Intelligent Selection
Let Braze pick the winning variant for you, and keep sending it.
Braze AI
What it does
Instead of running an A/B test, waiting, reading the result and promoting the winner by hand, Intelligent Selection lets Braze do it inside the campaign. You author the variants, set the outcome that matters, and Braze steadily shifts traffic towards whichever variant is performing best. It is a quick win: no extra data model to build, available now, and it turns every multi-variant message into a self-optimising one.
What feeds it
- ·Two or more message variants in a single campaign
- ·A defined conversion or primary event to optimise towards
- ·Enough send volume for Braze to tell the variants apart
How to switch it on in Braze
- 1Open the campaign and add the variants you want Braze to choose between.
- 2Under the variant settings, turn on Intelligent Selection rather than a fixed percentage split.
- 3Set the conversion event Braze should optimise for, and its conversion window.
- 4Launch. Braze serves the variants, learns, and reallocates traffic to the strongest as results arrive.
- 5Review the campaign analytics to see which variant won and why.
How Fuse builds for and utilises this
- ·We structure the conversion events so Braze has a clean outcome to optimise towards, not a vanity click.
- ·We design the experiment paths and variant strategy so the algorithm has genuinely different options to learn from.
- ·We set the guardrails: conversion windows, minimum volumes and when to intervene if a variant misbehaves.
- ·We read the results back through Currents into the warehouse, so what won here informs the next campaign, not just this one.
What to measure
A capability that fires on people who would have converted anyway proves nothing. Hold a slice out, measure only the gap above them, and let the recipe earn its place on evidence.
- ·Conversion rate of the optimised campaign against your previous manual A/B approach
- ·Time saved by not promoting winners by hand
- ·Incremental lift, held honest with a control group on /prove
Pairs with
On its own it is useful. Alongside these models it compounds.