Is Iterable an Agentic CEP? A Feature Review
Nova Agent and Nova Intelligence, scored against the four capabilities that define an agentic CEP: goal input, autonomy, monitoring, learning.
·
Blogs
·
Is Iterable an agentic CEP?
Short answer: further along than most, but still reactive at the core. In April 2026, Iterable launched Nova Agent, extending its earlier Nova Intelligence suite with campaign building, auditing, personalization, QA, and experimentation triggered through natural-language prompts. Iterable says over 90% of its customers already use some form of AI agent inside their programs, and the company has been genuinely aggressive about shipping agentic infrastructure faster than most CEPs in this comparison set.
But "ships AI agents fast" and "is an agentic CEP" are different claims. Using the same working definition we apply across this site, an agentic CEP takes a goal, plans the execution, runs it, watches for behavioral shifts on its own, and adjusts based on outcomes — largely without a human configuring every step. That's four capabilities: goal input instead of instruction input, autonomous execution, proactive monitoring, and outcome learning. This piece scores Nova against each one, using Iterable's own product materials plus independent third-party testing of the Nova MCP server.
What Iterable has built toward agentic execution
Nova Intelligence is Iterable's umbrella for AI features: Predictive Goals (churn risk and conversion probability scoring), Send Time Optimization, Channel Decisioning, Brand Affinity scoring per user, and Journey Assist, which turns a text description into a draft journey flow. Nova Agent, added in April 2026, sits on top of that suite as a conversational layer — a marketer can ask it to summarize a journey, review an experiment, or audit a campaign for QA issues, and it responds using natural language.
Iterable also shipped an open-source Model Context Protocol (MCP) server, letting tools like Claude Desktop query campaign and journey data conversationally. That's a meaningful step ahead of platforms that still require opening a dashboard to check performance — a marketer, or an external AI tool, can simply ask a question and get an answer sourced from live account data.
Where it gets more nuanced is autonomy. Independent testing of Iterable's MCP server found it can tell you how a journey performed when you ask. It does not proactively flag that a send underperformed unless someone goes looking. Goal-setting is also explicitly manual: Predictive Goals requires a marketer to define the business objective first; Nova analyzes data around that goal rather than proposing one on its own.
Scoring Iterable against the four capabilities of an agentic CEP
1. Goal input instead of instruction input
Partial. Iterable's own materials describe the model as "you define the strategic outcome; agents handle the ongoing work to move customers toward it" — which sounds like true goal input. In practice, that framing applies most cleanly to Predictive Goals, where a marketer still has to name the objective (reduce churn, increase conversions) before Nova does anything. Journey Assist works from a text description of the experience you want, which is closer to a detailed instruction than an open-ended goal. The system does not yet propose a campaign or journey from a business outcome with no further input.
2. Autonomous audience and content execution
Partial, similar pattern to most of the field. Send Time Decisioning, Channel Decisioning, and Frequency Optimization run continuously once enabled, and Nova Agents generate brand-aligned copy variations without a human writing each one. That's real autonomous execution on the content and delivery-timing side. Audience building is weaker: per independent testing, Iterable "reacts to events" and needs a human or an external workflow to define which events and segments matter — it does not identify a segment worth targeting on its own.
3. Proactive behavioral monitoring
Weak, and this is the clearest gap versus Iterable's own marketing language. Iterable's blog states the system "monitors real-time signals: customer behavior, engagement trends, and contextual triggers," which implies proactive detection. But independent testing of the actual MCP server found the opposite in practice: it can answer a question about Tuesday's send when asked on Thursday, but it will not surface that Tuesday's send underperformed unless a person prompts it first. There is no publicly documented case of Nova independently flagging an anomalous cohort, a churn spike, or a broken journey step without being asked.
4. Outcome learning and self-adjustment
Not demonstrated as a closed loop. Iterable's marketing describes agents that "feed outcomes back into future decisions, refining performance over time." Independent testing found a narrower reality: engagement models are reassessed on a schedule (weekly, per that testing), and there's no evidence Nova rewrites campaign strategy on its own based on what happened last week. A marketer still reviews results and manually adjusts targeting or timing — the system optimizes within parameters it was given, rather than changing the parameters themselves.
Iterable vs an agentic-by-design CEP
Capability | Iterable (Nova Intelligence + Nova Agent) | Agentic-by-design CEP |
Input model | Marketer defines the goal or the journey description; Nova executes within it | Goal or outcome, platform plans the full execution |
Content generation | Yes — real-time, brand-aligned copy variations | Yes, plus audience and timing decisions |
Audience building | Reacts to defined events; human sets the logic | System-proposed, human-reviewed |
Anomaly detection | Answers when asked; not proactive per independent testing | Surfaces shifts before a report is run |
Learning loop | Scheduled model reassessment; no self-rewriting strategy | Adjusts based on measured outcomes |
Where it sits | Extensive AI layer added across an instruction-based core | Agentic execution is the core architecture |
How Iterable compares to Braze on the same framework
Iterable and Braze land in a similar place on this scorecard, for similar reasons: both have shipped real, production AI agent infrastructure, and both fall short on the same two capabilities — proactive anomaly detection and closed-loop outcome learning. See the full Braze scoring breakdown for the side-by-side detail. Iterable's edge is breadth (Nova Agent's conversational layer spans more of the workflow, from QA to experimentation) and its MCP server, which lets external AI tools query account data directly — something Braze hasn't shipped in the same form. Neither platform will independently tell a marketer their campaign is underperforming before they ask.
The verdict
Iterable is agentic in coverage, not agentic in architecture. Nova Intelligence and Nova Agent represent one of the more complete AI feature sets among traditional CEPs — predictive scoring, send-time and channel decisioning, conversational campaign QA, and an open MCP server most competitors don't have. For teams that want AI assistance layered onto an existing instruction-based workflow, that's a genuinely strong toolkit.
What it is not, based on Iterable's own documentation and independent testing of the live product, is a system that starts from a business outcome, builds the full campaign itself, watches for problems without being asked, and rewrites its own strategy based on results. Those three gaps — autonomous audience building, proactive anomaly detection, and closed-loop learning — are exactly the four capabilities this framework uses to separate "AI features on a traditional CEP" from a platform that is agentic by design. If your team wants goal-in, campaign-out execution where the system also tells you when something's wrong before you go looking, that's a different category than what Nova currently ships.
See also
What Is an Agentic CEP?
What Is an Agentic CEP?
What Is an Agentic CEP?
An agentic CEP receives goals, not instructions — then plans, executes, and monitors campaigns autonomously. Here's what that means in practice.
An agentic CEP receives goals, not instructions — then plans, executes, and monitors campaigns autonomously. Here's what that means in practice.
See what Sortment can do for your goals.
See what Sortment can do for your goals.
Book a 30-minute call. We'll show you how the pilot works with your data and your stack.
Book a 30-minute call. We'll show you how the pilot works with your data and your stack.
AGENTS
CASE STUDIES
RESOURCES
AGENTS
CASE STUDIES
RESOURCES
Is Iterable an agentic CEP?
Short answer: further along than most, but still reactive at the core. In April 2026, Iterable launched Nova Agent, extending its earlier Nova Intelligence suite with campaign building, auditing, personalization, QA, and experimentation triggered through natural-language prompts. Iterable says over 90% of its customers already use some form of AI agent inside their programs, and the company has been genuinely aggressive about shipping agentic infrastructure faster than most CEPs in this comparison set.
But "ships AI agents fast" and "is an agentic CEP" are different claims. Using the same working definition we apply across this site, an agentic CEP takes a goal, plans the execution, runs it, watches for behavioral shifts on its own, and adjusts based on outcomes — largely without a human configuring every step. That's four capabilities: goal input instead of instruction input, autonomous execution, proactive monitoring, and outcome learning. This piece scores Nova against each one, using Iterable's own product materials plus independent third-party testing of the Nova MCP server.
What Iterable has built toward agentic execution
Nova Intelligence is Iterable's umbrella for AI features: Predictive Goals (churn risk and conversion probability scoring), Send Time Optimization, Channel Decisioning, Brand Affinity scoring per user, and Journey Assist, which turns a text description into a draft journey flow. Nova Agent, added in April 2026, sits on top of that suite as a conversational layer — a marketer can ask it to summarize a journey, review an experiment, or audit a campaign for QA issues, and it responds using natural language.
Iterable also shipped an open-source Model Context Protocol (MCP) server, letting tools like Claude Desktop query campaign and journey data conversationally. That's a meaningful step ahead of platforms that still require opening a dashboard to check performance — a marketer, or an external AI tool, can simply ask a question and get an answer sourced from live account data.
Where it gets more nuanced is autonomy. Independent testing of Iterable's MCP server found it can tell you how a journey performed when you ask. It does not proactively flag that a send underperformed unless someone goes looking. Goal-setting is also explicitly manual: Predictive Goals requires a marketer to define the business objective first; Nova analyzes data around that goal rather than proposing one on its own.
Scoring Iterable against the four capabilities of an agentic CEP
1. Goal input instead of instruction input
Partial. Iterable's own materials describe the model as "you define the strategic outcome; agents handle the ongoing work to move customers toward it" — which sounds like true goal input. In practice, that framing applies most cleanly to Predictive Goals, where a marketer still has to name the objective (reduce churn, increase conversions) before Nova does anything. Journey Assist works from a text description of the experience you want, which is closer to a detailed instruction than an open-ended goal. The system does not yet propose a campaign or journey from a business outcome with no further input.
2. Autonomous audience and content execution
Partial, similar pattern to most of the field. Send Time Decisioning, Channel Decisioning, and Frequency Optimization run continuously once enabled, and Nova Agents generate brand-aligned copy variations without a human writing each one. That's real autonomous execution on the content and delivery-timing side. Audience building is weaker: per independent testing, Iterable "reacts to events" and needs a human or an external workflow to define which events and segments matter — it does not identify a segment worth targeting on its own.
3. Proactive behavioral monitoring
Weak, and this is the clearest gap versus Iterable's own marketing language. Iterable's blog states the system "monitors real-time signals: customer behavior, engagement trends, and contextual triggers," which implies proactive detection. But independent testing of the actual MCP server found the opposite in practice: it can answer a question about Tuesday's send when asked on Thursday, but it will not surface that Tuesday's send underperformed unless a person prompts it first. There is no publicly documented case of Nova independently flagging an anomalous cohort, a churn spike, or a broken journey step without being asked.
4. Outcome learning and self-adjustment
Not demonstrated as a closed loop. Iterable's marketing describes agents that "feed outcomes back into future decisions, refining performance over time." Independent testing found a narrower reality: engagement models are reassessed on a schedule (weekly, per that testing), and there's no evidence Nova rewrites campaign strategy on its own based on what happened last week. A marketer still reviews results and manually adjusts targeting or timing — the system optimizes within parameters it was given, rather than changing the parameters themselves.
Iterable vs an agentic-by-design CEP
Capability | Iterable (Nova Intelligence + Nova Agent) | Agentic-by-design CEP |
Input model | Marketer defines the goal or the journey description; Nova executes within it | Goal or outcome, platform plans the full execution |
Content generation | Yes — real-time, brand-aligned copy variations | Yes, plus audience and timing decisions |
Audience building | Reacts to defined events; human sets the logic | System-proposed, human-reviewed |
Anomaly detection | Answers when asked; not proactive per independent testing | Surfaces shifts before a report is run |
Learning loop | Scheduled model reassessment; no self-rewriting strategy | Adjusts based on measured outcomes |
Where it sits | Extensive AI layer added across an instruction-based core | Agentic execution is the core architecture |
How Iterable compares to Braze on the same framework
Iterable and Braze land in a similar place on this scorecard, for similar reasons: both have shipped real, production AI agent infrastructure, and both fall short on the same two capabilities — proactive anomaly detection and closed-loop outcome learning. See the full Braze scoring breakdown for the side-by-side detail. Iterable's edge is breadth (Nova Agent's conversational layer spans more of the workflow, from QA to experimentation) and its MCP server, which lets external AI tools query account data directly — something Braze hasn't shipped in the same form. Neither platform will independently tell a marketer their campaign is underperforming before they ask.
The verdict
Iterable is agentic in coverage, not agentic in architecture. Nova Intelligence and Nova Agent represent one of the more complete AI feature sets among traditional CEPs — predictive scoring, send-time and channel decisioning, conversational campaign QA, and an open MCP server most competitors don't have. For teams that want AI assistance layered onto an existing instruction-based workflow, that's a genuinely strong toolkit.
What it is not, based on Iterable's own documentation and independent testing of the live product, is a system that starts from a business outcome, builds the full campaign itself, watches for problems without being asked, and rewrites its own strategy based on results. Those three gaps — autonomous audience building, proactive anomaly detection, and closed-loop learning — are exactly the four capabilities this framework uses to separate "AI features on a traditional CEP" from a platform that is agentic by design. If your team wants goal-in, campaign-out execution where the system also tells you when something's wrong before you go looking, that's a different category than what Nova currently ships.