Cut Renovation Proposal and Project-Gallery Writing in Half with Claude Code

Friday night, 10 p.m. A site visit was scheduled for the next morning, but the salesperson I know was still at the office. What was he doing? Hunting for a file: “Where did I save that kitchen-remodel proposal from before?” He copied last year’s estimate, forgot to change the client’s name in one spot, and almost printed it out with a different customer’s name still on it. I’ve heard a version of that story more times than I can count.

Remodeling work is supposed to be about the job site and coordinating the crew. Yet somehow the evenings dissolve into paperwork instead. Proposals, the explanation text on estimates, project-gallery updates, captions for before-and-after photos. Every one of them lives or dies on “does this land with the customer,” and yet there’s never enough time, so it all turns into recycled boilerplate.

Claude Code is surprisingly good at pulling you out of that “recycled-text purgatory.” Today I’ll walk through how far a remodeling company’s salespeople and tradespeople can hand off proposal writing and project-gallery updates to AI, based on what I actually tested myself.

Key takeaways

• For proposal drafts and project-gallery copy, the fastest setup is letting Claude Code write 80% and a human fix the remaining 20%.
• Hand it your site-visit notes and a few past cases, and it drafts the customer-facing pitch, the project timeline, and the reasoning behind a rough quote.
• The final price, anything touching structure or building codes, and how you handle personal data are things a human must always stop and decide.
• If you set up a workflow that never feeds the customer’s address, name, or photos straight into the AI, you avoid privacy accidents.
• Each proposal dropped from 1.5 to 2 hours down to 30 minutes. At 20 proposals a month, that’s over 20 hours saved.

Who at a remodeling company this actually helps

The reader I have in mind is a salesperson at a 5-to-30-person contractor or remodeling firm who handles both the pitch and the job site. Or a tradesperson who wants to share their work on a blog or gallery page but finds writing painful. The smaller the company and the more likely it has nobody dedicated to marketing, the more this hits home.

The path to winning a remodeling job usually looks like this:

1. Receive an inquiry or referral.
2. Do a site visit to confirm the customer’s wishes and the building’s condition.
3. Build a plan and a rough quote, then package it into a proposal.
4. Present the proposal and estimate, then sign the contract.
5. Do the work and hand it over.
6. Shoot the finished photos and post them to the project-gallery page.

Of these, step 3 (writing the proposal) and step 6 (updating the gallery) are the two biggest bottlenecks. Neither is your actual trade, yet they eat the most time. And here’s the thing: these two handle nearly identical information. Site-visit notes flow into the proposal, into the job, and into the gallery page. The same project’s information just keeps changing shape as it moves along. That’s exactly why it’s easy to hand off to AI as a bundle.

The rework and headaches you keep hitting

Here are the “yep, that’s us” moments I’ve heard from people on the ground. Sound familiar?

• You copy an old proposal and hand it over with a previous customer’s name or price still buried in it.
• Your proposal copy always uses the same phrasing, and customers can smell the template a mile away.
• The project gallery has been frozen for six months. The photos are still asleep on a laptop somewhere.
• The job description a tradesperson wrote is too technical, so it never lands with the customer.
• You can only explain the basis for a rough quote out loud, which later turns into a “he said, she said” mess.

Every one of these traces back to the same root: no time to write from scratch. Hand that off to AI, and you get time back for the on-site coordination you actually want to be doing.

Use case 1: Turn site-visit notes into a proposal draft

This is the one with the biggest payoff. Hand it the wishes and building condition you jotted down during the visit, and it drafts the customer-facing pitch, the project timeline, and the reasoning behind the rough quote.

The key is deciding your “template” up front. Lock the proposal’s section structure, then pour each new project into that same shape, and the quality stops swinging around. Copy the prompt below, rewrite it in your own company’s voice, and use it.

You are a veteran salesperson at a remodeling company.
Using the site-visit notes below, write a draft proposal for the customer.

[Output template]
1. The customer's wishes, organized (in 3 lines)
2. Overview of the plan we're proposing
3. The project timeline (numbered, in plain customer-facing language)
4. The basis for the rough quote (per line item, one sentence on why that cost applies)
5. Requests and cautions for the customer during the work

[Rules]
- If you use a technical term, restate it plainly in parentheses
- I will finalize the prices, so mark every figure with "*to be confirmed"
- Do not fill gaps with guesses; for anything not in the notes, write "needs site confirmation"

[Site-visit notes]
(Paste the notes here. Address and name can be redacted.)

That last rule, “do not fill gaps with guesses,” matters a lot. Leave it out, and the AI will happily invent assumptions like “assuming the building is about 20 years old” or “a full bathroom replacement is standard here.” In remodeling, that kind of thing causes accidents, so make it write “needs site confirmation” for anything it doesn’t actually know.

Use case 2: Turn a tradesperson’s job notes into gallery copy

The usual reason a project gallery stalls is “we have the photos, but nobody can write the text.” So have the AI convert the short notes a tradesperson left on site into readable, customer-facing copy.

The tradesperson’s notes can be fragments. Even a scrap like “bathroom, cracked tile, switched from built-in to prefab unit bath, 1 tsubo, 3 days” turns into something usable if you ask like this.

You are the PR person at a remodeling company.
Turn the tradesperson's on-site notes into customer-facing copy for the project gallery.

[Output template]
- Headline (under 60 characters, search-aware, include "area name + type of work")
- The problem the customer had (2-3 lines)
- What the work involved and the touches we were proud of (restate any technical terms)
- The change after the work (weave in a customer quote if there is one)
- A note to others facing the same problem

[Rules]
- Don't inflate with results or numbers that aren't in the notes
- Avoid over-the-top absolutes like "guaranteed" or "completely"

[Tradesperson's on-site notes]
(Paste the notes here)

What’s nice about this template is that it forces the headline to include the area name and the type of work. Customers search with phrases like “Austin bathroom remodel,” so if you miss that, your gallery can grow forever and still never turn into inquiries. If you want to go deeper on the writing itself, pair this with advanced prompt engineering for Claude Code to make building these templates a lot easier.

Use case 3: Strip personal data before you hand it over

This one is unglamorous but the single most important habit. You want to avoid pasting a customer’s address, name, and phone number straight into the AI. So keep one small script that swaps personal data out for placeholders before you paste anything.

Below is a check script that runs on Node.js. It reads a text file of site-visit notes, masks phone numbers, emails, and ZIP codes, then checks whether anything is still showing and prints the result. Copy it, save it as mask.mjs, and run it with node mask.mjs memo.txt.

import { readFile } from "node:fs/promises";

const file = process.argv[2] || "memo.txt";
let text = await readFile(file, "utf8");

// Replace phone numbers, emails, and ZIP codes with placeholders
const rules = [
  [/\(?\d{3}\)?[-.\s]?\d{3}[-.\s]?\d{4}/g, "[PHONE]"],
  [/[\w.+-]+@[\w-]+\.[\w.-]+/g, "[EMAIL]"],
  [/\b\d{5}(?:-\d{4})?\b/g, "[ZIP]"],
];

let hits = 0;
for (const [pattern, label] of rules) {
  text = text.replace(pattern, () => { hits++; return label; });
}

console.log("--- Masked text ---");
console.log(text);
console.log(`\nReplacements made: ${hits}`);
if (hits === 0) console.log("Note: nothing was detected. Check names and addresses manually.");

This script only catches phones, emails, and ZIP codes. Names and addresses have no fixed shape, so a human checks those last. Even so, just building the habit of “do one extra step before you paste” sharply cuts down on accidents where personal data flows straight to the cloud. If this is your first time touching Claude Code, get the install and basics out of the way first with the Claude Code getting-started guide, and this script will be ready to run in no time.

What to delegate to AI vs. decide yourself

Let me draw a clear line. Leave this fuzzy and the accidents start outweighing the convenience.

Task	Delegate to Claude Code	A human must decide
Proposal draft	Generating and rewording copy	Whether it’s factual and free of exaggeration
Rough quote	Writing up the reasoning	Locking in the price itself
Structure and codes	General explanatory text	Whether seismic, fire, or permit checks apply
Project gallery	Writing headlines and body copy	Choosing photos, getting publishing consent
Personal data	Processing text after masking	Final check that no name or address remains

The memory trick is simple: money, safety, and people’s names are where the human stops. Everything else that’s writing work, hand it off freely. With that split, even a tradesperson won’t get lost. If you want to keep it as an internal rule, follow the CLAUDE.md best practices and drop the table above straight into your project’s rule file, so the AI runs against the same standard every time.

What changed, before and after

This is the felt experience of a small remodeling company I know (three salespeople).

Before, each proposal took 1.5 to 2 hours. The time dissolved into hunting for old files and writing copy from a blank page. Gallery updates happened “whenever the mood struck,” and in six months only two new cases went up.

After, they mask the site-visit notes, hand them over, and a draft appears in about five minutes. A human polishes it in 20 to 25 minutes. Together that lands around 30 minutes. The gallery now grows by four or five cases a month from the tradespeople’s notes, and inquiries from local-area searches are slowly ticking up.

To put a rough number on the ROI: at 90 minutes saved per proposal and 20 proposals a month, that’s 30 hours. At a $25-per-hour rate, that’s roughly $750 a month worth of time freed up. More than the tool cost, the bigger win for people on the ground is using that freed time to fit in one more site visit.

Security and personal-data notes

Remodeling handles addresses, floor plans, even household makeup. It’s a job that’s practically a brick of personal data. That’s exactly why you should hold to just three things.

• Never paste raw customer data such as name, address, phone, or floor plans. Mask it with the script above first.
• Before posting photos to the project gallery, get the customer’s publishing consent in writing. Check that no nameplate or license plate is caught in the frame.
• If you’re using this at a company, confirm you’re on a plan that does not train on your input data, rather than a free consumer plan.

If you want to dig deeper into this “foundation for delegating safely,” the Claude Code intro for non-engineers lays out how to think about permissions and backups. For the official take on handling personal data in the US, the FTC’s privacy and security guidance is a reliable primary source.

FAQ

Q. Can a tradesperson who’s bad with computers actually use this? A. For the part where you ask it to write, you just give instructions in plain English. The only hard part is the initial install, so get a tech-savvy person to help once, and after that it’s just paste-the-notes-and-ask on repeat.

Q. Can I hand a customer the proposal the AI made, as is? A. No. The price and anything touching structure or building codes always get a human check. The AI is strictly “the fast-draft department,” and a human carries the final responsibility.

Q. Can I have the AI learn from our past proposals? A. If you mask your past proposals first and then hand them over, it can draft something that mimics your tone and structure. Codify the templates you reuse using the approach in CLAUDE.md best practices, and you won’t have to paste them in every time.

Q. How much time do I actually save per month? A. At 20 proposals a month, the company I watched freed up roughly 20 to 30 hours. Once gallery updates start flowing, the bump in inquiries follows later as a bonus.

Q. Any tips for getting even faster? A. Claude Code productivity tips collects the small moves that work on the ground: reusing templates, keeping instructions short, and handing off checks to a command.

How it went when I actually tried it

I actually wrote up a fictional bathroom-remodel site-visit note myself and ran all three steps end to end. First I put it through the masking script, and two items, a phone number and a ZIP code, got cleanly swapped out. The name stayed, because it has no fixed shape, which reminded me yet again that you have to delete that by hand. Not leaving it all to the script is the right call.

Next I had it produce the proposal draft, and the project timeline and the reasoning behind the rough quote took shape in five minutes. There was one spot where it flatly wrote “a prefab unit bath can be swapped in half a day,” which doesn’t match reality on site, so I fixed it. The line stays firm: a human stops on price and schedule.

On the gallery side, the fragments of the tradesperson’s notes turned into a plausible headline and body, and the area name went in properly too. The biggest payoff was that the psychological hurdle of writing simply vanished. Writing from scratch versus editing text that’s already there, the difference in how reluctant a tradesperson is to start is night and day. If you want to lift your whole company’s proposal and gallery output, starting with training and consulting to build templates around your own workflow is the fastest route, I think.