Designing Chat and Concierge Interfaces That Hand Off to Humans Well
- Blog
- Designing Chat and Concierge Interfaces That Hand Off to Humans Well
Insights
Designing Chat and Concierge Interfaces That Hand Off to Humans Well
How to design chat and concierge experiences where the bot greets, helps and then passes people to a human, a booked call or a document upload without losing the thread.
Chat interfaces are back at the centre of digital service, this time with language models behind them. The models are good enough that a bot can now answer a surprising share of questions on its own. That makes the remaining moments, where a person really needs another person, even more important to get right. A concierge experience is judged less by how clever the bot is and more by how gracefully it gets out of the way.
This guide walks through the moments we design for in every chat and concierge project: the greeting, the conversation itself, the handoff to a human, booking a call, and handling attachments. The patterns apply whether the chat sits on a public website, inside a banking app or in an internal service desk.
A Greeting That Sets Expectations
The first message does more work than any other. It should say who or what the person is talking to, what it can help with, and how to reach a human. People are far more forgiving of a bot that is honest about being a bot than one that pretends otherwise and then fails.
Name the assistant plainly and say it is automated, in the first line.
Offer three or four starting points based on the most common reasons people open the chat.
Show a visible way to reach a person from the very first screen, even if most people never use it.
Avoid long welcome messages. One short paragraph and a set of suggestions is usually enough.
Keep the Conversation Legible
Long chat threads get hard to follow quickly, especially on mobile. Break answers into short messages, use structured cards for anything with options or steps, and summarise what has been agreed before moving on. When the bot is pulling information from a policy, an account or a knowledge base, say where it came from. A small source label does a lot for trust.
Design for uncertainty as well. When the bot is not confident, it should say so and offer a next step, rather than producing a fluent but wrong answer. A simple line such as asking whether the person would like to speak to someone is often the most helpful thing it can say.
The best handoff feels like the conversation simply continued with someone who already knows what was said.
Designing the Escalation to a Person
Escalation is where most chat experiences fall apart. The person explains their problem to the bot, gets handed over, and then has to explain it all again. Every handoff should carry the context across, and the interface should show the person that it has.
Trigger escalation on request, on repeated failure, and on sensitive topics such as complaints, bereavement or fraud.
Pass a short summary, the full transcript and any details collected to the human agent before they reply.
Tell the person who they are now talking to, roughly how long they will wait, and what happens if they leave the page.
On the agent side, the summary should appear at the top of the conversation in plain language, with the transcript available underneath. Agents should be able to correct the summary, which also gives the team useful feedback on how the bot is doing.
When the Right Answer Is a Booked Call
Some conversations should not be finished in chat at all. A complex mortgage question, a design brief or a sensitive account issue often goes better on a call. Treat booking as a first class outcome of the conversation, not a failure. Show available times inside the chat, confirm the booking in the thread, and send a calendar invite with the context already attached, so the person on the call can start where the chat left off.
Outside office hours, be honest. Offer the next available slot or a callback, and say when someone will reply. A clear promise kept is better than a vague one broken.
Attachments and Documents
Concierge flows increasingly involve documents: a photo of a receipt, a scanned form, a brief or a contract. Make uploads easy to find, show progress, and confirm what was received with a thumbnail and file name. Where the bot reads the document, show what it extracted and let the person correct it before anything is submitted.
Be clear about where files go and who can see them. In regulated sectors, a short line explaining how documents are stored and used will answer a question many people have but rarely ask.
Accessibility and Tone
Chat windows are often bolted onto pages late in a project, and accessibility suffers. Make sure the chat works with a keyboard and a screen reader, that new messages are announced, and that text can be resized without breaking the layout. Write in the same voice as the rest of the brand, keep sentences short, and avoid forced cheerfulness, especially when someone is clearly frustrated.
Measuring What Matters
Containment rate, the share of conversations the bot finishes alone, is the metric most teams watch. It is useful but incomplete. Track how often people have to repeat themselves after a handoff, how long they wait for a human, and how satisfied they are at the end. A slightly lower containment rate with much better handoffs is usually the better experience.
We design chat, concierge and service experiences for enterprise teams, from the first greeting to the agent console. If you are planning one, talk to us, see our work, or read more about our design services.
Joel Koh
Digital Lead at One X Design, passionate about human-centered design for AI products.
Share this article
Related Posts