Automate Product Data Enrichment with n8n & AI Agents

Opening Problem Statement

Meet Sarah, a building surveyor tasked with managing a massive inventory of building materials and products for her company. She regularly conducts on-site surveys, capturing photos of various objects to log their details in Airtable. However, manually identifying the product name, model, material, color, and condition from these photos eats up hours each week, leading to inconsistent data and frustrating errors.

Imagine spending 4-6 hours per week tediously researching each item by cross-referencing photos, googling product details, and diving into multiple websites — time that Sarah could spend focusing on higher-value work instead. Mistakes in recording product condition or model can cost her team money and cause delays in procurement or maintenance schedules.

This exact challenge is the origin of the n8n workflow we are about to explore. It automates the tedious task of enriching product data from survey photos, using a combination of AI vision, reverse image search, and web scraping to populate Airtable with accurate and detailed product attributes automatically.

What This Automation Does

This workflow runs through several precise steps that deliver significant automation benefits for Sarah and anyone with similar survey data needs. Here’s what happens when it runs:

• Scans Airtable rows to find product entries that have images but lack enriched AI-generated data.
• Analyzes the product photo using OpenAI’s vision model to extract preliminary descriptions and attributes.
• Invokes an AI agent that performs reverse image searches and web scraping automatically to gather in-depth product information from the internet.
• Parses and enriches product data fields with specific attributes like title, description, model, material, color, and condition.
• Writes the enriched data back into the original Airtable base, marking those rows as processed to avoid duplicate work.
• Automates workflow triggers and routes custom AI tool calls ensuring a seamless, scalable process.

By automating these tasks, Sarah saves around 4-6 hours per week and reduces manual errors drastically, improving data quality and team productivity.

Prerequisites ⚙️

• Airtable account with a base and table containing product images and relevant fields.
• OpenAI API account for AI vision and chat models.
• SerpAPI account to perform Google reverse image searches.
• Firecrawl API account for web scraping and parsing webpages into markdown.
• n8n account to build, test, and run the workflow.
• Optional: Self-hosting option for n8n workflows (e.g., via Hostinger)

Step-by-Step Guide

Step 1: Configure the Manual Trigger

Navigate to the When clicking “Test workflow” node. This is a manualTrigger node that starts the workflow when you click “Execute Workflow” in n8n. This lets you test the entire process on-demand.

The node has no additional configuration needed. When triggered, it passes control downstream to fetch data.

Outcome: Workflow waits for manual trigger input.

Common Mistake: Forgetting to actually trigger the workflow manually during tests.

Step 2: Fetch Relevant Airtable Rows

Go to the Get Applicable Rows node, which is an airtable node configured to query the Airtable base appbgxPBurOmQK3E7 and table tblEHkoTvKpa4Aa0Q.

The filter formula AND(Image!="", AI_status=FALSE()) fetches only rows with a non-empty image field and where AI enrichment is not yet done.

Outcome: Retrieves a list of products with images pending AI processing.

Common Mistake: Incorrect base/table ID or typos in the filter formula can result in no data being fetched.

Step 3: Analyze Product Image Using AI Vision

Open the Analyse Image node, an openAi node configured to analyze the first image URL from the fetched row.

The prompt asks the model to focus on the product (using the row’s title) and extract key attributes like description, model, material, color, and condition.

Parameters: Uses the image passed from Airtable, temperature 0 for deterministic results.

Outcome: Receives a detailed textual analysis of the object from the AI.

Common Mistake: Not providing a valid image URL or API credential errors.

Step 4: Use an AI Agent to Enrich Product Data With Advanced Tools

Go to the Object Identifier Agent node, an agent node leveraging an OpenAI chat model. It has embedded tools for:

• Reverse image search using a custom tool workflow that calls SerpAPI.
• Web page scraping via the Firecrawl scraping API.

This agent receives the analyzed image text and a photo URL to autonomously query both tools, gather contextual information and fill any missing attributes (title, model, material, etc.)

Outcome: An AI agent automatically researches and discovers deeper product data online.

Common Mistake: Forgetting to configure or authenticate tool nodes properly, resulting in API errors.

Step 5: Route AI Tool Calls Based on Requests

The Actions Router node is a switch node. It routes calls to either the Google reverse image search API or the Firecrawl scraper based on the tool the AI agent requests.

Set the routing rules to match route keys like “serp.google_reverse_image” or “firecrawl.scrape” so the proper HTTP requests are triggered.

Outcome: Clean segregation and routing for multiple API tools configured inside the AI agent.

Common Mistake: Route key mismatches cause tool calls to fail or land in fallback.

Step 6: Perform Google Reverse Image Search

Review the SERP Google Reverse Image API node configured to send GET requests to https://serpapi.com/search.json with parameters:

• engine = google_reverse_image
• image_url =

Authentication uses a SerpAPI credential stored in n8n.

Outcome: Returns web pages where images similar to the input photo exist.

Common Mistake: Missing or invalid SerpAPI keys cause authentication failures.

Step 7: Parse Reverse Image Search Results

The Reverse Image Search Response node (a set node) extracts useful details such as title, link, and description from the image results array and strings them into a JSON string for easier consumption by the agent.

Outcome: Prepares well-structured search results for AI usage.

Step 8: Web Scrape Product Pages for Details

The Firecrawl Scrape API node sends a POST request to https://api.firecrawl.dev/v0/scrape with a JSON body containing the URL to scrape, taken from the AI agent’s scraping requests.

Uses HTTP header authentication with Firecrawl API key.

Outcome: Retrieves webpage content in markdown format for text extraction by the AI.

Common Mistake: Incorrect payload structure or missing API credentials cause scraping errors.

Step 9: Handle Web Scraping Success or Failure

Use the Scrape Success? if node to detect if scraping succeeded based on a true success boolean flag in the response.

If successful, the Firecrawl Scrape Success Response node trims the markdown content to 3000 characters and sends it back to the agent.

If failed, the Firecrawl scrape Error Response node sets an error message response.

Step 10: Update Airtable Rows With Enriched Data

The final Enrich Product Rows airtable node updates the original row by matching its ID and writing back AI-generated title, model, material, color, condition, description, and marks AI_status as true.

Outcome: Automated, accurate enrichment of survey inventory data in Airtable.

Step 11: Manual Trigger for Executing Edits

Optional nodes like Edit Fields and Execute Workflow Trigger are configured to manually route and execute workflow invocation with advanced data handling.

Customizations ✏️

• Change Airtable Base/Table: In the Get Applicable Rows and Enrich Product Rows node, update airtable base and table IDs to your own database.
• Add More Product Attributes: Adjust the AI prompt in Analyse Image and Object Identifier Agent to extract extra fields like size, weight, or brand.
• Switch AI Model: In OpenAI Chat Model1, select different OpenAI models like GPT-4 or GPT-3.5 according to your API plan and accuracy needs.
• Customize Tool URLs: Modify the SERP Google Reverse Image API and Firecrawl Scrape API nodes URLs or parameters if you want to switch service providers.
• Trigger Automation: Replace manual trigger with a scheduled trigger or Airtable webhook for fully automatic runs.

Troubleshooting 🔧

Problem: “No data fetched from Airtable”

Cause: Incorrect Airtable base/table IDs or wrong filter formula.

Solution: Double-check the base and table ID in the Get Applicable Rows node settings. Validate the formula syntax. Test with a simple query first.

Problem: “OpenAI API Error or no image analyzed”

Cause: Invalid API key or image URL inaccessible.

Solution: Verify your OpenAI credentials in n8n. Ensure the image URLs are publicly accessible and valid.

Problem: “SerpAPI Authentication Failed”

Cause: Missing or expired API key for SerpAPI.

Solution: Renew or re-add your SerpAPI credentials. Confirm proper credential usage in the HTTP Request node.

Problem: “Firecrawl API scrape failure”

Cause: Incorrect POST body or HTTP headers missing.

Solution: Check your request body parameters and ensure correct API key is added in headers.

Pre-Production Checklist ✅

• Verify Airtable base and table IDs and schema match expected fields.
• Test API credentials for OpenAI, SerpAPI, and Firecrawl.
• Trigger manual test and check intermediate node outputs for accuracy.
• Ensure image URLs in Airtable are valid and publicly accessible.
• Backup Airtable data before bulk updates in case rollback is needed.

Deployment Guide

After thorough testing, activate the manual trigger node or configure a scheduled trigger in n8n for periodic execution.

Use n8n’s execution logs and error notifications to monitor success and identify failures.

Adjust your API quota settings if running large batches, as calls to OpenAI, SerpAPI, and Firecrawl might consume your plan limits.

FAQs

Can I use Google’s built-in reverse image search node instead of SerpAPI?

This workflow uses SerpAPI because it allows custom parameters to tailor the search. You might try Google’s native node, but configuring it for precise reverse image searches might be limited.

Does this workflow consume many API credits?

Yes, especially for OpenAI calls and web scraping via Firecrawl. Monitor usage carefully to avoid unexpected costs.

Is my data safe since this workflow uses external APIs?

Only URLs and query data are sent to third-party APIs. No sensitive personal data is transmitted. Ensure compliance with your company’s data policies.

Can this handle a large volume of product images?

It can handle moderate volumes. For very large datasets, consider batching and monitoring API rate limits.

Conclusion

By following this comprehensive guide, you’ve automated the tedious and error-prone task of enriching building survey product data using an AI-powered n8n workflow. The workflow leverages OpenAI vision and chat models, SerpAPI for reverse image search, and Firecrawl for webpage scraping, all seamlessly integrated via custom AI agents.

This automation not only saves you several hours per week but greatly improves the accuracy and completeness of your Airtable product catalog. Imagine reallocating your team’s efforts to more strategic work and leaving the manual data entry behind!

Next, consider expanding this workflow to automatically generate maintenance schedules, track inventory levels automatically, or integrate real-time alerts for changes in product availability.

Happy automating!