With a customer base numbering in the hundreds of millions, more than 20 brands, and nearly 100 product lines, the challenge of customer engagement is not simply reaching more people. It is making better decisions before every interaction: who to engage, when to engage them, and what to say.
For one leading consumer goods group, that challenge had become increasingly difficult to manage.
The company had run more than 6,000 marketing campaigns, while some of the most complex customer journeys in its traditional marketing automation (MA) system had grown to more than 200 nodes.
As these journeys became more complicated, the underlying business logic did not become easier to reuse. Reviewing historical journeys, adapting existing strategies, and rebuilding campaigns became increasingly labor-intensive.
The company also explored adding AI models to its existing MA system, but the underlying workflow remained largely unchanged. Every new campaign still required teams to translate business requirements into rules, manually define audiences, and build journeys node by node.
To change this, the company adopted DeepZero Agentic MA, an agentic marketing automation system in which multiple AI agents work toward the same business objective across audience selection, journey orchestration, content generation, and performance analysis.
Instead of treating each campaign as a standalone workflow, Agentic MA connects decisions and outcomes across campaigns. Results from one cycle are fed back into the next, allowing audience strategies, journeys, and content to improve over time.
A repurchase campaign for one of the group's diaper products shows how this works in practice.
AI Audience Agent: Turning a Business Goal into a Target Audience
Traditionally, audience selection required marketers to translate campaign requirements into combinations of customer tags, transaction history, and behavioral conditions. The quality of the resulting audience depended heavily on the operator's experience.
With Agentic MA, marketers can start with the business objective itself.
For a repurchase campaign, the AI Audience Agent interprets the intent behind the campaign and analyzes historical data across customer profiles, tags, orders, purchase timing, behavioral events, and model outputs.
It then identifies customers who are more likely to repurchase or move to the next product size and automatically generates the corresponding audience strategy.
What previously required marketers to manually translate business intent into targeting rules can now begin directly from the business need.
AI Journey Agent: Combining Large and Specialized Models to Orchestrate the Next Best Journey
Once the audience has been identified, the next question is what should happen next.
Should a customer receive a replenishment reminder? Is it time to recommend the next diaper size? And when is the right moment to reach them?
Previously, these decisions were largely based on predefined product-cycle rules, while more complex journeys still had to be assembled manually, one node at a time.
With Agentic MA, the AI Journey Agent begins with the repurchase objective.
A large language model interprets the campaign goal, identifies candidate products, and invokes the appropriate specialized models. These models then evaluate factors such as the customer's product line, current size, purchase quantity, and purchase timing to predict the likely repurchase window, potential size progression, and recommended product.
Based on those predictions, the system automatically generates the marketing journey, including timing, replenishment and size-progression branches, and subsequent content steps.
The journey is validated before launch and can then move directly into operation.
AI Content Agent: Creating Content Around the Purchase Decision
Once the journey has been determined, the next task is deciding what to communicate.
The AI Content Agent retrieves relevant brand guidelines, product selling points, and content rules from the company's AI knowledge platform. It then combines that information with the campaign strategy, recommended product, and available promotional offers.
Content is generated according to the characteristics of each channel, including SMS and WeCom, so that the message reflects both the customer's current need and the context in which it will be delivered.
Rather than generating marketing copy in isolation, the content agent works from the same purchase decision that drives the audience and journey strategy.
28% Higher Repurchase Conversion—and a Feedback Loop for the Next Campaign
The process does not end when the customer makes a purchase.
New order data and journey performance are returned to the AI Data Analytics Agent, which analyzes questions such as:
Which customer segments converted more effectively?
Which messages performed better?
Which journey paths should be retained or adjusted?
Those findings then inform the next round of audience selection, journey orchestration, and content generation.
The repurchase journey is now running continuously, and repurchase conversion has increased by 28% compared with the company's previous fixed-rule approach.
The significance is not only the improvement in a single campaign. The larger change is that marketing decisions no longer have to be rebuilt from scratch every time a new campaign begins.
Turning Every Purchase into an Input for the Next Growth Decision
Agentic MA is not designed simply to automate a set of predefined customer journeys.
It brings multiple AI agents together around a shared business objective, connecting audience decisions, journey orchestration, content, and campaign outcomes around the customer's next purchase need.
Each interaction produces new signals. Each campaign creates new evidence. And those results become inputs for the next decision.
That is how marketing automation moves from executing workflows to continuously improving how growth decisions are made.
It is also what AI for Growth means at DeepZero: applying AI not only to automate marketing operations, but to turn intelligence into measurable business growth.
