":
    print("🚀 Starting Monday Marketing Crew...")
    result = marketing_crew.kickoff()
    print("\n✅ Crew complete. Digest saved.")
    print(result)

What the Monday Digest Looks Like

## 📊 Monday Marketing Digest — June 13, 2026

### 🔴 Reviews Requiring Attention (2 reviews)
⚠️ James T. (2★, June 11) — Complaint: waited 25 min past appointment time with no acknowledgment
⚠️ Mike D. (1★, June 9) — Complaint: wrong colour applied, return visit required, still not resolved

### ✉️ Draft Responses (ready to post)

**Reply to James T.:**
Hi James, thank you for taking the time to share your experience. We're truly sorry you waited 25 minutes — that's not the standard we hold ourselves to, and we completely understand your frustration. We'd love the opportunity to make it right. Please DM us and we'll arrange a complimentary styling service at your next visit. Warm regards, Nataliia and the Team

**Reply to Mike D.:**
Hi Mike, we're so sorry about your colour experience — this should never have happened and we take full responsibility. We'd like to correct this completely at no charge. Please call us directly at [PHONE] so we can arrange this as a priority. We genuinely value your trust and want to earn it back. Warm regards, Nataliia and the Team

### 📸 This Week's Instagram Captions
[5 ready-to-post captions with hashtags]

### ✅ This Week's Action Checklist
1. Post both review responses on Google Business Profile today (before 12pm)
2. Schedule 3 of the 5 Instagram captions for Tue/Thu/Sat
3. Follow up with Mike D. directly by phone — priority
4. Check wait time procedures with front desk team
5. Send thank-you DM to Sarah M. (5-star review)

Running the Crew on a Schedule

Option A: Cron Job (Linux/Mac)

# Run every Monday at 7:00 AM
crontab -e

# Add this line:
0 7 * * 1 cd /path/to/your/project && /usr/bin/python3 main.py >> /var/log/marketing_crew.log 2>&1

Option B: GitHub Actions (cloud, free tier)

# .github/workflows/weekly_digest.yml
name: Weekly Marketing Crew

on:
  schedule:
    - cron: '0 7 * * 1'  # Every Monday at 7:00 AM UTC
  workflow_dispatch:       # Also allow manual trigger

jobs:
  run-crew:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: actions/setup-python@v5
        with:
          python-version: '3.11'
      - name: Install dependencies
        run: pip install crewai crewai-tools langchain-anthropic python-dotenv
      - name: Run Marketing Crew
        env:
          ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}
        run: python main.py
      - name: Upload digest
        uses: actions/upload-artifact@v4
        with:
          name: weekly-digest
          path: weekly_digest_*.md

Comparison: CrewAI vs LangGraph vs AutoGen vs n8n

Common Mistakes to Avoid

FAQ

Related Articles

• LangGraph: Build an AI Booking Agent with Memory and State
• LangChain RAG: Answer Customer FAQs Automatically
• AutoGen Multi-Agent Marketing Automation for Local Business
• AgentOps: Monitor and Debug Your AI Agents
• OpenRouter Guide for Local Business Automation 2026