TL;DR
- I built a direct mail campaign for a deck contractor using Zillow, Cursor, and OpenAI's image model... no manual code written.
- The system pulls recently sold high-value homes, generates AI before-and-after deck mockups for each property, and produces a print-ready mailable offer personalized to that homeowner.
- Total build time: four to six hours. The whole thing would have cost six figures to build five years ago.
- ROI math is simple: the campaign costs $3,000 to $5,000 to execute, and a single deck job pays $10,000 and up.
- The same approach works for kitchen remodels, landscaping, HVAC, and any home service business where a visual "before and after" closes the gap.
Most direct mail for contractors looks like this: a generic postcard, a stock photo of a deck, maybe a QR code to a website that hasn't been updated since 2019.
The homeowner opens it, thinks "maybe someday," and throws it in the recycling.
What if instead of a stock photo, they opened the envelope and saw their own backyard... with a deck already rendered into it?
That's what I built. And it took one day.
What the System Does
Here's the pipeline from data to doorstep:
Step 1: Find your prospects. Filter Zillow to recently sold homes above your target price point. We used $500k and up in Fayetteville, Arkansas, filtered to the last six months. Someone who just bought a $600,000 home and has the backyard for a deck is exactly who you want to reach before anyone else does.
Step 2: Generate the mockups. An AI agent scrapes each listing, downloads the exterior photos, and uses OpenAI's image generation model to overlay a photorealistic deck onto the home. Before and after. Each property gets multiple angles.
Step 3: Approve what's good. A lightweight web app lets you swipe through every render and approve or decline. Jackson called it "Tinder for decks." That's accurate. About half will be usable. Some renders put the deck in the driveway. Those get declined. The ones you approve go to the next stage.
Step 4: Generate the mail piece. Each approved property gets a print-ready page: the before-and-after image front and center, a personalized header with the homeowner's address, sales copy, a QR code, and a strong offer. Print on cardstock, fold it into a premium folder, mail it.
Ninety properties went through the system. Forty-four got approved. That's 44 personalized, visual, premium direct mail pieces... generated in one day.
The Offer That Makes Them Call Back
The visual is what gets opened. The offer is what gets a response.
Four things need to land for a cold direct mail piece to work:
Value. What do they get just for saying yes? We offered a complimentary design visit from an experienced contractor, plus a $1,500 design credit toward any project. They get a real deliverable, not a sales pitch.
Risk reduction. Nobody wants to feel like they wasted an afternoon. So we added a guarantee: if you don't feel the visit gave you at least one valuable idea for improving your home, we'll send a $50 dinner gift card as a thank you for your time. Most people won't ask for it. But knowing it's there removes the hesitation that kills most cold offers.
Urgency. We're only extending this visit to 15 selected homes this month. That's not a manufactured number... you genuinely can't follow up on 200 personalized offers in a weekend. The scarcity is real.
Personalization. Their house. Their backyard. Not a stock photo. This separates the piece from everything else in the mail stack before they've read a single word.
If you want to see what AI tools make something like this possible from day one, the AI tools checklist is the right starting point.
How I Built It (Without Writing a Line of Code)
I used Cursor, an AI coding agent you download to your computer for free. You open a project folder, describe what you want in plain English, and it writes the code. The front end. The back end. The image generation scripts. The admin approval app. All of it.
Here's roughly how the build went:
Data collection. I pasted a Zillow search link into Cursor and said "go through this list of homes, download the exterior images, and save them with the address and price in a metadata file." It opened a browser on my computer, crawled every listing, and organized everything into folders. I didn't write a single line of code for any of that.
Image generation. I tested prompts separately in ChatGPT first... taking real Zillow photos and running different instructions against them until I got a result I liked. The key insight: "add a deck to this photo" produces garbage. A specific, professional brief produces something usable. I described it as if I was giving instructions to a skilled graphics artist, with rules like "do not alter the yard, trees, or existing structure" and "the original image must remain visually identical in the before version." Once I had a prompt that worked, I pulled it into Cursor and said "write a script that processes every property, filters out homes that already have a deck, and runs this prompt through OpenAI's image model."
Admin review app. Once mockups were generating, I needed a fast way to approve them. I described what I wanted ("swipe left, swipe right, show me before and after side by side, let me approve or skip") and Cursor built the app. It asked me clarifying questions, I answered them, and then it ran. I went in, approved the ones I liked, and moved on.
Mail piece. I described the layout I wanted, gave it a reference example, and said "generate a print-ready HTML page for each approved property, fits on one page, includes the address, before-and-after images, offer text, and a QR code placeholder." Cursor built it. I tweaked it. The whole thing now exports cleanly to a PDF or prints directly.
When I hit issues, I described what was wrong and Cursor fixed it. The only thing I had to supply myself was an OpenAI API key and a willingness to sit down and work through it.
The Math That Makes This Worth Doing
Say you print 40 of these on nice cardstock, put them in premium folders with the company logo, and mail them. You're probably looking at $3,000 to $5,000 all in once you factor in printing, postage, and your time.
A single deck project runs $10,000 and up. Most jobs bring referrals and follow-on work because now you're inside the homeowner's network. One conversion pays for the whole campaign.
And once the system is built, you can run it every week on new Zillow listings. Jackson pointed this out in the episode and he's right: the real leverage isn't the 40 mailers you sent today... it's that the pipeline keeps running. You could even set it up on Cursor Cloud to run automatically on a schedule, even when your laptop is closed.
This project cost hundreds of thousands of dollars to build five years ago, if it was buildable at all. Now it's a day's work and the cost of API calls. That gap is not going to close.
It's Not Just Decks
The same system works for:
- Kitchen and bathroom remodels. Zillow posts interior photos constantly. The "before this remodel" image is already there.
- Landscaping. Before-and-after on a yard is one of the most visual things you can show a homeowner. This would be huge.
- HVAC and roofing. The prompt needs adjusting, but the pipeline is identical.
- Windows and siding. Show the house before and after a refresh. Curb appeal is easy to visualize.
The underlying pattern stays the same: find people who recently bought in your target market, generate a visual of what their property could look like with your work done, and make the offer compelling enough that saying yes feels like an obvious call.
What to Actually Try This Week
Not sure where AI fits in your business first? Take the strategy diagnosis quiz and it'll point you at the highest-leverage place to start before you go build anything.
If you're a contractor or you work with one, the move is simple:
- Go to Zillow and filter recently sold homes in your area above your ideal project price.
- Download Cursor (free tier is plenty to test this).
- Start with one property. Describe what you want. Ask Cursor to generate a before-and-after mockup using OpenAI's image model.
- If the mockup lands, you now have proof of concept. Build the system to do it for 90 more.
The hardest part is starting. The second-hardest part is writing a prompt specific enough to get a good image. Neither of those things requires a developer... they just require you to sit down and try it.
AI gets you 90% of the way there. You just have to show up for the other 10.