Customer service copilot inside the existing workflow

Context

The customer service team of a logistics provider handled thousands of tickets each day. These tickets were about shipment status, estimated arrival times, new claims, and exceptions. Earlier chatbots failed because they were outside the tools that agents used. Agents did not move to a separate AI tool. The break in the workflow removed any productivity increase that was possible on paper.

Approach

The project team designed the AI as an assistant inside the ticket system that agents already used. It was not a separate application. The system showed suggested responses, draft text, and shipment data with each ticket. The project team trained the model on the actual response patterns of the service team. As a result, the suggestions had the same tone and approach that the service team used.

The project team released the tool in stages. For two weeks, the tool operated in shadow mode. A small group of agents saw the suggestions, but customers did not get them. Quality assurance (QA) grades of samples then confirmed the quality.

After this, the wider team got optional access. The QA results stayed stable for two more weeks. Only then did the suggestions become the default first draft of each response. Agent feedback had an effect on each phase.

Outcome

• Agents now handle approximately 60% more tickets in each shift.
• QA grades for response quality stayed stable or improved slightly during the rollout.
• New employees learn the workflow with the AI as the default. Adoption is no longer a separate change-management problem.

Principle

Adoption. The technical model was good in lab tests. The most important factor was that the model was inside the workflow that agents already used. Adoption had as much effect on the design as accuracy. The lesson: people usually do not accept an AI feature that makes them change tools.

We show this case study with permission from the client, as an example only. Specific commercial details stay confidential. If you are a qualified counterparty, you can ask for these details under a non-disclosure agreement (NDA).

← Back to Orchestron. · More case studies · Discuss an opportunity