Improve / Feedback AI
When you notice that the AI is not yet delivering the right answers or seems to be missing knowledge, the issue is usually not the AI itself.
In most cases, it comes down to configuration, prompting, instructions, or missing/incorrect source data.
In this article, we’ll walk you through the core process of improving and properly training the AI in ArminCX, step by step.
1. Give Direct Feedback in the Inbox
One of the fastest ways to improve AI quality is by using direct feedback on AI responses.
Within the inbox, you can:
Like or dislike AI-generated replies
Provide an alternative response when the AI answer is not ideal (by clicking on dislike)
This feedback helps the AI understand which answers are helpful and which are not. Over time, this creates a stronger foundation and improves how the AI responds to similar questions in the future.
Best practice:
Regularly review AI replies, especially in the early phase, and actively provide feedback.
2. Maintain High-Quality Sources
AI can only answer questions based on the knowledge it has access to.
If certain topics are missing or answers are inaccurate, the first thing to check is the Sources.
Important points to consider:
If the AI has no information about a topic, it cannot answer correctly
Outdated or incorrect content can lead to wrong or misleading responses
Too much irrelevant data can reduce answer quality
Using Files (e.g. PDFs, documentation) or Q&A entries is highly recommended to:
Provide structured, reliable knowledge
Keep information up to date
Cover common customer questions in detail
Best practice:
Regularly audit your sources and remove outdated content.
3. Use Clear and Strong Instructions
Instructions are one of the most powerful tools to shape AI behavior.
They define:
How the AI should communicate (tone, style, formality)
How specific topics should be handled
What the AI is allowed or not allowed to do
You can think of instructions as onboarding a new employee:
You tell the AI how to behave, what rules to follow, and how to respond in different situations.
Examples include:
How to handle complaints or sensitive topics
How to talk to clients
How detailed or concise answers should be
Best practice:
Write instructions clearly, explicitly, and from the perspective of a support agent.
Final Thoughts
Improving AI performance in ArminCX is an ongoing process, not a one-time setup.
By combining:
Continuous inbox feedback
Well-maintained knowledge sources
Clear, structured instructions
you enable the AI to deliver more accurate, helpful, and brand-consistent responses over time.
If you treat the AI like a real team member and continuously guide it, the quality will steadily improve.