Can You Trust the Number?
Once real money is involved, it's the question that decides everything. Any AI demo looks great. What matters is whether you can put its answer in front of your bank or your auditor and have it hold up. That's what we build for.
The demo always works. Then real money shows up.
Type a question into any AI tool and you'll get an answer in seconds, clear and confident. Ask it "what's my warranty cost per closing in Heron Landing?" and a number comes right back. Impressive.
Now ask it again an hour later. Sometimes the number's different. Ask where it came from and it can't really tell you. And when it hits a question it doesn't have the data for, it doesn't say "I don't know." It makes something up, in the same confident voice it used when it was right.
That's fine when you're drafting an email. It's a real problem the moment someone acts on the number before anyone notices it's wrong. Approves the draw. Sets the price. Walks it into the board meeting. A confidently wrong number is more dangerous than no number at all, because people believe it.
The one idea worth taking away
An AI should never be the thing doing your math. Its job is to figure out what you're asking and pull the right information. The number itself gets calculated the same, provable way every time. Ask the same question tomorrow and you get the same answer, and you can always see where it came from. That one boundary is what separates a party trick from something you can run a business on.
Where a Wrong Number Actually Bites
In homebuilding, the numbers don't stay on a screen. They go somewhere that matters.
Draw requests
Money leaves the building on the strength of a number. If the AI that checked it was guessing, you find out after the money's already gone.
The board & your investors
A margin figure that changes every time you pull it isn't a metric. It's a liability. A board number has to read the same today, next quarter, and when someone challenges it.
The bank
Your lender doesn't accept "the AI said so." Every number you hand them needs to trace back to something they can check for themselves.
The audit
When the PBC list lands, "where did this come from?" is the whole exercise. Numbers that can't show their work turn a two-week audit into a six-week one.
What Separates a Toy From a System You Can Run Your Business On
Three things. None of them have to do with how smart the AI sounds. They're all about whether you can trust what it hands back.
The same answer every time
Ask the same question twice and you get the same number. Today, next week, and when the board asks again in six months. The trick is that the AI never does the arithmetic itself. It works out what you're asking, and a proven calculation runs the math the same way every time. The answer holds still because nothing is being re-guessed.
Traceable back to your data
Every figure links back to where it came from: the invoice, the contract clause, the ledger line. When someone asks where a number came from, you can click straight through to the record behind it. That's what makes it hold up in front of a lender or an auditor. It shows its work.
The right people see the right numbers
Payroll doesn't leak to sales. A division manager sees their own division, not the whole company's margins. That kind of access control is built into the system from the start, so "who's allowed to see this?" always has an answer. A document the AI hasn't been cleared to share stays out of reach until someone clears it.
This Isn't a New Problem to Me
Getting the numbers right isn't new to me. It's what I've spent 35 years doing, in industries where a wrong figure costs millions and "the computer made it up" was never going to fly.
- Over $1 billion a year in payments, validated. I architected a state health agency's platform that checked provider claims against the contract, flagged the exceptions, and paid the rest. It's the same check-it-against-the-rules work that sits behind your draws and AP, and it earned a national award for state IT.
- Eleven years running models where a mistake cost real money. I built and ran the machine-learning systems behind a nine-figure portfolio at one of the world's largest quantitative funds. It ran live, every market day, with no room for a wrong answer.
- Thirty-plus systems unified into one trustworthy view. Today I'm the technology leader at an FDA-regulated manufacturer, building a platform that pulls 30+ systems into a single real-time data model, with AI running in production right now in a business where mistakes carry real weight.
5 Questions to Ask Any AI Vendor
Whether or not you ever work with me. If a vendor can't answer these cleanly, keep your checkbook closed.
1. "When I ask the same question twice, do I get the same answer?"
If the number moves between Tuesday and Thursday, it can't go in a board deck. You're looking for "yes, always." "Usually" isn't good enough.
2. "Where did this number come from — can you show me?"
Every figure should trace back to a source you recognize. If they can't show you, don't trust it.
3. "What happens when the AI doesn't know?"
The right answer is "I don't know," or "here's what I found." A confident guess is the most expensive mistake this technology makes.
4. "Who can see what?"
Your payroll and your buyers' data shouldn't be one prompt away from the wrong person. That has to be built into the system.
5. "What do I own when we're done, and can my team maintain it?"
If it only keeps working while the vendor is the one who understands it, you don't own a solution. You own a dependency. You should own what gets built.
Want AI You Can Actually Put in Front of Your Bank?
Start with a 30-minute call. No pitch deck. Just a straight conversation about where your numbers come from today, and where AI can take work off your team without leaving you wondering whether to believe it.
Set Up a 30-Minute Call