Ask a marketing team if they’re ready for Braze Decisioning Studio data requirements, and you’ll usually get a nervous laugh. They picture a fully mature CDP (Customer Data Platform). Years of squeaky-clean historical data. A 360-degree customer profile that took two years and three vendors to build. None of that is actually the bar.

BrazeAI Decisioning Studio™ is a reinforcement-learning decisioning layer that sits between your first-party data and your customer engagement platform, personalizing channel, message, offer, timing, and frequency for each individual customer. We’ve already covered how it plays with Klaviyo elsewhere on this blog, so we won’t rehash the basics here. This piece goes one layer deeper, into the data itself.

Here’s the thing. The real bar for a first-party data foundation isn’t size or completeness. It’s whether the data relevant to your decisioning use case is usable, connected to the right customer, and capable of producing a clear outcome signal. Braze’s current guidance identifies activation and conversion event data as the bare minimum for Decisioning Studio. Customer profile data and relevant features can provide additional context and improve model performance. A dedicated CDP and a fully enriched customer profile are not universal prerequisites.

Let’s cut to the chase. 

What data does BrazeAI Decisioning Studio use?

Before we talk fields and schemas, let’s reframe the problem. Decisioning Studio needs usable signals that connect customer context, decisions, and outcomes.

The engine needs to connect the actions it takes with customer responses and business outcomes. Braze’s current guidance identifies activation events and conversion events as the bare minimum data for Decisioning Studio. Engagement data can provide additional behavioral context when it is relevant to the decisioning use case.

This feedback is what enables the decisioning agent to learn. Braze describes AI decisioning as using reinforcement learning and contextual bandits to test available actions and update how it chooses among them based on outcomes.

If you’re already a Braze customer, Braze says your Braze data is integrated into Decisioning Studio. For data outside Braze, the setup depends on your architecture and use case; Braze supports bringing relevant warehouse and other first-party data into Decisioning Studio. Conversion data is especially important when the business outcome happens outside Braze.

What data principles matter for Decisioning Studio?

Braze’s documentation points to several practical requirements that affect how useful the model can be: reliable identity, relevant event and outcome data, appropriately refreshed customer features, and sound data quality. These are practical readiness principle, not a single official four-item checklist.

Reliable customer identity. Data needs to be attributable to the customer at the level required by the decisioning use case. Braze emphasizes that identity resolution needs to happen before data is passed to its AI decisioning system. For Braze customers, customer features are aggregated to the customer level and can be associated with users through Braze identifiers such as external ID.

Keep event signals and customer features conceptually distinct. Events capture actions that happen at particular times, while customer features summarize the customer’s state or behavior. For Decisioning Studio, preserving useful event timing can matter for learning and attribution, while features are generally maintained as customer-level aggregates.

Refresh customer features on a time-driven schedule. Braze’s current guidance says features should be updated on a time-driven schedule rather than an event-driven one. The appropriate cadence depends on the feature and use case; daily may be appropriate in some implementations, but it is not a universal requirement.

Use relevant, trustworthy data. Braze says event data does not have to be 100% perfect, but the data still needs to be useful for the decisioning use case. Relevant signals, reliable identity, sensible feature values, and usable conversion timing matter more than simply maximizing the amount of data sent to the model.

Picture a brand that only tracks a “total emails sent” counter instead of individual send events. Decisioning Studio has no way to know which specific message came before which conversion. Or a brand whose “days since last login” feature only refreshes when someone actually logs back in, so it silently goes stale for anyone who’s lapsed. Both are principle violations, not edge cases. 

What data is required vs. optional for Decisioning Studio?

Strip it down, and Braze’s current guidance identifies two data categories as the bare minimum: activation events and conversion events. Customer profile data and relevant features provide important context and can improve model performance, but they are not presented by Braze as a universal requirement that every use case must satisfy in the same way.

Conversion data is especially important because it provides the outcome the decisioning agent is trying to optimize. Braze notes that conversions can include a value such as revenue net of promotion costs, and the timing of conversions is important for decisioning. Attribution can use increasingly indirect signals—from a recommendation ID, to product or channel information, to offer information, and finally proximity when more direct signals are unavailable.

Other signals, such as login history, product usage, transactions, marketing engagement, survey responses, propensity scores, or environmental context, can be useful when they are relevant to the decisioning use case. Braze recommends selecting features for relevance and quality rather than sending every available attribute.

What does BrazeAI Decisioning Studio NOT require?

This is where most of the anxiety actually lives, so let’s take it apart myth by myth.

Myth 1: “You need a dedicated, CDP-grade identity resolution system first.” 

Reality: A separate CDP is not a universal prerequisite, but identity resolution still matters. Braze says identity resolution needs to happen before data is passed to its AI decisioning system. The practical requirement is that the signals used by the agent can be reliably associated with the intended customer.

Myth 2: “You need years of clean historical data before you can start.” 

Reality: Backfilling exists for specific situations- a new feature you want to add, recovering from a pipeline outage, a change in how you define a metric- not as a blanket entry requirement. And here’s the honest answer on lookback windows, because we’d rather tell you straight than guess: there’s no universal number. Braze is explicit that the required history depends on your specific use case, and recommends confirming it with their AI Decisioning Services team rather than assuming a fixed benchmark applies to you.

Myth 3: “Everything has to update in real time.” 

Reality: Not every signal needs to update in real time. Braze’s feature guidance says customer features should be refreshed on a time-driven schedule rather than an event-driven one. The correct cadence depends on the feature and use case. Avoid turning a daily example into a universal product requirement.

Myth 4: “Every possible attribute needs to be captured before launch.” 

Reality: Braze’s current minimum is activation and conversion event data. Other customer attributes and behavioral features should be selected based on whether they help the agent distinguish among the available actions and improve the decisioning use case.

Myth 5: “The data has to be flawless.” 

Reality: Braze says event data does not have to be 100% perfect. That does not mean data quality is irrelevant: the signals still need to be relevant, usable, correctly attributable, and fresh enough for the use case. Braze’s feature guidance specifically calls out coverage, granularity, freshness, validity, sparsity, and correlation.

Think about a mid-market retailer starting with the minimum outcome loop: activation events plus a conversion feed, then adding relevant customer features that help the agent distinguish among the actions it can take.

What about support-ticket history and NPS scores? Well, they are pushed to phase two, on purpose.

What does Decisioning Studio data readiness actually mean?

Decisioning Studio’s data foundation can be smaller than many teams assume. Braze’s current guidance says the bare minimum is activation and conversion event data. From there, reliable identity, relevant customer features, appropriate feature freshness, useful event timing, and sound attribution can strengthen the decisioning system. A mature CDP, years of history, and a flawless dataset are not universal prerequisites.

Readiness for AI decisioning is a question of data discipline, not data volume. Teams that get the structure right can start smaller, and faster, than they assume.