How IT company cut 30 hours of routine work with HelpDesk Assistant

Need for optimising customer support

As their client base grew, the company needed to improve how quickly and effectively they handled support tickets. Helpdesk operators were frequently overwhelmed, managing a mix of simple queries and complex technical issues—often without clear visibility into customer context or previous interactions. 

The support process was functional but slow, leading to internal inefficiencies and delays in client responses.

The goal was to get rid of routine tickets effectively.

Challenges for setting up the HelpDesk AI Assistant

Imagine you were a new employee in a company and you were given 500 PDFs and 10,000 old tickets to give you the knowledge to start working on customer support. This was the starting point of the HelpDesk AI Assistant.

Starting point:

• No integrated knowledge base—just 500 scattered PDF files, and even those didn’t cover everything. This made it difficult for the HelpDesk Assistant to deliver accurate support.
• A lot of the knowledge unwritten, missing experience, and knowledge of the operators in a written form.

And with this incomplete setup, the HelpDesk Assistant could not generate the right answers and happened to hallucinate a lot.

First pilot of HelpDesk Assistant

First, the like/dislike ratio of HD Assistant HelpDesk was 21/79. It was due to the missing context and all the above-mentioned challenges. 

If a customer sent a print screen, the current HelpDesk Assistant could not “read” it, so there was obviously a lack of information and context.

Second pilot of HelpDesk Assistant

The second pilot aimed to improve the HelpDesk Assistant by indexing images from PDF files for better context and using a feedback loop for continuous learning.

About 20 customer support operators were testing another HelpDesk Assistant — the “Smart” version for one month after being trained to evaluate the process in cooperation with the HelpDesk Assistant using “like” and “dislike”. After the training and clarifying the process, we asked the operators to add some comments and context to resolved tickets to enlarge their knowledge base.

The approach also changed. To reduce the risk of hallucinations, the HelpDesk Assistant was told not to answer unless it was sure of the right answer. Instead, it now asks follow-up questions to gather more context. The updated version, called “Smart,” also runs on ChatGPT 5—an upgrade from the older model used before.

After the second pilot, the like ratio has obviously improved and so did the operators' satisfaction with the HelpDesk Assistant.

The solution

The company implemented HelpDesk Assistant, part of the Easy Redmine suite. Integrated directly into their support workflow, the AI-powered assistant helps operators instantly access relevant data, suggest responses, and prioritise tickets based on urgency and complexity.

After changing the AI model from the “Fast” one to the “Smart” one, HelpDesk AI responses have improved significantly. The Smart HelpDesk Assistant enhancements were as follows:

• Contextual replies: The assistant could now consider the full conversation history, not just the latest message.
• Image recognition: AI gained the ability to read images attached to tickets and generate responses based on their content.
• Consistency and speed: Enhanced algorithms reduced response times of operators and improved consistency across agents.

The HelpDesk Assistant for drafting helpdesk ticket responses based on internal resources and helpdesk history. It could analyse ticket content and generate draft replies using company knowledge and historical ticket data. This marked the beginning of AI-driven customer support on the platform.

Key outcomes

• Faster ticket resolution: AI suggested answers and automated insights cut down average response times.
• Less operator fatigue: Repetitive queries are now handled with fewer clicks, allowing human agents to focus on edge cases.
• Higher client satisfaction: Faster turnaround on tickets meant fewer escalations and better feedback scores.
• Operational clarity: Managers now have clearer visibility into ticket types, response patterns, and resource allocation.
• Storage of company know-how: Collecting through resolved tickets, often saved in a new knowledge article.

If the operators add know-how to the ticket when using the HD assistant, the know-how remains in the company even after the HD operator leaves.

Estimated HelpDesk Assistant ROI

We created an estimation of ROI based on the agreed metrics with the client. In the HelpDesk AI model example, 300 AI-generated answers per month receive a 60% like ratio, meaning 180 of them are successful.

Each liked answer saves an average of 10 minutes, totalling 1,800 minutes or 30 hours of work time saved monthly. With an internal hourly rate of 20 EUR , this time saving equals 600 EUR in value.

After subtracting the 120 EUR monthly AI cost, the HelpDesk team achieves a net ROI of 480 EUR per month, effectively saving 30 hours of manual effort and boosting productivity.

For teams handling 1,000+ support tickets per month, the ROI potential scales dramatically—turning marginal gains into a significant operational advantage with significant return on investment.

Conclusion: Boosting productivity with HelpDesk AI

The improvement of the AI HelpDesk Assistant within Easy Redmine reflects a strategic shift towards AI-driven service management. Starting from basic AI-assisted ticket replies, the platform has evolved to offer advanced, context-aware, and multi-modal (text and image) support tools. HelpDesk Assistant now handles routine ticket drafting, leverages historical data, and provides more accurate responses.

These enhancements have not only streamlined internal workflows but also elevated the quality and speed of customer support. Faster, more consistent, and contextually relevant replies have improved customer satisfaction and reduced resolution times.