What deflection rate should I expect?

It depends almost entirely on how much of your repetitive volume is already documented. Rather than aim at a benchmark, measure your own normalised ticket volume before and after — that is the number that survives scrutiny.

How long before I can tell?

Give it several weeks. Weekly ticket volume is noisy, and the first fortnight is distorted by the novelty of a new widget on the site.

Why not just use answer rate?

Because a bot that never refuses scores perfectly on it while inventing answers. Answer rate is useful for spotting trends; it is not evidence of deflection.

Grow

Measure deflection in a way you can defend

Almost every published deflection number is overstated, usually by counting conversations that never would have become tickets.

Start freeNo credit card. 2 minute setup.

The easy number counts the wrong thing

Dividing bot conversations by total contacts assumes every conversation would otherwise have been a ticket. Most would not — people ask an assistant things they would never have written an email about, which inflates the figure and makes it indefensible in a review.

Deflection claimed as a share of all bot conversations
No baseline from before the assistant existed
Ticket volume unchanged despite a high reported deflection rate
Satisfaction not tracked alongside the automation numbers

Establish a baseline first

Record ticket volume per week, normalised against whatever drives it — active customers, orders, sessions. Without this, any later claim is unfalsifiable, and the normalisation matters because a growing business would have had more tickets anyway.

Measure the change in tickets, not the count of chats

The honest metric is the change in normalised ticket volume after launch. That captures actual displacement rather than incremental questions the assistant invited by existing.

The Analytics summary cards showing conversations, answered rate, unanswered, negative ratings, handoffs and leads over the selected period. — These tell you what the assistant did. They cannot tell you what it prevented — for that you need your own ticket volume from before it launched.

Normalise by a driver

Tickets per hundred active customers, not raw tickets.

Allow several weeks

One week is noise; a month is a signal.

Watch satisfaction at the same time

Falling tickets with falling satisfaction is not deflection, it is customers giving up. These two numbers only mean something together, and reporting one without the other is how automation projects get reversed a year later.

Attribute the wins to content

When you fill a gap, that question's frequency should fall in the report and in your queue. Tracking that link is what turns a one-off number into an ongoing case for the content work that produced it.

Common questions

What deflection rate should I expect?: It depends almost entirely on how much of your repetitive volume is already documented. Rather than aim at a benchmark, measure your own normalised ticket volume before and after — that is the number that survives scrutiny.
How long before I can tell?: Give it several weeks. Weekly ticket volume is noisy, and the first fortnight is distorted by the novelty of a new widget on the site.
Why not just use answer rate?: Because a bot that never refuses scores perfectly on it while inventing answers. Answer rate is useful for spotting trends; it is not evidence of deflection.