The AI Decision Gap in Braze Campaigns
A lifecycle marketer is building a re-engagement Canvas for a DTC brand, with three email variants, a Content Card fallback, audience filters for recent purchasers, and Intelligent Timing enabled. Braze can determine who enters the flow and when each person should receive a message. The marketer still has to define the rules that decide which content variant each individual sees.
That distinction matters. Most lifecycle teams do not struggle to create more campaign assets. They struggle to decide which asset is most appropriate for a particular customer at a particular moment. A discount may be right for one inactive shopper, while product education is better for another. A reminder may work for someone who browsed yesterday, but feel irrelevant to someone who has only opened messages without clicking for three months.
Braze gives teams the infrastructure to build these experiences. The remaining gap is the decision layer between a customer’s current signals and the message variant delivered to them.
Where Braze stops and the decision problem begins
Braze is very good at campaign orchestration. A marketer can build segments from detailed behavioral and profile data, use Liquid templating to personalize copy, connect external data through Connected Content, and coordinate email, push, in-app messages, SMS, and Content Cards through Canvas flows.
Those capabilities answer important questions:
- Who qualifies for this campaign?
- Which channel should carry the message?
- When should the message be sent?
- What profile attributes or event data should be inserted?
They do not automatically answer a different question: which of several valid messages should this individual receive now?
Today, that choice is commonly represented through campaign branches, audience filters, Liquid conditionals, or tests. For example, a team might send a stronger offer to customers with low purchase frequency, product education to customers who browsed a category, and a reminder to customers who abandoned a cart. These rules can be sensible and useful. They are also designed in advance, usually around a handful of signals that a marketer has chosen to prioritize.
The problem is not that those rules are wrong. It is that customer intent is rarely expressed through one clean attribute. It appears in combinations of recency, frequency, channel response, product interest, purchase history, discount sensitivity, and changes in engagement. A static branch structure can only cover so many combinations before it becomes difficult to maintain and still leaves many people in a default path.
A re-engagement Canvas in practice
Consider a synthetic but familiar setup. A mid-size DTC brand has a Canvas for customers who have not purchased in 90 days. The team has four message variants:
- A 15 percent incentive for customers who have responded to discounts before.
- A product discovery message for customers who have browsed recently.
- A social-proof message for customers who opened recent emails but did not click.
- A low-frequency reminder for customers showing very little engagement.
The marketer can implement these paths in Braze with segmentation, Liquid, and Canvas branching. The team can also run an experiment to compare the variants across a population. That is a strong operating model for learning which messages work on average.
But the average result does not necessarily identify the best message for each person. A customer who opened two emails last week, viewed a product twice, and previously used a discount may fit multiple branches. A customer who has shown no recent activity may still have a strong preference for one product category. The campaign needs a choice among plausible messages, not only a qualification decision.
A per-person decision layer evaluates the available variants against the signals available at send time. It can select the message most likely to produce the desired outcome for that customer, while still leaving Braze responsible for orchestration, delivery, frequency controls, and reporting. The marketer defines the candidate content and the business objective. The decision layer handles the final selection.
This is a capability gap, not a criticism of Braze
It is worth being precise here. Braze is not failing at its core job. Its job is to help teams collect customer data, build audiences, design cross-channel journeys, personalize messages, and measure engagement. It does that exceptionally well for many lifecycle programs.
The gap is narrower: Braze does not yet provide a general-purpose, native layer that chooses among multiple content variants for every individual based on their changing behavioral signals at send time. Marketers can approximate this with rules, experiments, Liquid, external services, and campaign branches. Those approaches can work, especially when the number of variants and signals is small. They become harder to manage when every new campaign adds another set of conditions.
This distinction also avoids a common implementation mistake. Teams sometimes try to put all decision logic into Liquid. Liquid is excellent for rendering known customer attributes, formatting values, and handling clear conditional personalization. It was not designed to become a continuously adapting policy for selecting among many possible messages. When that logic grows, the template becomes difficult to review, test, and explain.
What Braze already does well
The case for an additional decision layer depends on respecting the strengths of the platform underneath it.
Intelligent Timing helps marketers avoid treating every customer as if the best send moment were the same. That is a meaningful form of individualization, particularly for teams managing high message volume across channels. Segmentation depth is another major advantage. Braze can bring together events, attributes, purchases, engagement history, subscription status, and other signals in ways that support detailed audience construction.
Canvas is also a strong foundation for lifecycle work. Teams can coordinate entry criteria, delays, branching, message steps, exit rules, and channel fallback without building the entire journey from scratch. Connected Content can bring in external information, while Content Cards provide an always-available surface for customers who are not ready to respond to a push notification or email.
These features reduce the operational burden of running campaigns. A decision layer should add to that system, not replace it. Braze remains the place where marketers govern eligibility, journey structure, consent, channel selection, and delivery. The additional layer focuses on choosing the most relevant content option within those boundaries.
How to add decisioning without adding campaign chaos
The practical approach is to start with a bounded set of variants in an existing Canvas. Do not begin by sending an unconstrained message generated for every customer. Begin with two or three approved options that have a clear relationship to the campaign objective.
For a win-back campaign, those options might be a product reminder, a category recommendation, and an incentive. The team can define guardrails such as margin protection, promotion eligibility, frequency limits, and exclusions for recent purchasers. The decision layer can then use behavioral signals to select among the approved options while Braze handles the journey and delivery.
Measurement should also reflect the decision being made. A single campaign-level click rate can hide whether the selection process is improving outcomes for different customer groups. Teams should track the chosen variant, the signals available at selection time, the eventual conversion or engagement event, and the fallback path when confidence is low.
For a small team, this structure is more maintainable than creating a new branch for every hypothesis. It keeps the marketer close to the content and business rules, while allowing the selection process to respond to more than one predefined segment. It also makes a useful distinction between testing content and testing decisions. A variant can perform well overall but still be a poor choice for a specific behavioral context.
We see the opportunity as complementary to the Braze stack. The platform already provides the campaign control plane. What many teams still need is a way to make the last content choice with the same customer-level precision they already expect from send-time optimization.
That is where lift often gets left behind: not in the ability to send another message, but in deciding which of several relevant messages should reach this person right now.