Plan requirement
| Subscription | Any plan |
| Access | Admin |
The difference between an AI agent people tolerate and one they use is design, and most of it is about brevity and exits.
Answer in the first reply
Not a greeting, not a menu, not a question about which category this falls into. The customer has already said what they want; the best reply is the answer.
Every exchange before the answer loses people, and the ones you lose are disproportionately the ones who were nearly helped.
Ask only what you need
One question at a time, and only where the answer changes what happens next. A question whose answer you do not act on is a question that should not be asked.
Keep replies short
Two or three sentences on messaging. Long replies get skimmed, and a skimmed instruction is a failed instruction.
Where the answer is genuinely long, give the short version and link the article.
Always leave the door open
A route to a person, available at every point, and not hidden behind a particular phrasing. Customers who cannot find the exit do not conclude the AI is thorough; they conclude you are avoiding them.
Fail in one step
When it cannot help, it should say so and hand over. Three attempts at rephrasing is worse than one clear "I cannot answer this, let me get somebody".
Write the way people speak
Short sentences, ordinary words, no internal terminology. If your team calls something a "provisioning request" and customers call it "setting up my account", write the second.
Test the frustrating path
Design for the customer whose question is not covered and who is already annoyed. That path is the one that produces complaints, and it is the one nobody rehearses.
Comments
0 comments
Article is closed for comments.