We recently added a Braze Agent Step to one of our re-engagement Canvases targeting driving students who haven’t booked a lesson for around 15 days.
The goal of the Canvas is straightforward: bring inactive students back and get them to book their next driving lesson.
For the push step, we’re now running a 50/50 experiment:
- 50% receive our existing static push
- 50% go through an Agent Step that generates the message based on each student’s actual situation
The Agent can use signals such as hours already driven, prepaid hours remaining, and the personal range of hours recommended by their instructor. Based on those inputs, it selects the most relevant angle — for example whether the student needs to top up, is close to their recommended objective, is still early in their journey, or needs a more reassuring message.
It then generates the push title and body within strict guardrails around factual accuracy, tone, allowed claims and character limits.
So instead of relying only on segment → predefined copy, we’re testing individual customer context → Agent reasoning → message angle → generated copy.
The business KPI remains the same: does the student book another driving lesson?
We launched the experiment this week, so there’s no performance data yet. The objective is to see whether this additional layer of contextual personalization can outperform our standard push and improve reactivation.
