One Year of AI Adaptive: Every Tool We Built and What It Replaced

One year ago, we started building custom AI tools for our dental practice. Not buying them. Not subscribing to them. Building them — from scratch, on our own hardware, to solve our specific problems.

Twelve months later, we have 16 tools in production, a monthly software bill that's a fraction of what it used to be, and a team that genuinely cannot imagine going back. This is the full inventory — every tool, what it replaced, what it cost to build, what it saves, and how the team actually feels about using it. No cherry-picking. No rounding up.

The complete tool inventory

A few notes before the table. Build costs include development time, testing, and iteration — not just the initial prototype. Monthly savings combine eliminated subscriptions and estimated staff time recovered, valued at realistic Australian wage rates. Staff adoption is scored 1–5 based on how quickly and willingly the team integrated each tool into their daily workflow.

That's the full list. Sixteen tools, all running in production, all built over twelve months. Some are simple scripts. Some are full applications with installers and user interfaces. All of them solve a real problem we were actually having.

The maths

We've broken this down in detail in our real cost of AI tools post, but here's the summary at the one-year mark:

Is $54,000 a precise number? No. The subscription savings are easy to measure — you either cancelled the subscription or you didn't. The staff time savings are estimates based on how long tasks used to take versus how long they take now. We're confident the real number is in that ballpark, but we're not going to pretend it's exact to the dollar.

The payback period of roughly 14 months means we're not quite at break-even on the full suite yet. But individual tools paid for themselves much faster. The AI phone receptionist covered its build cost in under a year. Voice dictation paid for itself in about five months once we stopped paying Dragon Medical licences. The smaller tools — daily briefing, branded voice notes, call recording — paid back within weeks.

The most valuable tool

The AI phone receptionist, without question. It's the tool with the most immediate, obvious, daily impact. Every single call gets answered. Every after-hours enquiry gets handled. Patients book appointments at 10pm on a Sunday without thinking twice about it. Staff walk in each morning to a schedule that's already been updated overnight. It just works, quietly and reliably, and it changed how our practice operates more than any other single tool.

If we could only keep one tool from the entire stack, this is the one.

The most surprising tool

The AI memory system. We built it as an experiment — a way to capture decisions, notes, and context so that nothing got lost between projects. We didn't expect it to become essential. But it did. Being able to ask "what did we decide about the referral letter format?" or "what was the issue with the X-ray upload last month?" and get an instant, accurate answer changed how we work. It replaced something we didn't even realise was a problem: tribal knowledge that lived in people's heads and nowhere else.

Adoption is still only 3 out of 5, because not everyone on the team uses it directly — it's more of a behind-the-scenes tool. But for the people who do use it, it's indispensable.

The honest bit: what didn't work perfectly

Staff rostering scored the lowest adoption at 3 out of 5. The team preferred the old way for longer than we expected. Deputy has a polished mobile app, familiar interface, and years of muscle memory behind it. Our replacement was technically better and significantly cheaper, but "technically better" doesn't always win against "I already know how to use this." Adoption improved once we spent time training and listening to feedback, but it was a good reminder that convenience beats capability when it comes to daily habits.

Three tools didn't make it at all. We wrote about those in detail — the all-knowing chatbot, the AI appointment reminder calls, and an automated treatment plan generator. Each one taught us something valuable, but none of them survived contact with reality. That's not failure. That's iteration. You don't build 16 successful tools without a few that don't work out.

Some tools took longer to stabilise than we planned. The email autoresponder went through three major revisions before the tone was right. The document scanner needed significant tweaking to handle the variety of paper quality we actually receive. Nothing was plug-and-play on day one. Everything required iteration, and most things required more iteration than we budgeted for.

What we'd tell someone starting today

Start with the pain, not the technology. Don't build an AI tool because AI is exciting. Build one because something in your business is genuinely broken, slow, or expensive. Our best tools solved real problems that staff complained about weekly. Our worst tools solved problems we imagined but nobody actually had.

Pick one tool first. Nail it. Then move on. We were too enthusiastic early on and tried to build several things in parallel. The tools we built sequentially, with full attention, were better and adopted faster than the ones we rushed. We talked about this in our nine-month retrospective, and the advice hasn't changed.

Budget for the human side. Training, documentation, feedback loops, and patience. These aren't optional extras. They're half the project. A tool that nobody uses saves nothing.

Own your infrastructure. Every tool we built runs on hardware we control. No API rate limits, no surprise price increases, no vendor shutdowns. When a SaaS product we relied on changed their pricing last year, it didn't affect us at all — because we'd already replaced it. That peace of mind is worth more than the dollar savings.

One year down

A year ago, our practice ran on a patchwork of subscriptions, manual processes, and workarounds. Today, it runs on 16 tools we built ourselves, talking to each other, automating the tedious work, and freeing our team to do what they're actually good at — looking after patients.

The $54,000 in annual savings is nice. But if we're honest, the real value isn't the money. It's the control. We know exactly how every system works. We can fix anything that breaks. We can adapt to any change in our business without waiting for a vendor to add a feature or submitting a support ticket that goes nowhere. That kind of independence is hard to put a number on, but it's the reason we'd do it all again.

Nineteen tools built. Sixteen survived. Three taught us what not to do. Zero regrets.

If you're thinking about building custom AI for your business — or even just trying to figure out whether it makes sense — we'd love to chat. No pitch, no pressure. Just an honest conversation about what's worked for us and what might work for you.