Booking assistance service

Customers' most common questions fell into two groups: the status of a booking or request, and changing a booking that was already confirmed. Many of these could be answered from company policy or from the customer's own account. But customers couldn't easily find those answers, so they raised a ticket and waited.

How long they waited could depend on their language. Support for some languages had limited hours and was only available by phone or email. A customer writing in one of those languages outside working hours had no one to ask until support opened again.

Bring the answer to the customer, inside the account area where they already are, and let people take over only when it matters.

Principles

• Answer from what's real. Every answer comes from the company's own policies or from the customer's own account: their pending, active and past bookings and requests.
• Guide, don't act. The assistant explains and points the way. It never makes or cancels a booking; when a booking needs to change, it sends the team a change request, so every change is confirmed by a person.
• Know the limits. When something needs a person, the assistant says so plainly and hands over, rather than guessing.
• Every language, any hour. All six languages, around the clock, closing the gaps in support hours.

The assistant lives in the account area. These four conversations, with invented customers and bookings, show how it behaves.

Support tickets per active customer fell by 78% over roughly three months after launch, compared with before. Measuring per active customer means the drop reflects the assistant, not a change in the number of customers.

Customers rated the assistant's answers in the chat, and complaints fell. And for the first time, customers could get help in every one of the six languages at any hour, not only when the support team for their language was working.

Designing the handover taught me that an assistant's limits are part of the experience, not a failure of it. A handover works when the assistant is honest about why it's handing over, and the customer never has to explain everything again.

If I did it again, I'd track which questions most often end in a handover from day one. Those are the clearest signals of where the help content is thin, and where the next improvements should go.