Question 1

What is AI decisioning?

Accepted Answer

AI decisioning uses reinforcement learning to make individual-level marketing decisions that maximize a KPI you choose (e.g., revenue, conversions, lifetime value). The agent experiments and learns via trial and error which actions perform best for each customer.

Question 2

How is AI decisioning different from rules-based decisioning?

Accepted Answer

Rules-based systems rely on static “if X then Y” logic that becomes complex and hard to maintain at scale and delivers less uplift. AI decisioning learns autonomously, handles many variables and data sources, and performs better at scale.

Question 3

How does AI decisioning differ from traditional “Next Best Action” approaches?

Accepted Answer

Traditional next best action uses supervised learning to create propensity scores, segments customers, runs A/B or multivariate tests to choose treatments per segment, then codifies rules. It’s laborious, statistically constrained at small segment sizes, and hard to maintain—often becoming “next best guess.” AI decisioning replaces this manual complexity with reinforcement learning that continuously tests and optimizes at the individual level.

Question 4

What kinds of decisions can AI decisioning make?

Accepted Answer

Frequency, send time, channel, incentive/promotion, and creative/message. It can also decide when to deploy promotions and when to avoid promotional budget.

Question 5

What data can AI decisioning use?

Accepted Answer

Large, multidimensional datasets. It learns which features are informative, including engagement (opens, clicks), purchases, browsing, device, and highly company-specific behavioral data (often the most predictive).

Question 6

What is Braze AI Decisioning Studio?

Accepted Answer

It’s presented as an AI decisioning platform where marketers set goals (e.g., profit, conversions, retention), ground choices in first-party data (in your systems or Braze Data Platform), and steer specialist “Composable Intelligence” agents with goals and guardrails. Agents test channels, messages, offers, and timing concurrently at enterprise scale, learn from every interaction, explain decisions, and remove the need to hand-build segments or static rules/journeys. Leading brands are seeing double-digit lift across acquisition, onboarding, retention, and win-back.

Question 7

How does AI decisioning compare to using LLM “agents” for decisions?

Accepted Answer

Three LLM uses are described: Data engineering: LLMs turn unstructured text into structured data, then decisions can be rules-based (e.g., via Braze Agent Console). Text generation: LLMs write personalized copy based on customer data. “Vibe decisioning”: asking LLMs to pick treatments (send day, creative). The speaker is skeptical because optimization is an empirical problem where LLMs aren’t set up to produce high-quality answers. LLMs are complementary to reinforcement learning: use LLMs for data engineering and copy; use reinforcement learning to optimize decisions (including which inputs/prompts to feed an LLM).

Question 8

Is AI decisioning “plug and play” or set-and-forget?

Accepted Answer

No. High-value, complex use cases require careful configuration and tuning by experts. Once live, marketers continue to iterate—updating creatives, adjusting guardrails, adding data, and aligning with evolving strategy.

Question 9

Will AI decisioning discover my strategy for me?

Accepted Answer

No. It executes tactics (what to say, when, how often) and can surface insights that inform strategy, but it does not define brand positioning or core strategic choices (e.g., whether to use promotions).

Question 10

Does AI decisioning only maximize clicks?

Accepted Answer

No. It maximizes whatever reward you choose (e.g., revenue, LTV, conversions). If you reward clicks, it will optimize clicks; if you reward revenue or LTV, it will optimize those.

Question 11

What implementation model does Braze use for complex AI decisioning?

Accepted Answer

A forward-deployed engineering model (Pro tier) that includes a forward-deployed data scientist to configure models to your KPI, data, and business specifics, validate setup, and ensure performance uplift.

Question 12

Why is AI decisioning hard even if I have a CDP or unified data?

Accepted Answer

Turning unified data into optimal, attributable decisions is a hard data science problem. Many companies invested in CDPs/warehouses but saw less value than expected because decisioning itself is complex; AI decisioning is designed to solve this.

AI Decisioning in Customer Marketing: Evaluating Reality vs. Hype

Scott Brinker

George Khachatryan

Frank Doyle

This webinar answers

Check out more Webinars

Grow with Braze Chicago

Grow with Braze Amsterdam

It's time to be a better marketer