Key Takeaways
- Agentic AI isn’t replacing marketers. It’s helping them make better decisions and execute faster.
- The strongest AI workflows combine specialized agents with reliable data and human oversight.
- Signal-driven CRM outperforms calendar-driven automation by responding to customer behavior in real time.
- Teams that invest in strong data and workflow foundations are best positioned to scale AI successfully.
Many CRM programs keep running simply because they work well enough.
The same workflows, schedules, and campaigns continue year after year, even as customer behavior changes. Hayley Edwards of Superhuman challenged that mindset during Activate, arguing that the programs delivering the strongest results are often the least questioned.
The sessions throughout Activate pointed to the same shift. Traditional automation executes predefined rules. Agentic systems evaluate customer context before deciding what action to take.
Instead of asking whether a campaign should run, they ask:
- Is this the right customer?
- Is this the right channel?
- Is this the right moment?
- Is sending a message the best action?
That shift changes the role of marketing automation from executing schedules to reasoning about customer context.
Takeaway: Agentic CRM makes decisions based on customer context rather than predefined rules, creating experiences that are more timely, relevant, and adaptive.
What “Agentic” Actually Means in a Real CRM Stack
Agentic AI has quickly become one of marketing’s most talked-about ideas.
In practice, it isn’t a single AI system making every decision. It’s a coordinated set of specialized capabilities that work together while marketers remain responsible for strategy, oversight, and business outcomes.
Why One Big Agent Is a Recipe for Chaos
When AutoScout24 began exploring agentic CRM, Michela Maffei’s team reached an important conclusion: No single AI system can effectively run an enterprise CRM. Instead, the company divided responsibilities across specialized systems:
- Iterable orchestrates campaigns and customer journeys.
- Platform-native AI provides built-in intelligence, such as send time optimization.
- Custom AI agents apply AutoScout24’s business rules, operational processes, and customer context.
That architecture gives every agent the context it needs to make reliable recommendations.
As Maffei explained, fragmented customer identities, disconnected systems, and inconsistent event data don’t disappear when AI is introduced. They become sources of unreliable decision-making. Building a unified CRM foundation ensured every agent worked from the same customer understanding before recommendations were introduced into production.
The result was an execution layer the team could trust because every recommendation was grounded in consistent data and clear operational guardrails.
How Babylist Turned Natural Language Into Faster Execution
Babylist focused on a different challenge: helping a lean CRM team move faster. As the company expanded from email into an eight-channel engagement program and annual email volume grew from 5 million to 1 billion sends, production work threatened to consume the team’s roadmap.
Using Iterable’s MCP to connect Claude with Iterable, marketers can describe campaign updates in natural language instead of completing repetitive production tasks manually.
The impact has been significant:
- Email creation decreased from about one hour to roughly 15 minutes.
- Overall email production is on track to decline by 50%.
- The team now spends more time testing, optimizing, and planning future initiatives.
Babylist also introduced AI into campaign QA, using the same workflow to identify broken links, incorrect URLs, template issues, and copy errors before launch.
Flynn stressed those gains started with preparation. The team invested time defining prompts, documenting workflows, and establishing repeatable processes before introducing AI into production. That foundation made faster execution possible while maintaining quality across every campaign.
Takeaway: Agentic CRM succeeds when specialized AI capabilities are built on reliable data, well-defined workflows, and human oversight. That combination allows teams to move faster without sacrificing quality.
From Calendar-Driven to Signal-Driven: What Changes When the System Starts Reasoning
The most common form of marketing automation is a calendar. Build the campaign. Pick a send date. Launch to everyone. Repeat.
That approach creates consistency, but it also assumes customers are ready at the same moment. The teams at Carwow and Superhuman challenged that assumption by replacing fixed schedules with systems that respond to customer behavior in real time.
Why Marketing Teams Get Blocked, and What Actually Unblocks Them
CRM has always depended on customer data. The challenge is getting that data into marketers’ hands while it’s still useful.
Roman Petrochenkov described a familiar reality at Carwow. Customer data existed across the organization, but activating it often meant submitting a ticket, waiting through prioritization, and relying on engineering resources before a campaign could move forward. By the time new audiences were available, the opportunity had often passed.
Carwow focused on removing that operational friction. The team rebuilt its data foundation around three principles:
- Consistent definitions. A semantic layer ensures metrics like users, revenue, and engagement mean the same thing everywhere.
- Reliable AI context. Giving AI well-defined data reduces the amount of information it has to interpret or guess.
- Faster activation. A composable customer data platform allows marketers to build and sync audiences without waiting on routine engineering work.
Luke Trainer-Clark summarized the broader challenge well. Marketing teams rarely lack ideas. They lack timely access to the customer data needed to act on them. When that barrier disappears, personalization, experimentation, and campaign execution all accelerate.
Importantly, Carwow wasn’t trying to remove the data team from the process. The goal was to remove them from routine activation work so they could focus on building the architecture that makes faster marketing possible.
Takeaway: Agentic CRM depends on more than AI. It depends on giving marketers timely access to reliable customer data so they can act while customer intent is still relevant.
Grammarly Replaced Campaign Schedules With Customer Signals
Superhuman questioned something most lifecycle teams rarely revisit: The promotional program that generated more than $10 million in annual revenue. For years, Grammarly’s lifecycle team followed a familiar pattern.
- Build the promotional campaign.
- Send it to millions of users.
- Repeat the following month.
The program worked, making it easy to leave unchanged.
Hayley Edwards challenged the assumption that every customer should receive the same offer on the same schedule. Instead, the team built a behavioral decision system that evaluates over 80 customer signals before deciding when someone is most likely to convert. Those signals include product usage, engagement trends, upgrade page visits, and other behavioral indicators.
The new system uncovered insights the calendar had hidden for years. For example:
- Users were most likely to convert in the mid-afternoon, not at the team’s long-standing 5:00 a.m. send time.
- Sunday, once considered the worst-performing day based on team lore, turned out to be the highest-converting day.
- Campaign operations became dramatically more efficient, reducing manual effort from 30 hours to 10 hours per campaign cycle.
Those insights changed more than campaign timing.
They allowed the lifecycle team to stop optimizing around infrastructure constraints and start optimizing around customer behavior. The time saved also gave marketers more capacity to experiment, test new ideas, and improve the program over time.
Edwards revealed another important lesson, questioning another assumption. Larger discounts increased conversions in some international markets but also increased renewal churn because the underlying pricing didn’t match local purchasing power.
The solution was localized pricing, which increased year-over-year revenue in India by 30% while maintaining renewal rates.
Takeaway: Calendar-driven automation assumes every customer follows the same path. Signal-driven systems evaluate each customer’s behavior and respond when the timing is right, producing better outcomes while giving marketers more time to improve the experience.
How to Scale Agentic AI Without Scaling the Risk
The biggest challenge with agentic AI is building systems that people trust enough to use in production.
The practitioners running these programs shared a common approach: AI earns responsibility over time. It starts by observing, then recommending, and only later supports execution after people have confidence in the outputs.
The Case for AI That Observes Before It Acts
At AutoScout24, AI doesn’t begin by making customer-facing decisions. It begins by learning. Michela Maffei described a staged rollout designed to build confidence before expanding AI’s role in production:
- Read-only mode: AI observes workflows, investigates issues, and makes recommendations.
- Sandbox testing: Teams validate recommendations in a controlled environment.
- Human approval: Marketers review outputs before they’re introduced into production.
- Scaled execution: Once trust is established, AI supports larger operational workflows.
That progression protects both customers and the business. AutoScout24 also maintains clear boundaries around sensitive decisions. AI never communicates directly with customers about consent or personally identifiable information.
Those decisions remain with legal, customer care, and business owners. As confidence in the system grew, the business saw practical benefits:
- Faster campaign production.
- Less time spent investigating delivery issues.
- Improved deliverability.
- More time for marketers to focus on strategic work instead of operational troubleshooting.
Maffei summarized the broader lesson simply: AI scales the system you already have. Strong CRM foundations become stronger. Fragmented systems become more fragmented. No AI model can compensate for poor operational design.
The Four Elements That Make Agentic AI Work
Across AutoScout24 and Carwow, one pattern emerged consistently. Successful AI adoption depends on four fundamentals:
- A clean foundation that gives AI consistent customer context.
- Strong governance that defines where AI can and cannot act.
- Human judgment for decisions involving customers, risk, and business strategy.
- Customer relevance as the objective behind every recommendation.
Roman Petrochenkov reinforced the same principle from a data perspective. AI performs best when it has less to guess.
Carwow’s investment in semantic definitions, governed data, and structured activation reduced ambiguity before AI entered the workflow. That architecture allowed marketers to move faster because the underlying information was already trustworthy.
The result is fewer decisions being slowed down by fragmented systems, inconsistent data, and repetitive operational work.
Takeaway: Speed and governance reinforce one another. Teams move faster when AI operates on reliable data, within clear boundaries, and alongside human judgment.
Automation Was the Floor. Agentic Is What’s Above It.
Babylist, AutoScout24, Carwow, and Superhuman approached agentic AI from different starting points. Each solved a different operational problem:
- Babylist accelerated campaign production with natural-language workflows.
- AutoScout24 rebuilt its CRM architecture around specialized AI agents.
- Carwow removed the data bottlenecks that slowed customer activation.
- Superhuman replaced fixed campaign schedules with behavioral decisioning.
Different use cases. One direction. Marketing teams are moving beyond systems that simply execute predefined workflows. They’re building systems that evaluate context, respond to customer behavior, and help marketers make better decisions at scale.
The shift to agentic execution is already underway. The competitive advantage won’t come from adopting AI first. It will come from building the data, workflows, and governance that enable AI to deliver reliable decisions every day.
Frequently Asked Questions (FAQs)
What is agentic AI in marketing?
Agentic AI combines customer data, business rules, and AI reasoning to recommend or execute the next best action. Unlike traditional automation, it evaluates customer context before deciding what should happen.
How is agentic AI different from marketing automation?
Traditional automation follows predefined rules and schedules. Agentic systems consider customer signals, business context, and operational guardrails before determining whether, when, and how to act.
What does an agentic CRM workflow look like?
An agentic CRM workflow typically gathers customer signals, evaluates context, recommends the next best action, and executes through existing marketing systems with human oversight where appropriate.
Why is governance important for AI in CRM?
Governance establishes the boundaries that allow AI to operate safely. Clear data standards, human approval processes, and defined responsibilities help teams trust AI outputs in production.
Do agentic AI systems replace marketers?
No. The examples from Babylist, AutoScout24, Carwow, and Superhuman show AI reducing repetitive work and improving decision-making while marketers continue to provide strategy, oversight, and business judgment.