Automating dental appointment reminders without losing human control

01

The business problem

An independent dental practice sends reminder messages to patients before their appointments. Every week, the office manager spends around an hour preparing and sending those reminders.

An hour does not sound like much. But it is an hour that has to be found every week. The task still follows the office manager when she is on holiday, and if it is forgotten, patients do not receive their reminders.

The practice was not looking for an abstract AI strategy. It wanted one less recurring administrative task.

02

What we observed

Before building anything, we examined how the process actually worked. The reminders were not simply copies of one template. Each message depended on:

• the patient’s appointment date and time;
• the planned treatment;
• whether the treatment was publicly funded or private;
• whether there was an amount to pay; and
• how to explain the procedure in language that a patient would naturally understand.

None of these decisions was individually complicated. Together, however, they introduced enough variation that a member of staff still had to consider every message.

This is common in administrative work: a process may look repetitive while containing small contextual decisions that make a rigid template inadequate.

03

What we built

The AI-assisted workflow:

• reads upcoming appointments from the practice-management system;
• understands the scheduled procedure;
• translates technical treatment descriptions into natural Bulgarian;
• creates a friendly appointment reminder;
• includes payment information only when relevant;
• prepares the message for review; and
• sends it through Viber after approval.

For example, the workflow avoids displaying an awkward payment value of zero. When there is nothing to pay, it simply omits the payment sentence. It can also turn formal treatment descriptions into language that is clearer for patients.

04

What surprised us

The AI was not the most difficult part. The surrounding software was.

Our first attempt to enter Bulgarian text into Viber by simulating keyboard input corrupted the Cyrillic characters and produced unusable messages. The reliable solution was to change the workflow.

The system now generates and verifies the Bulgarian reminder, copies it safely to the clipboard, opens the correct Viber conversation, and pastes the message. A person approves the initial message before the remaining reminders can be processed.

The practice-management system introduced its own quirks as well. Some information was easier and more reliable to access through direct record links than through modal windows in the user interface.

These details matter. Real automation projects rarely succeed by simply adding an AI model to an existing process. They require an understanding of the systems, exceptions, and workarounds through which the work is actually performed.

05

What others can learn

Businesses do not necessarily need “AI projects”. They need operational friction removed. Sometimes AI is the right tool. Sometimes ordinary automation is more appropriate. Frequently, a dependable solution combines both.

The important work is:

1. understanding the real process;
2. identifying which decisions genuinely require judgement;
3. automating the predictable steps;
4. retaining human approval where it adds control; and
5. handling the edge cases that appear in real systems.

Technology is part of the solution. Understanding how people actually work is the more important part.