Ted Sihpol quoted a waterfront renovation at $1,100 a square foot. A week later the client sent him a 21-page AI-generated report priced at $1,200 to $1,300, with three scenarios and a projected return on each. Ted's words: "I think I shorted myself 100 bucks a square foot." This issue is about how he fixed that and what you can take from it.

Ted runs Renaissance Partners out of Fairfield County, Connecticut. Four self-performing carpenters, 99% residential remodeling, nearly every job within ten miles of his house. He came out of Wall Street, he's 58, and he is not who you'd expect to be running AI estimating. He's also cut his estimate time from three to four hours down to ninety minutes, rebuilt his subcontractor agreement in one sitting, and closed a job the morning after the proposal landed. This issue walks through the exact workflow: what he built, why it works, and how to build the same thing for your business.

Ted quoted a waterfront renovation. He felt good about the number.

Then the client sent him a report. 21 pages. AI-generated. Three project scenarios, a projected return on each, and a price range of $1,200 to $1,300 a square foot. More thorough than Ted's proposal. More options than Ted had offered. Priced higher than what Ted had quoted.

He had undercut himself by $100 a square foot. On a waterfront renovation. Because his client had done research he hadn't.

High-end residential clients in 2026 are not waiting for contractors to tell them what things cost. Some of them are running the same tools and arriving at better numbers faster. The ones who don't build their estimating model around their own real costs are the ones getting beat by their own clients.

Here is exactly what Ted built and how to replicate it.

Step 1: Build the context document before you touch a single estimate.

This is where most contractors skip ahead and wonder why the output is generic. Before you ask AI anything about a specific job, you need a document that tells it who you are and what your work actually costs.

Your context document should include:

Your cost per square foot by material category (open-cell spray foam, closed-cell spray foam, crown installation, trim runs, drywall, tile, etc.)

Your labor rates by trade and by your market

Your service area and any market-specific cost factors

Your target margin, not your markup percentage

That last line matters. Ted builds estimates around the margin he wants, not a markup. A 25% markup is not a 25% margin. If your estimating model is built on markup, you are consistently underpricing without knowing it. Fix the math in the context document before anything else.

Save this document in a Claude Project. It becomes the foundation every estimate runs on.

Step 2: Feed the job into the model, not the other way around.

Most contractors open a chat window and ask what a screened porch costs. Ted does the opposite. The context document is already loaded. He feeds in the project specifics: scope, square footage, material selections, and any known constraints. The model outputs line items he pastes straight into BuilderTrend.

Use this prompt structure once your context document is in place:

"Using my cost rates and margin target from the context document, build an estimate for the following project: [scope description, square footage, materials, location]. Format the output as line items I can paste into BuilderTrend: trade category, description, quantity, unit cost, extended cost, and margin."

An estimate that used to take three to four hours now takes ninety minutes. Ted has won two projects using this process. On the most recent one, the proposal was viewed 30 minutes after he sent it and signed the next morning.

Step 3: Use AI to check your math, not replace it.

Ted still runs his spray foam math longhand for 20 minutes and uses AI to verify it, never the other way around. His framing: everybody can buy a router table. Buy the cheap one and you spend the rest of your life sanding chatter marks off raised panels.

The same principle applies to estimating. AI is a verification layer for a builder who knows his numbers, not a substitute for a builder who doesn't. Run your critical line items manually first. Then run them through the model and look for variance. The gaps are where you learn something.

Step 4: Rebuild your subcontractor agreement in one sitting.

Ted took his insurance carrier's version of a subcontractor agreement, merged it with the AIA master document, and built a protected template that auto-fills the sub's name once. The whole process took one session.

Use this prompt:

"I have two subcontractor agreement documents. The first is from my insurance carrier. The second is the AIA standard subcontractor agreement. Merge the two into one clean, legally coherent template that incorporates the insurance requirements from the first document and the standard protections from the AIA version. Format it as a fill-in template that auto-populates the subcontractor's name, trade, and project address."

Ted sent the rebuilt agreement out over a weekend and got three signatures back, including from a title contractor who had been stonewalling him for months. The issue was never the legal language. It was the friction of a messy document and a slow follow-up process.

1. The context document is the whole game. Generic questions produce generic estimates. A model that already knows your foam costs, your trim rates, your market, and your margin target produces line items. Build the context document once. Update it when your costs change. Every estimate after that is faster and more accurate than the last.

2. Margin and markup are different numbers. 25% markup gives you a 20% margin. 33% markup gives you a 25% margin. Most builders think in markup because that's how they were taught. AI estimating models work better with margin targets because margin is what actually shows up in your bank account. Rebuild your context document around the margin you want and let the model back into the price.

3. Unsigned subcontractor agreements are a process problem. If you have subs who haven't returned agreements, the issue is rarely the terms. It's the friction: a long document, a manual process, no follow-up. One clean auto-fill template sent the same day you bring a sub onto a job changes the conversion rate. Ted got a signature from someone who'd been slow-walking it for months.

4. Know your math before you hand it off. AI is a faster calculator, not a smarter one. It calculates what you tell it to calculate. If your cost inputs are wrong, the estimate is wrong faster. Run your most critical line items manually, verify them with the model, and treat any variance as a question worth answering before the proposal goes out.

5. Your clients may already know what the job costs. The 21-page report Ted received was not an anomaly. High-end clients have access to the same tools builders do. Some of them use those tools before they call anyone. If your proposal does not reflect what the market will bear in your area, you will either leave margin on winning bids or lose to clients who know you underpriced it.

Beyond the Bid Circle is a peer community for builders scaling with systems and AI—the same builders who've escaped the 60-hour grind. Swap war stories, shortcuts, and real strategies with people who get it.