The contractor who sends the first quote wins roughly 60% of jobs. Your competitor is not better at their trade — they are just faster at pricing. Here is the architecture that closes that gap permanently.
What this automation actually does
The system starts when a homeowner submits a job photo — via a Jotform or Typeform intake that you embed on your site or send as a link — along with a short text description of the problem. That form submission fires a webhook to n8n, which passes the photo URL and scope text to GPT-4o Vision. The model reads the image, identifies the visible work items (cracked pipe section, burnt breaker box, rusted HVAC coil, whatever is in frame), and returns a structured JSON list of line items.
The critical step — and what makes this safe to deploy on real jobs — is grounding. Every line item the AI returns gets matched against your own Airtable or Google Sheets price list: parts at your cost with your markup, labor rates by task type, contingency rules for job complexity. The AI is only allowed to price from that list. It cannot guess, invent, or hallucinate a number that does not exist in your master rates. The final estimate assembly prompt enforces this as a hard rule. From there, n8n renders a clean branded HTML-to-PDF quote, attaches it, and fires it to the homeowner via Twilio SMS and Gmail — usually within 3 to 5 minutes of the photo being submitted.
Who needs this (and who doesn't)
This is built for owner-operators and small-crew trades businesses — electricians, plumbers, HVAC techs, general contractors — doing between $500K and $5M per year, where the owner or a senior estimator is personally writing quotes by hand, often at night, after a full day on the tools. The pain is not that they lack technical skill — it is that the quoting bottleneck is invisible until close rate data surfaces it. Jobs lost to a faster competitor never show up in the books as 'lost because we were slow.' They just do not show up at all.
The pattern also works for any trade where a visual scope is the primary input: roofing damage assessments, landscaping bids, pool repair estimates, pest inspection reports. Any job where a photo of the problem is the first thing a customer can provide is a candidate for this stack.
- Good fit: trades owner running 1-15 trucks who writes quotes manually or delegates to one estimator
- Good fit: any business where the first-to-quote advantage is measurable and the job scope is visually assessable
- Good fit: contractors already using Airtable, Google Sheets, or any structured price list — even a spreadsheet from 2019 works
- Good fit: businesses where average job value is $300+ (the ROI on faster quoting is immediate)
- Skip if: your quoting requires licensed engineer sign-off on every line item before delivery
- Skip if: your price list changes daily based on live supplier feeds — the lookup step needs a stable master list
How to build it
Build the Photo Intake Form
Create a Jotform or Typeform with three fields: customer name, phone/email, a file-upload field for the job photo (accept JPG/PNG/HEIC), and an optional 'describe the problem' text field. Set the form action to POST to an n8n webhook URL. This is the entry point — keep it simple enough that a homeowner can fill it out from their phone at the job site in under 60 seconds.
Set Up Your Master Price List in Airtable or Sheets
Create a table with columns: Item Code, Description, Unit (each / LF / HR), Material Cost, Labor Cost, Standard Markup %, and Job Type tag (plumbing / electrical / HVAC / general). Populate it with your real rates — even 50 line items is enough to start. This table is what makes the AI's output usable on real money. Every price the system generates traces back to a row in this table.
Wire the GPT-4o Vision Extraction Node in n8n
In n8n, create an HTTP Request node that sends the form photo URL to the OpenAI Chat Completions endpoint using the gpt-4o model with vision enabled. The system prompt instructs the model to return a JSON array of work items with fields: item_description, estimated_quantity, unit, job_type, and confidence (high / medium / low). Low-confidence items are flagged for manual review rather than priced automatically.
Build the Price-List Lookup Logic
For each item in the GPT-4o JSON output, n8n queries your Airtable or Sheets table for a fuzzy match on Description and Job Type. Use an n8n Function node or a dedicated Airtable Search node. If a match is found, attach the Material Cost, Labor Cost, and Markup %. If no match is found, add the item to a 'Review Required' list and exclude it from the auto-estimate total. Never let the AI fill in a price from its own knowledge.
Assemble the Estimate with Contingency Rules
Pass the priced line-item array to a second GPT-4o prompt that formats it as a line-item estimate: subtotal, a configurable contingency percentage (typically 10-15%), markup tier applied, and a total. The prompt is instructed to output only from the provided data and flag any item it cannot resolve. This is also where job complexity rules apply — for example, if more than 3 high-complexity items are present, bump contingency to 20%.
Generate the Branded PDF
Use an n8n Code node to render the estimate data into an HTML template that includes your logo, business name, license number, payment terms, and a quote validity date. Pass the HTML to a PDF-generation service (html-pdf-node, Puppeteer on a small VPS, or a hosted service like PDFMonkey) and store the output file temporarily. This is the deliverable the homeowner receives.
Deliver by SMS and Email + Log the Quote
Use the Twilio n8n node to send an SMS: 'Hi [Name], your estimate from [Business] is ready — [PDF link]. Total: $X. Valid 7 days. Reply YES to confirm or call us.' Send the PDF as an email attachment via Gmail. Write a row to your tracking sheet: customer name, quote total, timestamp, and status = 'sent'. Trigger a 48-hour follow-up SMS automatically if status has not changed to 'won' or 'lost'.
Run a Dry-Run on a Real Past Job
Before going live, take 3-5 photos from real past jobs you have already quoted manually. Run them through the system and compare AI output to your actual quotes. Check that line-item matches are accurate, that no items are hallucinated, and that the total is within 10-15% of your manual estimate. Fix any price-list gaps you find. This calibration step is what separates a toy from a deployable system.
Approach | Cost (monthly) | Setup time | Best for |
|---|---|---|---|
| n8n + GPT-4o + Airtable (self-hosted) | $30-80 API fees + VPS | 8-16 hours | Contractors who want full control and a one-time build cost |
| n8n Cloud + GPT-4o + Google Sheets | $50-120 (n8n cloud + API) | 4-8 hours | Contractors who want no VPS management with a lower complexity build |
| Rebar / purpose-built HVAC SaaS | $300-600+ | 1-2 hours (onboarding) | Larger HVAC operators who want a vendor-managed solution and do not need customization |
| Manual estimating (status quo) | Owner time: $50-150/hr opportunity cost | 0 (already doing it) | Nobody — this is the problem, not the solution |
Get the Free Build Kit: Prompts, Price-List Schema, and n8n Blueprint
The teardown above teaches the pattern. The kit gives you the exact GPT-4o prompts, a pre-filled Airtable schema with a starter trades catalog, the branded HTML estimate template, and the full n8n node map — so you are building on a tested foundation, not starting from scratch. Free to download.
You are an expert trade estimator assistant. You will receive a job-site photo and an optional text description of the problem.
Your task: identify every distinct work item visible in the photo or described in the text. Return ONLY a valid JSON array. Do not include any explanation or prose outside the JSON.
For each work item, output an object with exactly these fields:
- item_description: string (clear, specific trade description, e.g. "Replace 1/2" copper elbow fitting")
- estimated_quantity: number
- unit: string ("each" | "LF" | "SF" | "HR" | "set")
- job_type: string ("plumbing" | "electrical" | "hvac" | "general")
- confidence: string ("high" | "medium" | "low")
Rules:
- Only list items you can observe or that are clearly described. Do not infer items that are not visible.
- If the photo quality is insufficient to identify specific items, return a single item with confidence: "low" and item_description: "Photo quality insufficient — manual review required".
- Do NOT include any pricing. Pricing is handled separately.
- Do NOT add commentary, apologies, or caveats outside the JSON array.Want NoFluff to Build and Deploy This for Your Business?
We build and deploy this stack for trade businesses end-to-end: intake form, price-list setup, n8n workflow, PDF branding, SMS/email delivery, and go-live testing on real jobs. You get a running system in 5-7 days, not a tutorial. If you are quoting more than 20 jobs a month and losing any of them to slow turnaround, the math is straightforward. Book a 20-minute scoping call.