OneClinic’s new AI-powered service reduces manual work – and AI usage
OneClinic Oy, specializing in the logistics of healthcare materials and pharmaceutical supply, has developed an AI-based tuning series for its OC Logistics system that speeds up and streamlines the system’s operation – while it also reduces excessive AI usage. Software architect Janne Heikkinen explains how this was achieved.
AI under the hood
“Perhaps it is an exaggeration to say that the new feature would reduce AI usage. But in the development work we considered not to overload the language model with overly long prompts,” Heikkinen describes. “Language models inherently require a lot of computing power. The software developer must understand their responsibility and ensure that a resource-hungry command structure isn’t hidden inside the system.”
OC Logistics’ new AI service operates almost entirely out of sight of the user. It does not open new chat windows or prompt fields for the language model. So what does it do?
“Put simply, the AI service inside the system receives the order confirmation that comes by email from the supplier, analyses its structure, and compares it with the purchase order found in the system,” Heikkinen says. “If it detects discrepancies relative to the purchase order, it highlights them for the user to inspect whether they are real deviations. This happens in under a minute. Without AI, the comparison is currently done manually, which is labour-intensive.”
According to Heikkinen, this is a raw comparison where data contents are examined and analysed quite straightforwardly. “If the deviation is real, a deviation report is generated using the language model, which leads to the subsequent actions,” Heikkinen notes.
Relieving the language model’s load
“Orders can look very different depending on the customer. For example, a PDF order confirmation might contain extraneous information or even unnecessary pages. So, the very first step is for the system to clean up and standardize them. With uniform content, it becomes easier to implement the comparison and prompt for the language model, which can also be standardized as a result,” Heikkinen explains. “At the same time, the data being compared is compressed into as small a form as possible, so the prompt does not carry unnecessary information for the comparison.”
So this is almost literally a tuning series that makes the processing of OC Logistics’ order confirmations substantially faster than before. Where did the development idea originate?
Everyone benefits from the development
“We at OneClinic have a fairly good understanding of how the welfare sector operates and the logistics bottlenecks where a lot of work is still done manually – meaning that with proper system development, it could be easily reduced and professionals’ time freed for more productive tasks,” Heikkinen articulates. “There are even more bottlenecks, but not everything can be fixed at once. For OneClinic’s clients, it’s great that every new system feature becomes available to all other customers as well – provided it has been tested and piloted first. The AI-tuned processing of order confirmations is currently in a pilot phase at a welfare services region, where it is estimated to save several man-years of valuable resources for other tasks. The results are very promising.”