Why late entry costs more than it looks
An argument, not a forecast: why arriving late to a retrieval channel is structurally more expensive than arriving late to a ranked one.
The argument for moving now is usually made emotionally, a growth chart, an adjective, a deadline someone invented. We want to make it structurally instead, because the structure holds regardless of how fast adoption turns out to be.
What you would need to know
A defensible business case needs three numbers: the share of your category's demand that resolves through an assistant, your share-of-answer once instrumented, and the rate at which both move.
We are not going to supply those from a benchmark, because a benchmark drawn from other people's categories would be a guess wearing a number. The first two are measurable for your category specifically, which is what an assessment does. The third nobody knows.
What follows does not depend on any of them.
Why late entry costs more than it looks
In a ranked-list channel, arriving late means paying more for the same position. In a retrieval channel, arriving late means something harsher: the answer has already stabilised around whoever was structurally available when the question was first asked well. Displacing an incumbent answer costs materially more than becoming one.
Layer on top of that the data asymmetry. A merchant instrumented twelve months earlier has twelve months of answer-monitoring history, knows which claims move recommendations in their category, and has already fixed the failure modes you have not discovered yet.
The gap is not twelve months of revenue. It is twelve months of revenue plus twelve months of learning, and only one of those can be bought back.
The honest counter-argument
Adoption curves are uncertain and some categories will move much slower than others. A merchant in a category where purchases are overwhelmingly tactile, local or impulse-driven has a legitimate case for waiting.
Which is exactly why the first engagement should be a measurement, not a build. Find out what your category actually looks like, and then make the call with evidence instead of vibes, including, quite possibly, the decision to wait.