Problem
XP, a Brazilian investment platform, opens about a million accounts a year. A similar number of users signed up but abandoned registration midway, or opened an account and never funded it. A year later they no longer remember starting, so emails and WhatsApp journeys rarely land. The journeys are built by hand in Adobe's marketing cloud: 20-60 hard-coded onboarding journeys, and every new offer or segment is an engineering ticket.
Example from the article: over 1M users signed up but never finished registration or never funded an account.
Solution in brief
Marketers define a goal (for example, account open), an audience and a pool of creatives. A reinforcement-learning-based decisioning agent reads the lakehouse data and picks which message to send to each user, in what order and when. XP compared it with its heavily A/B-tested average journey and reported +72% account opens and +35% activation. A real-time data foundation had to come first.
Questions this case answers
- How to re-engage users who stopped mid-funnel by choosing message, order and timing per user from a pool of creatives instead of one fixed average journey
- How to optimize a channel for a conversion that happens months after the first touch
- How to re-activate users who signed up but never finished registration or never funded an account
- How to replace dozens of hand-built onboarding journeys with 1:1 decisioning over a pool of creatives
Tools and methods
- Hightouch AI Decisioning — picks message ID, order and timing per user.
- Databricks lakehouse — one store for online and offline first-party data.
- Hightouch events and connectors — web events and ad data into the lakehouse in real time.
- Synthetic conversions — predicted investment value sent to Meta and Google as the signal.
- Conversion API — server-side signal sync to Meta and Google Ads.
Google or a one-time email code. No password.