Why We Built Velzyx

The honest answer is that I kept watching small businesses pay a tax for being small. Not money, exactly. Time. Attention. The slow erosion of the thing that made them good in the first place. I built Velzyx because I thought I could remove that tax in three specific places, and because I had stopped believing that horizontal platforms were going to do it for them.

I want to walk you through how I got here, because the founding story matters more when the company is asking you to trust it with a part of your business that actually works. Voice. Lead capture. Underwriting. These are not nice-to-haves. If we get them wrong, real people in real practices and brokerages lose real revenue, and they should know who we are and what we actually believe before they let us anywhere near it.

The dental office that wouldn’t leave my head

In 2024 I was sitting in a dental office in Costa Mesa waiting for my own appointment. The front desk had two people on it. They were good at their jobs. They were also, that morning, completely underwater. One of them was on the phone trying to explain why a hygiene appointment got moved. The other was checking a patient in while three more people stood in the lobby, and the phone kept ringing, and every time it rang she had to make a decision: pick up, or finish the human standing in front of her. She picked the human, which was right, and the call rolled to voicemail. I watched five calls roll to voicemail in twenty minutes.

I asked her later, half curious, what happens to those calls. She said, “Some of them call back. Most of them don’t.” And then she said something that stuck with me: “The new ones almost never do.” New patients. The most valuable call type in the building. Rolling to voicemail because the front desk was doing the right thing for the patient in front of them.

I went home and pulled the math on a back-of-envelope. The practice was, conservatively, missing six to eight new patient calls a week. Average case value in the kind of work they did sat north of two thousand dollars. Even if you assume most of those callers would have shopped around anyway, the lost revenue from the ones who would have converted was meaningful. Not catastrophic. But it was the kind of slow leak that you stop noticing because it’s been there forever, and the practice owners I started talking to didn’t see it as a fixable problem. They saw it as the weather.

Three industries, one pattern

What got me was that this exact shape kept showing up everywhere I looked.

A friend of mine sells eight-figure homes on the coast. He is, by any measure, excellent at his job. But his website was a builder-template thing with a hero video that took six seconds to load, and his lead follow-up was a Google Sheet his assistant updated when she had time. He was losing inquiries from buyers with real money because his digital surface area looked nothing like the product he was selling. When a buyer Googles a $12M listing at 9pm on a Sunday and lands on a site that feels like a 2014 agent page, that buyer’s next click is your competitor. You don’t hear about that loss. It happens silently, in a browser tab, while you’re at dinner.

Then a commercial real estate analyst I’d known for years told me he’d spent two days underwriting a single retail deal. Two days. Pulling rent rolls, normalizing leases, modeling expense growth, running sensitivities. He’s sharp. The work was real. But two days of analyst time on a deal that, in the end, his shop passed on, is two days that didn’t get spent on the deal they would have done. CRE underwriting is, for the most part, the same handful of operations repeated against different documents. There is real judgment in it, but the judgment is maybe twenty percent of the time. The rest is reformatting and rechecking.

Three different industries. Three completely different jobs. But the same shape: a small business with a great operator at the center, getting slowly worn down by work that should not require their best hours.

The tax of being small isn’t money. It’s your best hours getting eaten by work that doesn’t deserve them.

The horizontal platform problem

The obvious objection at this point is: this is what big AI platforms are for. Just plug in a general assistant. Let the model figure it out.

I tried that. I really did. In early 2025 I spent about three months building what was effectively a thin wrapper around a general language model and pointing it at the dental front-office problem. It worked, in the demo. It fell apart in the field.

It fell apart because dental scheduling is not a generic scheduling problem. It is a problem about hygiene recall cycles, insurance verification timing, the difference between a limited exam and a comprehensive, whether the operatory you’re booking has the equipment for what the patient needs, whether the patient is overdue and the hygienist wants to see them sooner, how long the doctor blocks for a crown prep versus a seat. A general model can talk about teeth. It cannot run a column in an operatory schedule without being taught exactly how that practice runs it. And every practice runs it slightly differently.

The lesson I took, which sounds obvious in retrospect but wasn’t at the time, is that the model is the easy part. The hard part is the workflow. The hard part is knowing that in this office, hygiene recall is six months but they push to four for periodontal patients, and that the doctor wants new patient exams blocked at 60 minutes not 30, and that the front desk needs the call summary in the practice management software before the patient gets to the door, not after. None of that is in any model’s training data. It has to be learned per business, and then it has to be wired into the actual systems the business already uses.

The bet

So the bet I’m making with Velzyx is straightforward, even if it’s unfashionable. I think specialized AI products, built deeply into specific verticals, are going to beat horizontal AI platforms for small and mid-sized businesses for the next several years. Not forever. But for long enough that building three of them is a reasonable use of a decade.

That’s why we’re three products instead of one. Aria Dental AI is the front office for dental and service businesses, built per practice. AgentCentric is the website and lead system for luxury real estate. AnalytixCRE is the underwriting tool for commercial real estate analysts. They share infrastructure underneath, but the surfaces are completely different because the jobs are completely different. I would rather have three products that each do one thing as well as it can be done than one product that does five things at seventy percent.

What “specialized” actually means

I want to be specific here, because the word gets thrown around. When I say Aria is specialized, I mean that when a new patient calls a practice we’ve onboarded, the system already knows the practice’s new-patient flow, the insurance plans they accept, the doctor’s preferred block lengths, which operatory has the laser, how the front desk wants the chart note written, and where in the practice management software the appointment needs to land. None of that is configuration in a settings panel. It’s built into how the agent runs, and we build it during onboarding by sitting with the practice and watching how they work.

When I say AgentCentric is specialized, I mean every site is built for the way luxury buyers actually shop, with the lead capture wired into a follow-up cadence that an actual top-producing agent helped us design, not a generic CRM template. When I say AnalytixCRE is specialized, I mean the underwriting model knows the difference between a triple net retail lease and a modified gross office lease the way an analyst does, and it surfaces the things a senior would flag, not the things a textbook would.

A moment I almost quit

I’ll tell you the moment I almost stopped. It was August of 2025. We had three early Aria deployments live. One of them had a bad week. The agent mishandled a call from a patient who was, it turned out, calling about something fairly emotional. Not a clinical emergency, but a real life thing. The agent gave a correct, polite, useful response, and it was completely wrong for the moment. The patient called the practice owner to complain, the practice owner called me, and I sat in my car in a parking lot in Newport Beach for forty-five minutes feeling like I had built something that was going to cause real harm at scale.

What I came out of that with, eventually, was the conviction that the product is not the AI. The product is the handoff. The thing we sell is not “an AI that handles your calls.” The thing we sell is a system that handles the calls that should be handled, and gets out of the way fast and gracefully for the calls that shouldn’t. We rebuilt the escalation logic the next week. We changed how the agent listens for emotional cues. We made the human handoff faster and cleaner. And we wrote into our own onboarding playbook that we will tell every practice, in plain language, exactly what the agent is bad at, so they can decide whether they want to deploy it at all.

I think a lot about that parking lot. The lesson wasn’t “don’t build AI for sensitive industries.” The lesson was “know exactly where your product ends.”

What Velzyx is, and what it isn’t

Velzyx is a holding company for three specialized AI products serving three specific industries. We are small on purpose. We sell directly. We build per customer where it matters and reuse infrastructure where it doesn’t. We do not have a sales motion built on fear or urgency. If you’re a dental practice owner reading this and Aria isn’t a fit for the way you run your front desk, I would rather you find that out on a call with me than on a contract.

We are not trying to be a platform. We are not trying to be everywhere. We are not building general AI for businesses. I am suspicious of any company that tells small operators that one tool will fix their phones and their website and their underwriting and their billing. The integration work to do any of those well is real and it is per industry. We chose three because three is what we can do well right now.

What I want this company to feel like

I want Velzyx to feel like the company you call when you want someone who has actually sat in the office, watched the work, and built the thing for the way your business actually runs. Not a vendor. Not a platform. A small team that takes one specific job seriously per industry and ships software that earns the right to handle it.

If that resonates, the door is open. If it doesn’t, that’s fine too. The point of writing this is not to convince anyone. It’s to put on record, in my own words, why we’re doing this, so that when we make a hard call later you’ll know the frame we’re making it from.

Thanks for reading.