How to Build a Copilot for Accurate AI Driven Diagnoses

Whether it’s Housing Associations, Local Authorities, or Repairs Contractors, there are customer service teams that often play a pivotal role in initiating property repairs. Their tasks frequently include raising repair cases and work orders, which require a specific classification code known as a Schedule of Rates (SOR) or Repairs Ordering Schedule (ROS). Traditionally staff have used tools with pictures to help them diagnose the issue and to retrieve the correct code or spreadsheets they look up codes on.

AI created vision of a current style user interface for reporting common problems

However, a recurring challenge emerges: misdiagnosing the repair needs during the initial contact. This often leads to dispatching the wrong operative, which in turn frustrates both the customer and the contractor. Operatives may show up ill-prepared, lacking the necessary tools, materials, or expertise for the task at hand.

Example of the traditional flows to identify the repair service needed.

With this challenge in mind, I started to consider the potential of integrating a Copilot — a digital assistant designed to aid agents in accurately diagnosing repair issues and matching them with the correct SOR/ROS code.

The concept

The core idea behind the Copilot is to streamline and improve the process of diagnosing repairs through intelligent automation. Users can simply copy and paste a repair description or provide the details directly to the bot.

The Copilot then analyses the information, using advanced language models to understand the issue. It cross-references this description with a database of common problems and their corresponding SOR/ROS codes. In just a few seconds, the bot responds with the appropriate repair code, ensuring the right contractor or operative is sent with the necessary tools and materials for the job. This helps to reduce errors in diagnosing issues and speeds up the overall repairs process.

PoC with ChatGPT

Creating a proof of concept (PoC) for a Copilot using ChatGPT is a great way to test its effectiveness before developing a full-scale solution. To start, you can use the pre-trained models available through OpenAI’s API to prototype the core functionality of your Copilot. With basic access to OpenAI’s GPT models, you can begin by feeding repair descriptions into the model and instructing it to respond with the appropriate SOR/ROS codes. This will allow you to quickly validate how well the AI can interpret the repair issues and generate useful recommendations.

By starting with a PoC, you can evaluate how well the custom model fits your requirements before scaling up to a more comprehensive solution.

Build your own with Microsoft Copilot Studio

With Power Automate, the Copilot’s capabilities would expand to include automating the entire work order process.

If you’re curious about the process of building your own Copilot to enhance customer service and streamline operations, I’ve detailed the steps in a previous post. In “Reducing Repair Misdiagnoses with Microsoft Copilot,” I walk through the key considerations, from setting up the digital assistant to integrating it with existing systems for real-time support. You can read the full guide here to gain insights into creating a Copilot tailored to your organisation’s needs.

Users can access the Copilot directly from within Microsoft Teams, without needing to switch between different applications. This convenience ensures that they can quickly respond to repair issues as they arise.

📚 References

For more information on creating and optimising your Copilot, refer to the following resources:

• Microsoft Copilot Studio Documentation
• Introducing GPTs | OpenAI

Feel free to reach out in the comments if you have any questions or need further assistance with your Copilot project. Happy building!