Real businesses. Still running.
These are real businesses running on software we designed, built, and still operate. Not concepts, not mockups. Every line reviewed by a senior engineer. Below: the problem each one solves, what we built, and the capability it delivers. We describe what's real and skip the inflated numbers.
Salisbury Bookkeeping
A fractional CFO practice for contractors, running on a marketing site and a complete in-house CRM we built from scratch. No HubSpot, no third-party marketing platform, no per-seat tax.
A small accounting practice that wants to grow has to stitch together a CRM, an email platform, a scheduler and an outreach tool, then pay per seat for all four and still do the follow-up by hand. The tools do not talk to each other, so the work of chasing a lead never actually goes away.
The whole stack, owned outright: a Next.js marketing site, a custom CRM with a visual workflow builder, an email engine that sends through the Gmail API with its own open and click tracking, a cold-outreach machine with warm-up pacing and deliverability controls, and a reply handler that classifies incoming mail with Claude and drafts a response for approval.
A live practice running entirely on software it owns, with an admin console where four AI agents (Email Manager, Workflow Doctor, SEO and GEO, Daily Briefing) watch the system and report what needs attention. The brand repositioned to fractional CFO work on the same platform, with no migration.
Salisbury Real Estate
A full custom real-estate CRM and public site for a Utah agent team, including a nightly pipeline that turns expired MLS listings into contactable leads.
Agents rent their entire business from someone else. The CRM, the website, the lead source and the follow-up all sit on platforms that own the data and charge per contact. The leads that matter most, expired and withdrawn listings, go to whoever contacts the owner first, and doing that by hand does not scale.
A public marketing site and a complete CRM on one codebase, with five nurture workflows for sellers and buyers, a nightly job that pulls expired, cancelled and withdrawn MLS listings and runs skip tracing to produce contactable leads, an off-market deal to social-post pipeline for investors, and email that sends from the team's own domain with open and click tracking we wrote rather than rented.
A team running on software they own, at salisburyre.com. Incoming replies are classified and drafted by AI for a human to approve before anything sends, and a four agent back office handles the routine work. The lead pipeline, the data and the sending reputation all belong to the business.
Auto Advisor
A consulting firm for auto shops and dealerships, plus the multi-tenant AI platform it installs. The software is included in the engagement, not sold per seat.
Independent shops and dealerships already run a shop-management system and a DMS, and they are not going to replace them. Consultants hand over a binder and leave; software vendors demand a rip and replace. Neither one changes what happens on the service drive on Monday morning.
A multi-tenant platform that sits on top of the systems a shop already runs, with a fifteen agent crew, a KPI engine, an action queue that turns findings into assigned work, a learning loop that improves from corrections, and per-employee access grants. Built for three segments at once: independent repair, used-car and aftermarket dealers, and franchise new-car stores.
A consulting business where the software is the delivery mechanism. The advice does not sit in a document, it arrives as work in a queue attached to the numbers that produced it, and the platform keeps running after the engagement instead of expiring with it.
Top Builder AI
An AI back office for field-service contractors that installs on the software they already run, and automates the part their platform never did.
Field-service platforms automated booking and dispatch and then stopped. Money, materials, people and paperwork are still handled by a person in an office, on a spreadsheet, at the end of a long day. That is where the margin leaks, and it is the part no vendor wanted to touch.
Eight self-learning agents that install onto ServiceTitan, QuickBooks and nine other field-service platforms, and run the back office rather than the front desk. They are advisory by default: the agent proposes the action and a human approves it, so the contractor keeps control while the routine work stops piling up.
A product that meets contractors where they already are. Nothing is ripped out, the existing platform keeps doing what it is good at, and the work it never automated gets handled by agents that learn the business as they go.
Neuron SEO
Our own automated AI-search engine. It measures where ChatGPT, Perplexity and Google's AI cite a brand and where they do not, then builds and publishes the content that closes the gap. We ran it on ourselves first.
Customers increasingly ask an AI assistant who to hire before they ever open a search page. If the assistant does not know a business exists, the job goes to whoever it names instead. Almost nobody can tell you which buying questions they are cited for today, which makes every content decision a guess.
A closed loop that runs on a schedule. It asks the real engines the real buying questions and records who gets cited, finds the questions where the brand is absent, writes a fact-dense answer page and a long-form video for the gap, fact-checks the result and validates its structured data, then deploys and publishes to the brand's own site and channels. Every step is logged.
Six brands running on it, twice weekly. Because the loop measures, the results are countable rather than claimed: our own citation wins are listed page by page, with the engine and the date, on a page generated from the data itself.
Custom agents and full automation
The thread running through every build above. Each business got its own agents, written for its own work, wired into the systems it already uses.
Most AI automation is a chatbot bolted to a website, or a chain of no-code steps that breaks quietly and tells nobody. Neither one touches the work that actually consumes a week: the follow-up nobody sent, the report nobody ran, the reply sitting unanswered in an inbox since Tuesday.
Agents that do specific jobs inside a specific business, and the plumbing that makes them trustworthy: work that runs on a schedule, a queue that retries what failed instead of dropping it, a record of every run, and an approval step wherever an action reaches a real customer. Reply classification and drafting, deliverability monitoring, KPI review, daily briefings, content production and publishing.
Automation you can leave running, because it fails loudly and asks permission before it speaks for you. And because we hand over the code and the repository, the agents are an asset the business owns rather than a subscription it rents.
What these systems actually deliver.
The same building blocks we ship — an answer-on-the-first-ring front desk, instant lead follow-up, and automatic appointment reminders — have a long, documented track record of recovering revenue for local service businesses. Below are real, third-party results from the category: a missed call answered, a lead reached in minutes instead of hours, a no-show that never happened.
In plain terms: these are documented results from the category — peer-reviewed studies and named industry research — not Neuron HQ's own client outcomes. We're a new AI-agent practice and don't yet have client numbers to show. What we can show is that the model works, with a source you can open behind every figure.
Why does answering a lead in 5 minutes matter so much?
Because the window closes fast. The MIT / InsideSales.com Lead Response Management study — three years of data across 15,000 leads and 100,000 call attempts — found that contacting a new web lead within 5 minutes instead of 30 makes you 21 times more likely to qualify it. Harvard Business Review's audit of 2,241 U.S. companies reached the same conclusion: firms that responded within an hour were nearly 7 times more likely to qualify a lead than those who waited even one hour longer. An AI agent answers in seconds — every time, day or night.
How many calls is a local business actually missing?
More than owners think. Invoca's platform data shows 27% of calls to home-services businesses go unanswered — roughly one in four new-customer conversations dropped before it starts. And the ones you put on hold don't wait: Invoca's Buyer Experience report found 75% of consumers hang up rather than stay on hold, with only 6% willing to hold for 30 minutes. An AI receptionist picks up on the first ring, books the appointment, and never sends a paying caller to voicemail.
Do automated reminders really cut no-shows?
They do, and the effect is large. In a randomized controlled trial published in the International Journal of Pediatrics, automated text reminders cut the no-show rate from 38.1% to 23.5% — a 14.6-percentage-point drop (p = 0.04). A 2020 systematic review of 20 studies backs it up: patient reminders produced an average 41% reduction in missed appointments, with 95% of studies showing a positive effect. Every recovered slot is revenue that would otherwise have walked out the door.
Is the phone still where high-value bookings happen?
Yes — especially for the bookings worth the most. Invoca's Buyer Experience research found 68% of consumers prefer calling a business over any other channel for the human connection, and 30% say they call specifically because they're most comfortable making a high-stakes purchase by phone. A new patient or homeowner is far more likely to book by voice than fill out a form — which is exactly why a front desk that always answers, in a natural voice, protects your highest-value revenue.
| What the system does | Documented result | Source & type |
|---|---|---|
| Instant lead follow-up (speed-to-lead) | 21× more likely to qualify a lead (5 min vs 30 min) | HBR (2011) · MIT/InsideSales study (15,000 leads) |
| Answer every call (AI receptionist) | 27% of home-services calls otherwise unanswered | Invoca platform data |
| Automatic appointment reminders | 38.1% → 23.5% no-show rate (−14.6 pts) | Int. Journal of Pediatrics (2016) · RCT |
| Reminders at scale (across studies) | ~41% average reduction in missed appointments | PAMJ-One Health (2020) · review of 20 studies |
| Always-on natural-voice front desk | 68% of consumers prefer to call a business | Invoca Buyer Experience Report · 500 consumers |
The takeaway: across independent studies, the wins come from the same three habits — answer fast, answer every time, and remind every patient. Those are exactly the jobs an AI agent does perfectly, around the clock. These are category results, not Neuron HQ's own client numbers — but they're the proof the model works before we build it for you.
Want this for your front desk?
Tell us where calls, leads, and appointments slip through today. We'll map exactly which AI agents to build first — and what they'd recover — for your business.
Does editing a page actually change whether AI cites it?
Yes, and this is the only peer-reviewed evidence on this page that the effect is causal rather than a correlation. Researchers from Princeton, Georgia Tech and the Allen Institute for AI built GEO-bench, roughly 10,000 queries drawn from nine datasets across 25 domains, and tested content changes against generative engines. Their finding, verbatim: “GEO can boost visibility by up to 40% in generative engine responses.” They also found tactic effectiveness differs measurably by domain, which is why this work is done per industry rather than from a universal checklist.
Is anyone actually arriving from an AI answer yet?
Adobe measured it across more than a trillion visits to US retail sites and found that in February 2025, traffic from generative AI sources increased by 1,200% compared to July 2024. The behaviour matters more than the growth rate: visitors arriving from AI answers showed 8% higher engagement, 12% more pages per visit, and a 23% lower bounce rate than other traffic. Adobe sells analytics, not AI-search tooling, which is what makes this the answer to “the people quoting these numbers are the people selling the service”.
What does an AI agent actually take off a team’s plate?
Bosch Service Solutions put classification in front of a customer-interaction centre run for a German car manufacturer. More than 90% of incoming emails are automatically and correctly pre-classified, and their operations manager reports the average handling time per email fell from more than 5 minutes to less than 1, which they equate to five full-time roles redeployed onto answering customers rather than sorting mail. That boundary is the honest one: the AI reads, routes and files; humans still handle the actual request.
Is renting software per seat actually the cheap option?
Zylo manages more than 40 million SaaS licences and $40 billion in SaaS spend, and reports that SaaS spend now averages $4,830 per employee, a 21.9% increase year over year. That is the rent-versus-own arithmetic in one number: per-seat pricing scales with your headcount whether or not the seat gets used, and it compounds against you at renewal. Software you own carries a build cost and a hosting cost, neither of which is indexed to how many people you hire.
Does automated follow-up actually produce leads, or just noise?
Randomised field experiments across three companies, sent to millions of recipients and published in Marketing Science, found that simply adding the recipient’s name to a subject line raised opens by 20% (9.05% to 10.80%), increased sales leads by 31% (0.39% to 0.51%), and reduced unsubscribes by 17% (1.2% to 1.0%). The unsubscribe result is the important half: a mechanical change carrying no information about the product moved lead volume up and opt-outs down at the same time, which is the evidence-based answer to “automated follow-up is just more spam”.
Different industries. Same way of building.
Learning, spa operations, construction finance, the job hunt — on the surface these have nothing in common. Underneath, they were built the same way: an idea scoped into a real app, the AI work done on our Neuron engine, and every line of code reviewed by a senior engineer before it shipped. That's the method behind Built with Neuron, and it's available for your idea too.
Your idea, shipped
as real software.
These four started exactly where you are — as a description of something that didn't exist yet. Tell us what to build. Fixed price, every line reviewed by a senior engineer, first version in about two weeks, and you own the code.