Let the next step be decided per-user, not by a fixed cadence

Refreshed daily and on every meaningful event, from lifecycle stage, whether sessions are trending up or down, how many features they use, recent intent such as a pricing visit or a support touch, and how they responded to past messages. It returns the action to take, the channel, the content theme, a confidence score, and the uplift it expects.

Active users whose confidence is 50 or above and who are not already in another high priority journey. It refreshes continuously, so people move between branches as their signal changes.

A short value tip for teaching, a customer story for nurture, a personal incentive such as a discount, trial extension or credit for an offer, and a feature spotlight with a deep link. Each pulls content blocks from the whole profile, not just a first name.

A tooltip pointing at the feature with a one line reason and a try it button, and a recommendation card built from what they tend to use. These only appear during an active session.

Fed by their own history, what similar users buy and use, and what is trending in their segment. It shows in the app sidebar, inside emails and in a page slot, and refreshes weekly.

At day 7, 14, 30 and 60 after signup and weekly after that, the journey reads the recommendation and takes that branch: teach, nurture, offer, nudge a feature, surface recommendations, hand to an agent, or wait and recompute. What happens next feeds back into the decision. They leave on conversion, on unsubscribe, or after three waits in a row.

Which actions the model picks and how often, engagement per branch, whether higher confidence really does mean higher conversion, the lift against a 5 to 10% holdout left on a fixed cadence, and which segments it serves worst.