Referral work is better work. A lot of operators would say the same.
You run on a handful of beliefs like that, built from years of jobs. Reports could always tell you part of it, never the whole thing.
You remember the referral who signed the estimate without blinking. You remember the lead you paid for that turned into three trips and a job you lost money on. The ones that got quoted, got done and got paid with no trouble don't stick with you the same way, and most of your money came from those.
Comparing how referral jobs performed against the leads you paid for has been possible for years. Most software does it. What nobody was going to do is sit down and cross-reference every estimate won or lost, every job done, and every conversation with a customer. But all of it together is what tells you whether referral work really is better work, where it stops being better, and what the ones who never call you back had in common. Now you can do it in one sitting.
You export your job history to a file, or connect your software directly, and hand it to whichever AI you use. If most of it lives in your email, that connects too. The records don't have to be tidy. You're asking about jobs that are already done and paid for, not building anything on top of them. Sorting through the mess is the AI's job now, not yours.
Most of what you believe will hold. You have watched a lot of jobs go right and wrong, and that counts for something. The likely answer is that you were right all along. But when a belief doesn't hold up, you find out where it's off. Referrals argue less about price but go quiet on your estimates just as often as anybody else. Or the tire-kickers all came from one place and the rest were fine.
Your beliefs set the budget. They shape how you deal with customers and how the work gets done. When was the last time you tested one against your own records?
THE VERDICT
Pick the belief you'd defend hardest without looking anything up. Then put your records in front of an AI and ask.
Do it before you act on it again.