Full-stack developer uses n8n + Claude to build real estate potential AI rating system, a real case of monthly income $3,500

Project goals

Use n8n + Claude Code to build an AI automatic potential customer scoring system to automatically screen and follow up customers with home purchase intentions for real estate agents, saving more than 20 hours of manual time every month and increasing the conversion rate by 3 times.

Identity Anchor

I'm Josh, a freelance full-stack developer with 5 years of experience. In early 2025, at a class reunion, an old classmate who was a real estate agent complained to me: I received 50-80 inquiries from Zillow, Redfin and other platforms every day, but 80% of them were invalid traffic - some asked about nearby house prices, some needed data to write papers, and even some competitors came to find out. Potential customers who are truly interested in buying a house are often drowned in a massive amount of messages, and lack of timely follow-up leads to loss.

I realized that this is actually a typical information filtering + automatic follow-up problem. If I can build a prototype to verify this requirement in two days, it may open up a new revenue channel.

Timeline

• Week 1: Research requirements + Use n8n to build MVP and connect to Zillow API to obtain potential customer data
• Week 2-3: Use Claude Code to write a lead scoring model and an automatic email reply system, and start testing on a small scale
• Weeks 4-6: Optimize scoring accuracy, add SMS notification function, and officially launch the SaaS version
• Weeks 7-12: Expand to 28 paying clients through word of mouth through local REALTORS® association

Scope of application and preconditions

• Familiar with the basic concepts of n8n (workflow, trigger, webhook)
• Understand the basic usage of Claude Code
• Basic API calling experience (REST API, JSON)
• The target customer group is independent real estate agents (not large intermediary chains)

If you are not familiar with n8n yet, you can first read our introductory tutorial: How to Build Automated Content Collection & Publishing with n8n + OpenAI to understand the basic concepts.

Overview of implementation steps

• Step 1: Connect with Zillow/Redfin API to obtain potential customer consultation data
• Step 2: Use Claude Code to build a lead scoring model (Hot/Warm/Cold)
• Step 3: n8n serial automation process: data collection -> scoring -> labeling -> email follow-up
• Step 4: Add WeChat/SMS reminder function (real-time notification of high-value potential customers)
• Step 5: Package as a SaaS multi-tenant system to support simultaneous use by multiple brokers

Task List

• Get the developer key of Zillow API, Redfin API
• Configure data collection webhook in n8n
• Use Claude Code to write potential customer rating prompt templates
• Set up automatic email follow-up template (3-email sequence)
• Configure CRM tag synchronization

Data collection and structuring

Prospect data is not collected directly through the N8N form, but through the Zillow Agent Hub API and Redfin Partner API to obtain the list of inquiries received by the broker. Each brokerage inquiry record contains the following structured fields:| Fields | Example Values | Usage | |------|--------|------| | Source Platform | Zillow / Redfin / Realtor.com | Classification Statistics | | Consultation type | House viewing appointment / valuation consultation / loan consultation | Intent identification | | Budget range | $250K-$400K | Purchasing power filter | | Timeline | Within 1 month / Within 3 months / Uncertain | Urgency score | | Contact | name + email + phone | Follow-up contact | | Past interactions | 0 times / 1 suite viewed | Completion rating |

Potential customer scoring algorithm

def score_lead(lead):
    score = 0
    if lead.get('budget_max', 0) >= lead.get('avg_listing', 300000):
        score += 40
urgency_map = {'Within 1 month': 30, 'Within 3 months': 20, 'Not sure': 10, 'Just looking': 0}
score += urgency_map.get(lead.get('timeline', 'uncertain'), 0)
    if lead.get('has_scheduled_tour'):
        score += 20
    if lead.get('previous_interactions', 0) > 0:
        score += 10
    return min(score, 100)

Strategy generation and implementation

After MVP was launched, Josh discovered a problem: pure rules scoring was too rough. Many high-quality prospects can only be discovered after a phone call, but agents simply don’t have time to call them one by one.

Solution: Integrate Claude Code in n8n, pass the potential customer’s original consultation text + historical interaction records to Claude, and let AI do in-depth scoring at the semantic level.

Breakdown of income sources

Cost vs Profit

Risk boundaries and bottom-up strategies

1. API limit risk: Zillow API has a daily query limit. Use n8n’s rate limiter node to cooperate with the queue mechanism to avoid being blocked.
2. AI misjudgment risk: Claude's score occasionally misjudges, potential customers with a score >= 80 will be notified directly to the agent, and those with a score of 50-79 will be included in the daily summary email
3. Customer churn risk: Launch a 3-month package to lock in at least 3 months of revenue

Failure loop

• First attempt: Using pure rule scoring (budget + timeline + behavior), the accuracy is only 60% -> Reason for failure: Ignoring semantic information, many home buyers’ true intentions are hidden in the consultation text
• Second attempt: Connect to Claude to do AI scoring, but directly adjust the API daily cost $15 -> Reason for failure: Complete analysis of each inquiry, killing the chicken with a knife. Change to: first use rules to filter out obviously invalid inquiries, and only send high-quality potential customers to AI scoring
• Final breakthrough: two-level architecture of rule filtering + AI depth scoring, accuracy increased to 88%, API cost controlled to $1.5 per day

Core Insights

Insight 1: Don’t start out as a SaaS platform

Josh originally wanted to make a platform for all agents. But the success of a tool lies in solving real pain points, not in how many functions it has. He first helped a friend's broker manually build the entire process. After confirming that the solution was indeed effective and getting a clear willingness to pay, he began to consider systematization and multi-tenant architecture.### Insight 2: n8n is an MVP artifact

It only took 1 week to go from idea to working MVP. n8n's built-in nodes (HTTP Request, Webhook, Email, Schedule, Function) cover almost all middleware needs without writing back-end code. The only AI scoring part that requires customization is perfectly solved by the prompt generated by Claude Code.

Insight 3: Pricing based on time savings, not features

Instead of pricing by function, Josh directly anchors the value of time saved by agents: each agent spends about 25 hours per month on prospect screening, and the time cost is $1,250/ months. $99/ Monthly pricing charges only 8% of the time savings value, making it an obvious good deal for brokers.

Tool linkage

Tool names that naturally appear in the text will be matched by the platform side to generate hover-cards from the maintained tools library: n8n Cloud, Claude Code, OpenAI, Claude

Internal link guidance

• Newbies should read the tutorial first: How to Build an AI SDR System with n8n + OpenAI -Tools used in this article: n8n Cloud