How AI Automates Our X-Ray Upload Workflow

Taking an X-ray is fast. Filing it is not. In our dental practice, we take around 30 X-rays a day — periapicals, bitewings, OPGs, the lot. The clinical part takes seconds. But the admin that follows? That's where the time disappears.

Every X-ray needs to end up in the correct patient's record in our practice management system. That sounds simple until you watch someone do it 30 times in a row, scattered across a full day of appointments. Export the image from the imaging software. Save it to the desktop. Open the practice management system. Search for the patient. Navigate to their imaging tab. Upload the file. Confirm. Delete the copy from the desktop. Move on to the next one.

Two minutes per X-ray. Thirty X-rays a day. That's a full hour of dental assistant time, every single day, spent on what amounts to digital filing.

An hour a day doesn't sound like much — until you add it up

Five hours a week. Twenty hours a month. 260 hours a year. That's more than six full working weeks of a dental assistant's time, spent dragging files from one place to another. Not assisting chairside. Not prepping trays. Not greeting patients. Just clicking through the same upload workflow, over and over.

At a loaded cost of around $40–45 per hour for a dental assistant in Darwin, that's roughly $10,000–$12,000 per year in staff time. For filing images.

We knew there had to be a better way. So we built one.

The solution: a watched folder that does the thinking

The concept is surprisingly straightforward. Instead of saving X-rays to the desktop and manually uploading them, our imaging software now saves directly to a shared network folder. That folder is watched by our AI system. The moment a new file appears, the automation kicks in.

Here's what happens, step by step:

1. X-ray saved to the watched folder — the clinician takes the image and it lands in the network folder automatically. This is the only step that involves a human.
2. AI reads the filename and metadata — imaging software typically embeds patient identifiers in the filename or the file's metadata. Our system reads both.
3. AI matches to the patient — using the filename pattern and the day's appointment schedule, the system identifies which patient the X-ray belongs to. If the filename contains a patient ID, it's a direct match. If it's less explicit, the system cross-references the appointment book — who was in the chair at the time the image was captured?
4. Automatic upload to the patient record — the X-ray is uploaded to the correct patient's imaging folder in our practice management system. Properly named, properly dated, properly filed.
5. Action logged — every upload is logged with a timestamp, the patient it was matched to, the filename, and the method used for matching. Full audit trail.

From the clinician's perspective, nothing changes. They take the X-ray exactly as they always have. The image just ends up where it needs to be without anyone lifting a finger to put it there.

Before and after

The numbers tell the story:

• Time per X-ray filing: dropped from ~2 minutes to zero (fully automated)
• Daily time on X-ray uploads: dropped from 60 minutes to 0 minutes
• Weekly time saved: 5 hours of dental assistant time returned to clinical work
• Annual time saved: 260 hours — over six weeks of full-time equivalent work
• Misfiled X-rays: effectively eliminated. The AI matches consistently. Humans, especially when juggling three patients and a ringing phone, occasionally file an X-ray under the wrong name. The AI doesn't get flustered.

We've talked before about how the real cost of software tools isn't the subscription — it's the staff time around them. X-ray filing is a textbook example. There was no software subscription to cancel here. The cost was pure labour, hidden inside the daily routine, never questioned because "that's just how it works."

How it handles the tricky cases

Any automation is only as good as its ability to handle things going wrong. We designed the system with three specific edge cases in mind:

• Unrecognised files — if an image lands in the folder and the system can't confidently match it to a patient, it doesn't guess. It moves the file to a "manual review" queue and sends a notification. A team member can match it in a few seconds with full context. Silent mismatches are far worse than a flagged file that needs a quick human decision.
• Duplicate detection — if the same image is saved twice (it happens — someone re-exports, or the software hiccups), the system detects the duplicate and skips it rather than uploading the same X-ray twice to a patient's record.
• Automatic backups — before any file is moved or processed, a backup copy is created automatically. If anything ever goes wrong, the original image is always recoverable.

What it can't do

We're always upfront about limitations because overpromising helps no one.

If an X-ray file has no patient identifier in the filename and no one is currently scheduled in the chair, the system has nothing to match against. It can't read the X-ray itself and identify the patient from their teeth (that's a different kind of AI problem entirely). In those cases, it flags the image for manual matching. This happens rarely — maybe once or twice a week — and it takes about 15 seconds to resolve when it does.

The system also can't handle images that arrive in formats the practice management system doesn't accept. If someone saves an X-ray in an unusual format, it flags it rather than trying to force an upload that would fail. Practically, this almost never happens because our imaging software outputs standard formats, but the safety net is there.

This pattern works far beyond dental

The underlying problem we solved is universal: capture something, identify who or what it belongs to, file it in the right place. We built it for dental X-rays, but the exact same architecture applies to any business with a "scan, identify, file" workflow.

• Photography studios — hundreds of images from a session need to be sorted into client folders. A watched folder that reads metadata and matches to the booking schedule could handle it automatically.
• Construction sites — site photos taken throughout the day need to be filed against the correct project, stage, and trade. Currently someone does that manually at the end of each day.
• Medical imaging centres — the same problem we have, but at scale. Every scan needs to reach the right patient record and the right referring practitioner.
• Document-heavy industries — law firms, accounting practices, real estate agencies. We covered this pattern in detail in our post on how our AI document scanner replaced Adobe Scan. The principle is the same: stop humans from being the routing mechanism between capture and filing.
• Veterinary clinics — same imaging workflow as dental, same filing burden. Different patients, identical problem.

If your team captures images or documents throughout the day and someone spends time afterwards sorting them into the right folders or records, that process is a strong candidate for automation.

Why this matters more than it looks

An hour a day of saved time is valuable on its own. But the real benefit is subtler than the numbers suggest.

When a dental assistant knows they don't have to stop between patients to upload X-rays, their workflow changes. They're not mentally tracking a backlog of unfiled images. They're not rushing through uploads between appointments and occasionally filing a periapical under the wrong Smith. They're not staying 15 minutes late because the afternoon got busy and they still have eight X-rays sitting on the desktop.

The cognitive load disappears. The task simply doesn't exist anymore. That's the difference between saving time and eliminating work — and elimination is always better than optimisation.

The bottom line

We went from a dental assistant spending an hour a day on X-ray uploads to a system that handles it in the background with zero human involvement. No more misfiled images. No more backlog at the end of a busy afternoon. No more 30-step daily routine of export, save, search, upload, confirm, repeat.

260 hours a year, given back to clinical work. Not by working harder or hiring more staff, but by letting the AI handle the filing that never should have been a human job in the first place.

If your business has a workflow that looks like "capture something, figure out where it goes, put it there" — whether that's X-rays, site photos, scanned documents, or anything else — get in touch. We'll walk through your current process, identify where the time is going, and show you what it looks like when the AI handles the filing. No obligation, just a practical conversation about getting your team back to the work that actually matters.