Solved, not deflected: what AI support should measure
Most AI support tools brag about deflection — the percentage of tickets a bot closes before a human sees them. It sounds efficient. In practice, it often means customers bounce off an unhelpful chatbot and quietly churn.
A different north star
LightReply doesn’t try to keep customers away from you. Every reply is drafted by AI and sent by a person, so the question we care about is simple: did the customer’s problem get solved?
We run LightReply on our own support at Brash Apps. Most tickets now start from a draft that’s ready to send, and the person sending it still decides what the customer hears.
Why a person stays on Send
A draft that reads well but promises a refund nobody issued is worse than no draft at all. Keeping a person on Send means every answer is checked against what actually happened. The AI does the typing; you make the call.
What we look at instead
Did the reply answer every question in the email? Did the customer have to write back? Are they still a customer next month? These are harder to count than deflection, and much closer to what keeps a small business alive.
Deflection optimizes for the company. Solving the problem optimizes for the customer. We’ll take the second one.
Keep reading
How LightReply learns to sound like you
Your sent replies are the best training data there is. Here's how we turn them into drafts you'd actually write.
A support workflow that fits a team of one
You don't need a 12-person support org. You need to clear the inbox before lunch. Here's a workflow that works.