Why the Shift Can No Longer Wait
In the age of instant messaging, users increasingly expect rapid responses. What used to be considered an acceptable response time can now already result in a potential loss of image. Customers expect answers that are fast, precise, stylistically consistent, and relevant at the same time.
A Challenge: The “Ticket Flood”
Our customer uses SugarCRM to manage their service processes. Support tickets are handled via the Cases module. When a customer complaint, a technical inquiry, or a cooperation request is submitted via the customer’s website, it is automatically recorded as a case in the system. The challenge lies in the volume and diversity of incoming requests: hundreds of response templates are stored as structured knowledge base articles, and service agents must select the appropriate one for each case.
The Goal: Faster Response Times and Reduced Workload, Technical Flexibility and Platform Independence
As part of the solution design, we defined two key technological principles to ensure future readiness:
- SugarCRM as an Open Foundation for the Customer Service Team
- AI-agnostic approach: Although we fine-tuned Google’s Gemini model during the concept phase due to its outstanding semantic capabilities, it was important for us to design the solution independently of any specific AI engine. Since customer requirements and security policies may vary (whether open source or closed enterprise models), the system is built so that the underlying algorithm can be replaced or extended at any time without affecting the user experience.
Intelligence Built on Real-World Patterns
To train the model, we used examples from real-world use cases (customer inquiries). As a result, the algorithm does not just analyze words but understands industry-specific issues and concerns. Part of the challenge is also determining how much training data is required to achieve a sufficiently high level of accuracy in decision-making.
The Concept: Intelligent Support Within the CRM
The planned process is not a standalone external tool but an organic extension of the existing workflow. In practice, it works as follows:
- Incoming and classification: The customer inquiry appears in SugarCRM as a Case, and basic categorization is performed.
- Software processing: The content of the properly classified Case and its associated customer history is analyzed in the background by our custom-developed code. The appropriately anonymized data is sent to the trained model, which—based on prior training—can determine with high probability which response template best fits the inquiry.
- Intelligent preparation: Our custom-developed code does not only make a suggestion but automatically generates an email draft linked to the Case, based on the selected template.
- Flexibility: The system can be extended with additional intelligent features such as personalized greetings or signatures. The number and content of templates may also change over time, which is why we chose a fine-tuning training approach capable of adapting to the evolving customer knowledge base.
- Personalized approval: The service agent is presented with a pre-formulated response, which they can either adopt as-is or refine and supplement.
The path to AI autonomy
The goal is to operate AI support as independently as possible, without requiring manual control for every decision, and ultimately to enable automated email sending.
Important: Guarding the Company Identity and Style
At the outset, the company was concerned that automation might lead to impersonal responses and undermine brand authenticity. In reality, AI—by generating a near-perfect draft within seconds directly in SugarCRM—gives employees back valuable time for genuine attention: for that extra sentence that turns a ticket number into a truly personal, human response.