FuseData Products
Braze recipe

Channel Optimisation

Reach each customer on the channel they actually respond to.

← All models
Braze AI

What it does

Channel Optimisation predicts the best channel for each customer, whether that is email, push, SMS or in-app, instead of blasting every channel and hoping. Braze reads how each person has engaged across channels and routes the message to the one most likely to land. It answers the 'which channel' for every message, cuts wasted sends and opt-outs, and improves as cross-channel history builds.

What feeds it

  • ·Cross-channel engagement events, so more than one channel is live
  • ·Reachability and subscription state per channel
  • ·Purchase or conversion events to tell engagement from real response

How to switch it on in Braze

  1. 1Set up a Canvas and add the channel steps you want Braze to choose between.
  2. 2Use an intelligent channel step so Braze selects per customer rather than sending on all channels.
  3. 3Define the fallback order for customers who are not reachable on their preferred channel.
  4. 4Set the conversion event so channel choice is judged on outcomes.
  5. 5Launch, and monitor which channels Braze is favouring for which segments.

How Fuse builds for and utilises this

  • ·We instrument engagement across every channel, so the model sees the full cross-channel picture rather than one loud channel.
  • ·We define eligibility: reachability, subscription state and fallback order, so no customer falls through the cracks.
  • ·We set the conversion event so channel choice is judged on outcomes, not opens.
  • ·We review the channel mix regularly, checking which channels Braze favours for which segments and whether that matches the economics.

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 per message sent, not per channel, so fewer better-placed sends can win
  • ·Opt-out and unsubscribe rates, which should fall as blasting stops
  • ·Incremental lift versus a send-everything control on /prove
Run the holdout simulator →

Pairs with

On its own it is useful. Alongside these models it compounds.

Model
Send-Time Optimisation
Model
Contact-Frequency Optimisation
Model
Next-Best-Action