If you’re here, chances are that you’re a marketer using Braze and struggling with how to make the most out of the platform’s robust AI capabilities. We get it. Most of our clients are feeling the same way. Given the depth and breadth of Braze’s AI capabilities, most marketing teams we work with have access to more AI features than they have time, data, or bandwidth to use well.
That gap is where the real cost shows up — the cost of underutilizing tools that you’re already invested in. Not because the tool doesn’t work, but because nobody is handing marketing teams a step-by-step guide on how, when, and why to leverage the different capabilities available to them.
The teams getting real results out of BrazeAI aren’t the ones who turned on the most features. They’re the ones who treated this like a sequence: start with what your data and team capacity can support today, bank the quick wins, and use that momentum to take on the harder stuff later. Crawl, walk, run. It sounds obvious, but the reality is that most teams still skip it.
This post walks through that sequence. We’ll cover what to activate first, what it actually takes to set up each layer, and what you’re missing by leaving any of it switched off.
Start with Intelligence Suite
The first thing we recommend to almost every brand looking to leverage Braze’s AI features is the same: start with Intelligence Suite.
Intelligence Suite — Intelligent Timing, Intelligent Channel, and Intelligent Selection — doesn’t need custom models, new data pipelines, or engineering time. These features run on the engagement history already accumulating inside Braze, so all you need to do is turn the feature on and it starts learning.
Intelligent Timing sends each message when that specific user is most likely to engage, based on their historical interaction patterns. Intelligent Channel routes each user to the channel they actually respond to — not the one your team defaults to. Intelligent Selection shifts send volume toward better-performing test variants in real time, instead of waiting for a scheduled end date.
What it takes to set up: Intelligent Timing and Intelligent Channel need existing engagement history to work from. If you’re new to the platform or have thin behavioral data, the system falls back to population-level patterns while it learns — which still works, just with less individual precision. Intelligent Selection needs enough audience volume across your test variants to produce meaningful signals. Most teams can stand all three up in a single sprint with no engineering involvement.
What you’re missing without it: Sending at the wrong time is one of the most consistent suppressors of engagement, and it’s one of the easiest things to fix. Teams without Intelligent Channel send every message to every channel by default — which inflates costs, burns SMS budget, and trains customers to tune you out. The Intelligent Selection miss is quieter but more expensive: without it, you’re actively sending underperforming variants to real customers while waiting for a test to close. The cumulative revenue impact of that delay — across every test, every quarter — adds up faster than most teams realize.
What it looks like in practice: A national grocery retailer enabled Intelligent Timing as a single-sprint activation. AI-timed sends drove a 10%+ lift in sales compared to scheduled sends, generating nearly $160K in incremental revenue from one test. The message didn’t change. The timing did.
Then, Fix Campaign Production with Generative AI
Once Intelligence Suite is running, the next question clients usually ask is: where are we losing the most time?
For most teams, the honest answer is campaign production. Copy variants. Liquid logic. Pre-send QA. When a campaign takes two weeks to build instead of three days, you run fewer experiments, miss more seasonal windows, and spend your best hours on execution. We see this pattern across retail, media, QSR, and financial services — and it almost always traces back to the same bottlenecks.
Braze’s Generative AI tools are built to remove that friction. The AI Liquid Assistant lets marketers describe what they want in plain language and get working Liquid code back — collapsing the ticket queue. AI Content QA reviews messages before send, catching issues that would otherwise require a manual review pass. The AI Copywriting Assistant generates on-brand copy across channels directly in the platform. AI Image Generator brings visual asset creation into the same workflow.
What it takes to set up: These tools are available without complex setup, but their output quality depends entirely on what you put in. Vague prompts produce generic output — and generic AI copy is, in some ways, worse than no AI copy, because it creates work to fix. Someone on your team needs to own prompt design and output review. For the Liquid Assistant specifically, we recommend someone with Liquid familiarity review the output before it ships. And the bigger shift isn’t turning on a feature — it’s redesigning the campaign production workflow around it. Teams that bolt these tools onto an existing process see marginal gains. Teams that rebuild around them see the real efficiency jump.
What you’re missing without it: The cost of slow campaign production is easy to underestimate because it’s spread across everything. Fewer tests. More missed seasonal moments. The Liquid dependency is where we see the most painful bottlenecks: marketers who know exactly what personalization they want to build but can’t execute it without a developer. That ticket queue isn’t just an inconvenience — it’s a direct constraint on how personalized your customer experience can be. Pre-send errors are the sleeper miss. Broken Liquid, misconfigured filters, off-brand copy — these cause real damage to the customer experience that’s hard to quantify and easy to avoid.
What it looks like in practice: A retail CRM manager needed to personalize an email by each customer’s most recent purchase and loyalty status. Normally: a developer, a ticket queue, and days of waiting. She used the AI Liquid Assistant, described what she wanted, got working code back, reviewed it, and launched. The personalization that used to require dev dependency now runs on the marketing team’s timeline.
Next, Move from Reactive to Predictive with Predictive Suite
If you’re only messaging based on what customers have already done, you’re responding to signals after the moment of highest leverage has passed.
The Predictive Suite — Predictive Churn, Predictive Events, and Item Recommendations — uses behavioral data already inside Braze to score users by future likelihood, so you can act before the behavior happens. Predictive Churn identifies at-risk users before they lapse. Predictive Events scores users by likelihood to complete a target action, so you can prioritize high-intent customers instead of blasting your entire list. Item Recommendations surface the right product or content for each user at the moment of send.
This is where the data conversation starts to matter more. And it’s where we see the most teams stall.
What it takes to set up: Before we recommend activation, we have a frank conversation with marketing teams about data readiness. Predictive Churn requires a clear definition of what “churned” means for your business — a question that sounds simple but often sparks longer internal debates than expected. It also requires enough users in each category for the model to learn from. A rough guide: a few thousand users on each side of the outcome. Activating Item Recommendations needs a populated Braze Catalog and enough purchase or interaction history to surface meaningful affinity signals. What makes these tools faster to stand up than third-party predictive solutions is that the data is already in Braze — no new pipelines, no integration overhead. A partner can add value by helping your team define the right outcome, sanity-check model output, and build the campaign logic to act on scores before they go stale.
What you’re missing without it: By the time a customer shows obvious churn signals — declining opens, lapsed purchases — the window to keep them is already narrowing. Predictive Churn lets you intervene earlier, when retention campaigns are more likely to work and less likely to require a deep discount to do it. Without behavioral affinity data driving what surfaces at send, brands default to bestsellers or editorially curated picks. Customers who receive recommendations that actually match what they want convert at meaningfully higher rates. And if you know who’s most likely to upgrade, book, or buy — and you’re treating them the same as everyone else — you’re spending acquisition-level effort on customers who were already close to converting.
What it looks like in practice: A streaming platform noticed cancellation rates ticking up. Instead of waiting for customers to leave and running a win-back, they used Predictive Churn to identify at-risk subscribers before they lapsed — triggering a retention campaign the moment a user crossed a risk threshold. Item Recommendations personalized each message to the subscriber’s viewing history. The retention push reached the right customers before the decision was made.
Build Team Velocity with BrazeAI Operator™
A lot of AI impact doesn’t show up in campaign performance reports. It shows up in the questions that used to slow teams down.
How do I structure this Canvas? Why isn’t my Liquid rendering? What’s the right filter for this segment? In most organizations, those questions funnel to a developer or a Braze expert who has their own queue. That’s where the drag accumulates.
BrazeAI Operator™ is a conversational AI assistant built directly into the dashboard. It answers questions, walks through setup, troubleshoots Liquid, and helps with Canvas logic — in plain language, at the moment you need it. It’s page-aware, so it already knows where you are in Braze.
What it takes to set up: Operator has almost no technical barrier to entry — it’s on by default and requires no configuration. The real question is adoption. The teams that get the most out of it are the ones who deliberately redirect the habit of filing a ticket or pinging a developer toward trying Operator first. That sounds simple, but habits are sticky. We’ve seen teams sit next to a powerful tool and not use it because nobody made it the default behavior. One other thing: Operator is built for procedural and technical questions — Canvas architecture, Liquid syntax, segment logic, feature setup. It’s not a strategic advisor. Use it for “how do I build this,” not “should I build this.”
What you’re missing without it: Every time a marketer files a ticket for a technical Braze question, they’re pulling engineering capacity toward a problem that didn’t require it — and waiting days for an answer they needed in the moment. Multiply that across a team of five CRM marketers over a quarter, and the drag is significant. When campaigns take longer to build because marketers can’t resolve technical questions independently, fewer experiments run. Less learning means the same campaigns keep going out to customers who’ve seen them before.
What it looks like in practice: A CRM manager building a re-engagement campaign for a fitness app wanted to personalize messages based on each user’s favorite workout category. She knew the logic. She didn’t know the Liquid. Before Operator, that question went to a developer. Instead, she described what she was trying to build directly in the dashboard and got the exact Liquid structure she needed. The campaign launched the same day.
When You’re Ready to Generate Personalization — Not Just Optimize It: BrazeAI Agents & Agent Console™
Everything up to this point has been about using AI to work smarter inside a defined campaign structure. Agents operate differently. They don’t optimize a message. They generate it — in real time, for each individual, based on live data.
BrazeAI Agents are AI-powered components that live inside Canvas steps or Catalog fields. You define the prompt, the brand guardrails, and which data sources to pull from. The agent synthesizes profile attributes, catalog records, and Canvas entry properties into a unique output for every customer — then stores that output in Canvas context so it’s accessible across every subsequent step in the flow.
BrazeAI Agent Console™ is where you build, configure, and deploy those agents directly inside Braze. Organizations can bring their own LLM — Anthropic, OpenAI, Gemini — or use Braze’s in-house model. You control which data gets sent to the agent, keeping output focused and token usage efficient. The Braze MCP Server extends this further — a secure, read-only connection that lets external AI tools access non-PII Braze data without leaving your environment.
What it takes to set up: You need a defined use case before you open Agent Console — not “we want AI-generated personalization” but something specific: “we want to generate a personalized product recommendation intro paragraph based on a user’s last three purchases and their loyalty tier.” Vague briefs produce vague output. The attributes and events you want the agent to reference need to be clean, consistently populated, and accessible in the Braze profile or Catalog — gaps in the data show up directly in the output. You also need documented brand guardrails before you write a single prompt. And plan for a review cycle before you go live. The first version of an agent prompt almost never ships unchanged. Teams that skip that step tend to be the ones that walk away from Agents, saying they didn’t work.
What you’re missing without it: This is where personalization goes from “we reference your first name” to “this message was built for you.” You can’t write a unique paragraph for 500,000 customers. An agent can. That’s not a feature — it’s a fundamentally different way of running lifecycle marketing. The brands that figure it out early will have a compounding advantage over the ones still writing segmented batch copy.
What it looks like in practice: A retail brand came to Stitch with a generic 30-day onboarding series that wasn’t driving downstream engagement. We redesigned it around an in-app personality quiz — capturing crafting style, materials preferences, and project interests — and fed that data into a Braze Agent embedded in the onboarding Canvas. The agent generated personalized email copy for each new user, matching them to a community guide aligned with their interests. What had been a standard welcome sequence became a differentiated first impression.
To learn more about Agent Console & see a real use case, check out our on-demand Braze Innovation Series webinar.
For Brands Are Ready for Program-Level Optimization: BrazeAI Decisioning Studio™
Most optimization work happens at the campaign level — test a subject line, adjust send time, pick a winning variant. That’s useful. It’s also limited. You’re optimizing one touchpoint at a time. Those decisions don’t talk to each other.
BrazeAI Decisioning Studio™ operates at the program level. It runs continuous experiments across an entire lifecycle journey — offers, content, channels, timing, frequency — simultaneously, for every customer, against a defined business metric. It’s powered by decisioning agents, which don’t pick one winning combination for a segment; they learn what works for each individual and adjust accordingly.
A quick note: decisioning agents are different from BrazeAI Agents. BrazeAI Agents work at individual touchpoints. Decisioning agents orchestrate whole programs — running experiments continuously, learning from results, and improving over time. For organizations with the right foundation, it’s the most powerful capability in the Braze stack.
We’re candid with marketing teams about this one: it’s not for everyone, and it’s not where most organizations should start. The teams it’s right for have done the foundational work — clean data, defined conversion events, enough audience volume for the models to learn from. Retention, loyalty, and high-frequency lifecycle programs tend to be the best fit — see if it is right for you.
What it takes to set up: This is the BrazeAI feature that requires the biggest lift, because Decisioning Studio is only as good as the foundation underneath it. You need a single, clearly defined business metric before you turn it on — like revenue per member, retention rate, conversion to second purchase. If your team can’t agree on what success looks like first, the tool has nothing to optimize for. Conversion events need to be cleanly tracked and firing reliably. On an audience scale: a program with 10,000 active users probably doesn’t have enough volume. A loyalty program with 500,000 does. The organizational readiness question matters too. Decisioning Studio shifts control away from “we decided this campaign will run this way” toward “the model is learning what works.” That’s a meaningful change, and it requires buy-in from stakeholders used to having direct control over what goes out. Teams that struggle with it are usually the ones that didn’t have that conversation first.
What you’re missing without it: Campaign-level optimization has a ceiling. A customer who received a discount offer last week and a re-engagement push yesterday is experiencing your program as a series of disconnected moments, not as a coherent journey. For brands running loyalty or retention programs at scale, the revenue delta between “we pick the best campaign for each segment” and “the model learns what works for each individual” is significant — and it widens with every month of model learning.
What it looks like in practice: A large hotel chain with millions of loyalty members deployed a Retention Decisioning Agent with one goal: increase revenue per member. The agent ran continuous experiments at the individual level — offer type, channel, timing, frequency, message tone — learning that one member converts on Tuesday push notifications with upgrade offers while another responds to biweekly emails timed to her booking window. Over time, it stopped guessing and started knowing.
Where to Start
The right starting point depends on your goals, your data, and your team’s actual capacity. What we see most often is teams activating in the wrong order — not because they made a bad decision, but because nobody mapped the sequence to where they actually were. That mismatch is usually what opens the gap between “we have these tools” and “we’re getting results from them.”
The sequence in this piece is the one we walk most teams through. Intelligence Suite first — least infrastructure, fastest results, and it builds the internal confidence that makes harder capabilities easier to justify later. Generative AI next, once your campaign process is tight enough to know where the real friction is. Predictive Suite and Operator are where the data and velocity work starts to compound. Agents and Decisioning Studio are where the foundation you’ve built either pays off or exposes the gaps you skipped.
When teams stall, it’s rarely because the technology stopped working. It’s because something foundational wasn’t in place — a clearly defined outcome, enough audience volume, a campaign workflow built to support what the tool produces. The more sophisticated the capability, the more expensive that gap becomes.
Most organizations are somewhere in the middle: Intelligence Suite running inconsistently, Generative AI tools turned on but underused, Predictive Suite or Agents on the roadmap but not yet activated. That’s a fine place to be. The question is whether you’re moving through the stack with a plan — or just turning things on as they come up.
If you’re not sure where your organization sits, or which BrazeAI™ capability is the right next step, that’s exactly the conversation we’re here for.
