Teams in the construction industry deal with a constant flow of paper delivery tickets. Every day, drivers drop off materials and hand over a slip. Someone on site has to collect these, take a photo, and email it back to the office. From there, an administrator manually types the details from each photo into a spreadsheet. This process is slow, prone to typos, and creates a significant delay between when materials arrive and when they are officially logged.

You can watch the full build here: https://youtu.be/-zd1\_1s1YdQ

The Manual Data Entry Problem

The core issue is turning a physical piece of paper into a structured digital record. When a photo of a delivery ticket comes in, it's just an image file. It isn't searchable, and the data isn't usable until someone manually reads it and enters it into a system like Google Sheets or an accounting platform. This administrative work consumes hours that could be spent on higher-value tasks. It also introduces a risk of human error, where a wrong number can cause issues with inventory and billing down the line.

Building the Automation Logic

To solve this, we didn't look for a complex, off-the-shelf software package. We built a direct and efficient automation using a few connected tools. The goal was to create a system where a site manager could simply post a photo of a delivery ticket into a Telegram group, and the data would appear in a Google Sheet moments later, with no manual typing involved.

We chose this approach because it meets the team where they already work-on their phones. Here is the logic we implemented:

• Data Capture via Telegram: We started by creating a simple Telegram bot. This bot is added to a group chat with the site team. Its only job is to watch for new images being posted. We configured its privacy settings to ensure it could read all messages, not just commands directed at it. You can find details on setting this up in Telegram's official documentation.

• Processing the Image: Once an image is posted, the automation, orchestrated in Make.com, triggers. The first step is to grab the highest resolution version of the photo available. This is a critical detail because the accuracy of data extraction depends heavily on the quality of the source image.

• Extracting the Data: The image is then sent to an AI model via an API call. We provide the model with a clear prompt, telling it exactly what information to find in the photo: the ticket number, the supplier name, the date, and the list of materials. To ensure the output is consistent, we defined a JSON schema. This forces the AI to return the data in a clean, structured format that our system can easily parse.

• Handling Errors and Duplicates: Not every photo is perfect. If the AI cannot read a blurry or poorly lit image, the system doesn't just fail silently. It automatically sends a reply back to the Telegram group, notifying the user that the image was unreadable and asking them to resubmit. We also built in a check to prevent duplicate entries. Every photo in Telegram has a unique File ID. The system checks this ID against our Google Sheet to see if it has already been processed, preventing the same ticket from being logged twice.

The Outcome

The system is now active. The team on site posts photos of delivery tickets directly into a Telegram chat as they receive them. The data is automatically extracted and populated into a central Google Sheet within seconds. The administrative team no longer has to perform manual data entry for these tickets, saving at least an hour of work each day. The data is also more accurate and available in near real-time, giving the project managers a clearer view of material deliveries.