One of the best uses of AI I have found did not start with AI. It started with standing in a driveway watching people work too hard.
The Takeaway
AI did not find this problem and it did not decide what to fix. I did that, by showing up three times and paying attention. What AI changed is that noticing was suddenly enough, because the distance between seeing a problem and having a working tool in someone’s hands collapsed.
What Happened
The first one
My son Levi is a loan originator and my partner. We work together as Team Duncan.
We sponsored a local caravan. Sponsoring means we brought breakfast, introduced ourselves and the company, and then rode along with everybody for the morning.
It was a great morning. Good people, good houses, worth the time.
It was also a little frazzled, and it took me a while to understand why.
A lot of the agents there had never been on a caravan before, so there was a learning curve happening live in a parking lot while everyone tried to keep up. We rode the whole route with the group, and here is what that looked like:
- People digging for a pen. Then digging for something to write on.
- Notes going onto whatever was handy, balanced on a steering wheel between stops.
- Somebody punching the next address into their phone while the car ahead pulled away.
- Anyone who lost the group having to work out where the group went.
Then we got back, and the real work started. One person trying to remember everything that got said at eight houses. Another person adding up scores by hand. All of it from memory, at the end of a long morning, when nobody is sharp.
It worked. It just cost far more than it should have.
The second one
We sponsored again. Different coordinator, also excellent at the job.
This time I found out what the coordinator actually does before anyone shows up. Three and a half hours, minimum. Identifying the properties, building the schedule, lining up who is going where. That is a half day of unpaid work every week, done by a volunteer, and almost nobody riding along knows it is happening.
That morning we missed a property entirely and had to double back, because it was not identified correctly on the schedule.
And there are real rules underneath all of it that most people do not realize:
- A property needs a representative present. No rep, the house gets skipped.
- If a rep does not complete the full route and come back at the end, that property is out.
- Out means no eligibility for the award.
Why the award matters
The winner gets a sign in the yard. Caravan of the Week.
That sign is not a trophy. It is advertising pointed at exactly the right audience, because every agent in the local MLS sees that property. And for the seller, it is proof their agent is out working on their behalf, which is the thing sellers most often suspect is not happening.
So the stakes are real. And the entire thing was running on pens, memory, and one volunteer’s morning.
What I built, and what I did not
Let me be precise, because the honest version is smaller than the impressive version.
I did not touch the coordinator’s prep. Those three and a half hours are still three and a half hours. Pulling properties, building the route, lining up the schedule. That work is unchanged and I would not claim otherwise.
What I built fixes everything after that. The clunky part. The field, and the arithmetic, and the reporting.
The coordinator’s side:
- Drop in the list of properties
- Add the participating agents and their emails
- Drag and drop to set the order houses appear
The agent’s side, on their phone, out in the field:
- Talk into it or type, whichever suits the moment
- Make notes walking up to the house instead of balancing a notepad on a steering wheel
- It saves as you go, so nobody loses anything
At the end:
- Scores calculate themselves
- Every listing agent gets a summary of what the group actually said about their house
- The winner is announced
No pens. No adding anything up. No trying to remember what somebody said three houses back.
The third one
Third caravan, I put the tool in live. Everybody was using it. Real event, real agents, real scores.
I was scared half to death.
Think about what failure looks like there. Not a bug report. A morning of work, from a dozen people, gone. Every note, every rating, at every house. A field full of agents who took a chance on the mortgage guy’s software and got nothing back for it. And I would have been standing right there when they found out.
There is no soft version of that. I put myself out there on it.
It worked. It worked really well.
Why It Happened
I want to be careful about the lesson here, because the wrong one is easy to grab.
AI did not find the problem. No model was going to tell me that agents are writing notes on steering wheels, or that a coordinator gives up half a day every week. I learned that by sponsoring, riding along, and asking questions of people who were too polite to complain.
AI did not decide what mattered. Voice notes came from watching people try to write while walking. Autosave came from watching how much a person stood to lose. Drag and drop ordering came from watching a coordinator build a route by hand. Every real design decision came out of that driveway, not out of a chat window.
AI did not know what it could not fix. Deciding to leave the prep work alone and go after the clunkiness was a judgment call that required understanding the process well enough to know where the software could actually help.
What AI did was make noticing sufficient. That is the whole story. A few years ago this idea dies right here in the post. I am a mortgage guy. I would have described it to somebody, gotten a quote, and let it go. Now the gap between seeing a problem and putting a working tool in someone’s hands is small enough for one person to cross.
The scarce thing used to be the ability to build. Now the scarce thing is understanding a process well enough to know exactly where it hurts. That part has always been free, and it has always been the harder half.
What This Means for Your Business
Go where the work happens. I did not think my way to this. I rode the route, three times. The best problems to solve are invisible from a desk, because the people living with them stopped complaining years ago.
Watch for workarounds, not complaints. Nobody said the process was broken. They just kept a pen in the console and a clipboard on the seat. A workaround is what a problem looks like after people have given up on it being fixed.
Know which part you are actually fixing. I could not give the coordinator her half day back. I could give everyone the rest of it. Being clear about that made the tool better, because I stopped trying to solve a problem I did not understand and fixed the one I did.
Then put it in front of real people and accept that it might fail. Mine went live at a real event with real agents and real consequences. That was uncomfortable and it was the only way to know.
One Thing to Try This Week
Think of one thing in your business that is annoying every single week and that everyone has quietly accepted.
The spreadsheet you rebuild. The message you retype. The report you assemble by hand every month.
Do not ask AI to help you do it faster. Describe the whole thing, start to finish, including the parts you think are unavoidable, and ask which part of it does not have to be a task at all.
The answer is usually not the whole thing. But it is usually more than you expected.
The caravan tool is built to be branded. If your brokerage or association runs caravans and any of this sounds familiar, it can carry your name on it.
About the Author
Clay Duncan is a Huntsville, Alabama Mortgage Loan Originator with Princeton Mortgage, NMLS #118739, and runs practical AI training for real estate professionals across North Alabama. He is currently building a suite of AI-powered tools for REALTORS®. This series covers what he learns doing it, including the parts that break.
If You Want to Go Deeper
Clay’s AI training for REALTORS® is built for practical adoption: what to use, when to use it, and how to make it stick. Sessions run for offices, teams, and partner groups.