We Gave Claude a Job. It Became Our Sales Coach.

Watch the full episode of AI Marketing for Remodelers and Builders above.

Most of what gets said about AI right now is theory. It's going to change everything. You should be using it. Great. How?

This is the opposite of that. On a Friday afternoon, I sat down with Claude and walked away about 15 minutes later with a tool that reviews our sales calls, scores them against a rubric, and writes up coaching feedback. Automatically. Every day. I don't write code. I didn't buy software. I described the outcome I wanted and let the thing build it.

I'm Spencer, founder and CEO at Builder Funnel. The full story is in the video above, where Kai and I walk through the whole build. Below, I'll break down how it actually came together, what it got wrong on the first pass, and how you can build the same kind of thing inside your own business this week.

In this blog you'll learn: how I built an AI sales-call coach in Claude with no code, why version one scored a closed deal at 2.5 out of 5, how I corrected it with plain feedback, what it costs, and how to apply the same approach to any recorded meeting in your business.

Table of Contents

• The Problem: There's Never Enough Time to Coach
• The Build: 15 Minutes, Zero Code
• Version 1 Was Wrong in the Most Useful Way Possible
• How It Got Good: Feedback on the Feedback
• Why Build It Yourself Instead of Buying a Tool
• What's Next: From Single Reports to Trends
• This Was Never Really About Sales Calls
• Go Build Something
• Frequently Asked Questions

The Problem: There's Never Enough Time to Coach

If you run a remodeling or building company, you know this feeling even if "sales manager" isn't on your business card.

Someone on your team runs the pre-qualification calls or sets the appointments. To make them better, somebody has to review those calls and give feedback. At Builder Funnel, that somebody is me. I watch the call, pause to take notes, flag the good moments and the missed ones, then walk the rep through it.

It works. It's also slow. The bottleneck comes down to a few things:

• Volume. You can only watch so many calls in a week.
• Lag. You might review a Tuesday call on Friday, after the rep has already run 15 more without that feedback.
• Consistency. Your notes on call number ten aren't as sharp as your notes on call number one.

The coaching is good. It's just arriving after the moment it could've changed something. More feedback, faster feedback. Every rep wants it. Almost no sales manager has the hours to give it.

The Build: 15 Minutes, Zero Code

Here's the part that caught me off guard. I wasn't trying to build software. I opened Claude (using Cowork, the desktop feature that sits between the simple chat and the full coding environment) and assumed I'd have to prompt my way into some script I'd then copy and paste somewhere to actually run.

Instead, I just described what I wanted, see my prompt below:

"It would be great if you could create a bot or an agent that every day combs through all the discovery calls for my sales team, analyzes them through some scoring system, and produces a report with the feedback."

Claude came back with a few clarifying questions. Discovery calls only, or all sales calls? What scoring scale, 1 to 5, 1 to 10? I picked my answers, waited a couple minutes, and Claude said it was done, and it would run every day at 5 p.m. I ran a sample on the spot. Claude connected to Fireflies, where we record our meetings, pulled a real discovery call, scored it across ten categories on a 1 to 5 scale, and handed back a clean, formatted report. Tables and all. Total time to a working version one was about 15 minutes.

What Are the Three Ways to Use Claude?

This is the part Kai walked me through on the episode. There are three ways in, and they're worth knowing before you start:

• Claude chat: the web or desktop app. Plug and play, no setup.
• Claude Code: runs in a terminal, built for coding. More power, more setup.
• Cowork: the middle ground. The power of Code with the simplicity of chat.

I used Cowork. If "open a terminal" makes you want to close the tab, that's the one to start with. If you want a wider view of the tools worth trying, we put together a rundown of the best AI tools home builders and remodelers should be using.

Version 1 Was Wrong in the Most Useful Way Possible

The first report was genuinely good. It also got a few things wrong, and that's the part worth slowing down on.

I gave Claude almost no direction on the rubric beyond "use a 1 to 5 scale." On its own, it broke the call into about ten categories:

• Rapport and tone
• Agenda setting
• Discovery and question quality
• Pain identification and impact
• Decision process
• Competitive landscape
• Talk-to-listen ratio
• Objection handling
• Clear next steps
• Overall effectiveness

It scored each one, then wrote up the top three things the rep did well and the top three to work on. The rep on that first call, Tony, scored a 2.5 out of 5. He also closed the deal.

Why Did a Closed Deal Score Only 2.5 Out of 5?

Because the AI flagged two things as problems that were actually working as designed.

• Jumping straight into questions. The AI dinged Tony for skipping the warm-up. That's our script. These calls run 15 to 25 minutes. There's no five minutes of "how's-the-weather" by design, because the business owner is there to solve a problem and so are we.
• A long pitch in the back half. It flagged that Tony "dominated the back half with seven to eight minutes of uninterrupted pitch." Also on purpose. We ask a lot of questions up front, figure out whether we can actually help, and only then walk through the next step.

Once I gave the agent this feedback, it adjusted. Think of it like coaching the coach.

How It Got Good: Feedback on the Feedback

Getting from version one to a version we'd actually run took maybe one to two hours of back and forth, spread across a couple days. None of it was technical. All of it was the kind of feedback you'd give a new sales manager you were training.

• I corrected the rubric. Anywhere the scoring didn't match how we actually sell, I said so, and it updated.
• I added a "how to get to a 5" section. Now every report doesn't just say a category scored a 2, it says exactly how it could've been a 5. Ask these two questions. Use this tone. Concrete beats vague every time.
• I looped in the rep. I sent the feedback to Tony and asked him to grade it. First round, we both spotted a few misses. Latest round, his reaction was "Wow, this is really good."

That last move matters more than it looks. The fastest way to make AI feedback worse is to hand it down from the higher ups and treat it as gospel. The fastest way to make it better is to put it in front of the person receiving it and ask them where it's wrong. Nobody wants to feel like an algorithm is grading them. Everybody's fine with a tool that helps them get better, especially when they had a say in how it scores.

Why Build It Yourself Instead of Buying a Tool

There are tools that score sales calls. Why create your own? A few reasons:

• You own it and you control it. This one does exactly what we sell, scored against our actual discovery script, in our language. When our process changes, we change the agent in a sentence.
• It's niche on purpose. Nobody else needs a tool that grades calls against your specific script. That's exactly why it's worth building, because no off-the-shelf product is ever going to bother with your one weird thing.
• The math is simple. You're looking at $100 to $200 a month for a Claude subscription. Compare that to what you'd pay a sales consultant to review the same volume of calls, or the cost of the deals a rep fumbles while waiting on feedback.

None of this means ripping out the platforms you already run on. The big tools earn their keep, and picking the right one still matters, that's why we wrote up the best construction CRMs for home builders and remodelers. The move is using AI to handle the specific things those platforms don't quite do for you. You love your CRM for 80% of what you need. AI is how you close the gap on the other 20% without waiting on someone's product roadmap.

What's Next: From Single Reports to Trends

Version one reviews one call at a time. The next step is what happens to all that feedback after, and it's the thing Kai pushed me on when I showed him the build.

Right now, each report stands alone. The opportunity is to give it a longer life. Store every report, then roll them up:

• Monthly recaps per rep, so coaching has a paper trail.
• Three-month trends, so a rep can see "I've scored low on discovery two months running, and here's my progress since."
• Team-wide patterns, so a manager sees the whole pipeline instead of one call at a time.

Think of it like a year-in-review for your sales team, built on real coaching data instead of vibes. If you're still mapping how those calls fit your pipeline in the first place, start with how to build a sales funnel that actually works for home builders.

This Was Never Really About Sales Calls

Sales calls are just the easiest place to start, because most of them are already being recorded. The same approach works anywhere you have a repeatable conversation with a standard you're trying to hit:

• Client check-in meetings your project managers run
• Design presentations
• Production meetings
• In-person consultations (most phones have a voice recorder you can set on the table, after you ask permission)

If there's an agenda or a right way to run it, you can record it, build a rubric, and finally get consistent review on meetings that have never been reviewed once in the history of your company. And in an industry where talent is the next big constraint, tools that make your existing people better are worth more than ever.

The first step is the boring one. Start recording your calls. Phone pre-qualification calls are the easiest entry point, use CallRail, use Zoom, use a recorder on the table. The recordings are the foundation here, and you can't build coaching, training, or marketing on a foundation that doesn't exist yet. Once they're in place, everything else opens up.

Go Build Something

People use AI to draft things, and it's fine. They use it to crunch data, and it's great. But the moment it really clicks is when someone builds something that does a real job in their business. That's when they realize how much was possible that they'd assumed wasn't.

I'll hand you the same challenge we gave our listeners. Pick one process you've always thought "I wish this were easier." Get Claude on your desktop, open Cowork, and describe the outcome you want. Not the how. The what. Then give it feedback like you would a new hire, and keep going. Give it an afternoon. That's usually all it takes.

Frequently Asked Questions

How do you turn Claude into a sales coach?

Describe the outcome you want in plain language, connect it to wherever your calls are recorded (we use Fireflies), and let it propose a scoring rubric. Then refine it the way you'd coach a new sales manager: tell it where its scoring is wrong, add a "how to reach a 5" section, and have the rep being reviewed sanity-check the feedback. No coding required.

How long does it take to build?

A working version one took about 15 minutes. Getting it to a version we trusted took another one to two hours of back-and-forth spread over a couple of days, all of it plain feedback, none of it technical.

What does an AI sales coach cost?

For a setup like this, you're looking at $100 to $200 a month for a Claude subscription. That covers reviewing every call, every day, which is more coverage than a busy sales manager can deliver by hand.

What do you need before you can do this?

Recorded calls. That's the real prerequisite. Phone pre-qualification calls are the easiest entry point. Tools like CallRail or Zoom handle the recording, and for in-person meetings a phone voice recorder works once you've asked permission.

Is this only useful for sales calls?

No. Any repeatable conversation with a standard you're trying to hit works: client check-ins, design presentations, production meetings, in-person consultations. If there's a right way to run it, you can record it, build a rubric, and get consistent review.

Building something with AI in your business? We want to hear about it. Email Kai at kbiami@builderfunnel.com and tell us what you're going to build after watching this. We'll test it out.

And if you'd rather have a team that turns AI into real pipeline and closed projects for your company, set up a meeting with us.