Question handling

Designing the handover so it never turns into a maze

Every other decision in configuring an assistant is downstream of this one. How readily it offers a person determines whether the widget is experienced as help or as an obstacle, and the difference between those two is almost entirely a matter of how the handover is designed rather than how good the answers are.

What they are really asking

They have decided this channel cannot solve their problem, and they want to know the shortest path to somebody who can.

  • can i speak to a human
  • i want to talk to someone
  • agent
  • is anyone actually there
  • this isn't helping
  • give me a phone number
  • can someone call me back
  • i need to talk to a real person
  • how do i contact support
  • let me speak to your manager

The material that answers it

An assistant is only as good as the document behind it, and for this question the document usually exists but is written in the wrong shape. What each one has to contain to be answerable:

Material that answers this question
The contact routes you will actually honour, with hoursA number that is answered, an address that is read, a form somebody works from. Listing a route nobody watches is worse than listing fewer routes, because the person will use it and then wait.
What happens after somebody sends an enquiryWho receives it, when they look, and what the first reply usually is. This is the most reassuring paragraph you can publish and hardly anybody publishes it, because it is obvious from the inside.
A response time you can meet, including out of hoursDifferent numbers for weekdays, evenings and weekends if that is the truth. One honest slow number outperforms an optimistic fast one, because the second is measured against and the first is planned around.
The address enquiries are sent to, and who watches itConfiguration rather than content, and the thing most likely to be set once and forgotten. An enquiry route pointing at a mailbox nobody opens is the quiet failure underneath most complaints about a chat widget.
What you want people to includeAn order reference, the address on the account, and the question in one line. Publishing the list lets the assistant prompt for it, which is the difference between a reply that resolves and a reply that asks for details.

How to handle it

Offer the person before they have to demand one

The instinct is to make the handover a last resort in order to protect the deflection rate. It backfires: somebody who has to ask three times arrives at the human already annoyed, and the conversation starts from a worse place than if they had been offered one immediately.

A good rule is to offer after the second unsuccessful exchange, and immediately for anything in the escalation set of whatever intent they are in. Nobody has ever resented being offered a person too early.

Recognise the request itself rather than leaving it to retrieval

This product handles the request directly. A small set of conversational messages, including greetings, thanks and somebody asking for a person, are recognised without going near the knowledge base, so the offer appears consistently and does not depend on your material happening to contain a page about contacting support.

That consistency is the property that matters. A handover that works because a contact page ranked well is a handover that stops working when the contact page is reworded.

Collect three things, not a form

The form takes a name, an email and a message, and that is deliberately small. A long form at the moment somebody has given up on self service is another obstacle, and the abandonment rate on it is the whole handover leaking away.

The assistant can do the work instead: prompt for the reference or the account address inside the conversation, so the message field arrives with the context in it and the form stays short.

Set a reply expectation you can actually meet

Say when somebody will be in touch, in the same message as the confirmation, and pick a number your team hits on a bad day rather than a good one. A promise of a couple of hours that becomes tomorrow generates a second contact and a worse mood.

The enquiry is emailed to the address you nominate and stored as well, so a failed send does not silently lose it. That matters most here, because this is the message where somebody has already decided the channel is not working.

When it stops being an answer

The request that arrives straight after a refusal

When the assistant has just declined to answer, the offer of a person should be in the same reply. A refusal followed by silence is the exact shape people describe when they say a chat widget was useless, and it is entirely avoidable in the wording of the fallback message.

Out of hours, when nobody is going to reply tonight

Honesty outperforms optimism. Say that the team is back in the morning and that the message is waiting for them. Somebody told plainly will wait. Somebody left to assume will send four more messages and open with that in the morning.

The person who will not use the form

Some people want a number and nothing else, and repeatedly offering a form to them is the maze in its purest form. If you have a number, the assistant should give it. If you do not, it should say so plainly rather than steering, because being steered is the thing they are objecting to.

How this one goes wrong

The maze

Somebody types agent. The assistant offers a help article. They type I want to speak to a human. It suggests searching the knowledge base. They type this a third time, in capitals. Every individual reply was an attempt to be helpful and the aggregate is a company that will not let a customer reach it.

The cost is not one conversation, it is the reputation of the channel: a customer who has been through the maze once will not use the widget again, including for the questions it answers perfectly well. Configure the handover to be offered early and honoured immediately, and the same assistant becomes the fastest route to a person rather than the obstacle in front of one.

The same question, trade by trade

The pattern above holds everywhere. The wording, the escalation line and the material behind it do not, so there is a page per trade.

Questions

Does offering a person early undermine the point of the assistant?
No, because the questions it answers well are ones nobody wanted to escalate anyway. The people asking for a person have a reason, and delaying them converts a solvable problem into a complaint about the widget.
What does the handover actually do?
It offers a short form taking a name, an email and a message. That is emailed to the address you nominate and stored as well, so nothing is lost if the email fails to send. It does not open a ticket in another system.
Can it transfer somebody to a live agent?
No. There is no live chat behind it. It takes a message and gets it to the address you choose, which is why setting an honest reply time matters more here than anywhere else.
Where should the handover be offered most aggressively?
Anything with a deadline, a safety consequence or an unhappy customer attached: deletion requests, complaints, allergens, accessibility adjustments and account recovery. In those intents the handover is the correct outcome rather than the fallback.

Keep reading

Try it on your own material

Upload a document or point it at your site, paste one line of HTML, then ask it something only your business could answer.