Skip to content
EN
Start a project

WorkAI ticket routing

Client project under NDA · AI

A model sorts the tickets now.

AI ticket routing is a language model we adapted for a client. It reads every new ticket, classifies it and passes it to the department that can resolve it.

Start a project
Illustration: from the left, blank cards drift in disorder toward a violet point in the middle. To its right, the stream splits into four lanes. On each, the cards lie in a row and end in their own tray. That’s how every ticket lands in the right department.

Starting point

Someone had to read every ticket, classify it and pass it to the right department. That took hours every week.

Tickets came in without a category and with unclear urgency.

What we built

  1. First, clean up the old tickets.

    The basis was tickets from many years. Before a model could learn from them, they had to be clean: duplicate and empty tickets went out, along with those that were settled on the phone in the end. Names and contact details were anonymized, and access rights were split into levels.

  2. Categories the departments already know.

    For labels, we took what the departments had been using themselves. The analysis also turned up categories that didn’t exist yet. Urgency follows the levels that are standard in the field.

  3. A model that only decides.

    We adapted a language model so that it only decides and writes no text: which category, which department, how urgent. We measured whether it gets it right in several rounds, each with its own yardstick. After each round we readjusted the model.

How a ticket lands in the right department. The model writes no text, it returns only three values.

Result

New tickets now go straight to the right department. Sorting by hand is gone, which saves several hours every week.

We tested on old and new tickets.

Data protection

The model runs on the client’s own premises. New tickets never leave the company.

Personal data was removed before training. In operation it needs hardly any computing power.

In a similar spot?

We build models for your own data under AI, automation & modern development. Zeasy RAG shows how we make knowledge available to look up instead.

Start a project