Journey Optimizer · decisioning
Which offer wins
Laura bought an hour ago but gets a first-purchase offer. The running offer has used its allowance and the new-customer audience is still out of date. Change the allowance and see which offer takes the slot and why.
Example assumptions
Example catalogue, offers, priorities and scores. Club runs before Promotions; their scores are not compared with each other. The audience is still out of date an hour after a purchase. Delivery and future purchases are not simulated.
Decision policy
Which offer comes out
What the decision policy returns in the chosen channel, and why.
Selection strategies
How each collection is filtered and ranked
Dates, eligibility and capping discard. What is left is ranked with the strategy's method.
Architecture
Where it fits in the platform
Diagram of the Journey Optimizer Decisioning engine. On the left, when the decision is made: in a code-based experience, when the page is requested, and in email, push or SMS, at send time; in both cases with the profile and context of that moment. That request reaches a decision policy, which sets how many items to return and in which order to try its selection strategies. Each selection strategy is a four-step funnel over the catalogue of decision items: the collection decides which items go in, eligibility filters them by audience or by a decision rule, capping removes the ones that have reached their limit, and ranking orders the rest by priority, by formula or with an AI model. If the strategy does not fill the requested items, the policy moves on to the next strategy. On the right, the result: the selected items, whose attributes are used in the message; the fallback, if no strategy returns anything; and the decision event, which records what was proposed.
The recommendation
Rules for what changes, audiences for what does not, sensible capping and always a fallback
The wrong offer is rarely the ranking's fault. It is usually eligibility computed too late, or capping that counts the wrong event.
- 01
Decision rules for what changes
Rules are evaluated at decision time. Audiences are not updated in real time: use them for stable conditions, not for “has already purchased”.
- 02
Capping on the right event
Decision event counts every decision, even when nothing is seen. In inbound channels, impression counts what was actually displayed. Up to 10 cappings per item.
- 03
Ranking that fits the channel
Priority to start with, a formula for business rules and an AI model where there is traffic. In email inside journeys, AI ranking is not available.
- 04
Always a fallback
Default content in every decision policy. If no strategy returns anything, the slot is not left empty.