When evaluating any AI-powered tool, marketers face a reasonable set of questions: What can it actually do? What won't it do? When the underlying technology inevitably evolves, can I count on consistent behavior?
These aren't questions about algorithms or model architectures. They're questions about trust.
With our Customer Data Agent, Amperity is giving marketers the ability to build segments and customer journeys through natural language. That's a significant expansion of what AI can do inside Amperity. It also means we need to be clear about the guardrails, permissions, and behavioral guarantees that keep this capability enterprise-ready.
From answering questions to taking action
Amperity launched AmpAI as a way to explore your customer data through conversation. You could ask complex questions, generate SQL queries, and visualize the results without needing to know query syntax. The AI helped you understand your customers better.
Customer Data Agent takes a fundamentally different step. It moves from informing decisions to executing them.
Where the original AmpAI could tell you how many customers made repeat purchases last quarter, Customer Data Agent can build the segment that captures those customers and construct a journey to re-engage them. It operates as an agentic system: one that doesn't just respond to questions but takes coordinated action on your behalf.
This is more than a feature enhancement. It's a shift in what the AI is capable of modifying. Customer Data Agent can now configure your tenant, creating new segments and journeys through multi-step plans that coordinate the necessary components automatically.
That level of capability requires an equivalent level of control. Here's how we built it.
Guardrails by design: permissions that travel with the user
The first principle behind Customer Data Agent is straightforward: the AI should never have more access than the person using it.
Customer Data Agent integrates with Amperity's existing permission framework at a deep level. Every action the AI takes on your behalf is subject to the same viewing and editing restrictions that apply when you work manually. If your user role doesn't allow you to create journeys, Customer Data Agent won't create them for you. If you can view segments but not edit them, the AI operates under that same constraint.
This isn't a surface-level check. Permissions are enforced at the request validation layer, meaning every action is authorized against your user role before execution begins. Capability and access stay aligned regardless of how a request is phrased.
Administrators still retain full control over who can use AmpAI. The feature can be controlled for specific users or for the entire tenant, giving your team the flexibility to roll out AI-assisted workflows gradually, starting with experienced users before expanding access.
The underlying principle: AI should amplify what you can do, not expand what you're allowed to do.
The approval layer: human oversight where it counts
Speed is valuable, but not at the cost of surprises. Customer Data Agent is designed to keep you informed and in control at every stage.
When you describe a segment target or journey request, Customer Data Agent doesn't immediately start building. Instead, it generates a plan: a structured outline of the segments it will create and the journey flow it will construct. You review this plan before anything executes.
This planning step serves multiple purposes. It lets you verify that the AI understood your intent correctly. It surfaces the logic behind the proposed configuration so you can evaluate whether it makes sense for your use case. And it gives you the opportunity to refine your request through continued conversation before committing to execution.
Once you approve a plan and execution begins, every output is visible for review. Created segments appear in a canvas view where you can inspect the underlying logic and see exactly how many customers match the criteria. Journeys are displayed in the same visual canvas you'd use when building manually, making it easy to audit the flow, timing, and channel configuration. This visibility isn't an afterthought; it's central to how the feature is designed, and you're never left wondering what the AI actually built.
But oversight doesn't end once you approve a plan. Customer Data Agent includes full undo capabilities that let you reverse course if the results don't match your expectations. You can undo the last change the AI made, rolling back a single action. Or, if you want a clean slate, you can delete all segments and journeys created during a conversation in one step.
This means experimentation carries minimal risk. You can explore what Customer Data Agent builds, evaluate whether it fits your needs, and remove everything cleanly if you'd rather take a different approach. There's no penalty for iteration.
The goal is transparency, not automation for its own sake. You should always understand what's about to change, and you should always have a clear path back if you change your mind.
Behavioral consistency across model updates
AI models improve over time. New versions bring better reasoning, more nuanced responses, and expanded capabilities. But for enterprise users, progress creates a legitimate concern: will the tool I rely on today behave differently tomorrow?
Amperity approaches model transitions with this concern front and center. Before any underlying model change ships to production, we run rigorous evaluation frameworks that measure AmpAI's performance across a comprehensive set of scenarios. These evals establish behavioral baselines and verify that updates improve capability without introducing regressions.
Our architecture also provides flexibility in how we adopt new models. Rather than custom fine-tuning (which would require retraining cycles every time a model updates), we rely on extensive prompt engineering to shape AmpAI's behavior. This approach lets us move to improved models more quickly while maintaining the consistency you expect.
What this means in practice: you can rely on Customer Data Agent to behave predictably even as the technology underneath continues to advance. Performance should only get better, never regress.
Practical implications for your team
Customer Data Agent is designed for marketers who understand campaign strategy and want faster paths from concept to execution. You don't need SQL expertise or deep technical knowledge to use it effectively. You do need clarity about what you're trying to accomplish.
For administrators, the key consideration is access control. Customer Data Agent respects your existing permission structure and you can determine which of your users can access AmpAI.
For practitioners, the best starting point is net-new segment and journey creation. Welcome series, win-back journeys, and milestone celebrations: these are scenarios where building from scratch aligns perfectly with current capabilities.
AI you can rely on
Customer Data Agent represents a meaningful expansion of what's possible inside Amperity. Natural language campaign building, coordinated segment and journey creation, and multi-step planning are capabilities that genuinely change how marketing teams can operate.
But capability alone isn't enough. Enterprise tools need to be trustworthy, predictable, and transparent. That's why we built Customer Data Agent with guardrails from the start: permissions that travel with the user, planning steps that keep humans in the loop, and evaluation frameworks that ensure consistent behavior across model updates.
Trust isn't a feature you add at the end. It's the foundation you build on.
For more information on Amperity’s AI capabilities, contact us for a demo.
Customer Data Agent FAQs
What is Amperity's Customer Data Agent?
Customer Data Agent is an agentic AI capability within Amperity that allows marketers to build customer segments and journeys using natural language. Rather than write SQL queries, you describe what you want to accomplish and Customer Data Agent creates a plan, executes it upon your approval, and delivers fully configured segments and journeys ready for activation.
Customer Data Agent builds on AmpAI, Amperity's conversational AI for data exploration. While AmpAI helps you ask questions and understand your customers, Customer Data Agent takes the next step by turning those insights into action.
What can Customer Data Agent create?
Customer Data Agent can create new audience segments and customer journeys through coordinated, multi-step plans. When you describe a campaign goal (such as "build a win-back journey for customers who haven't purchased in 90 days"), Customer Data Agent identifies the segments needed, builds the visual segment, and constructs the journey flow connecting those segments to your activation channels.
You can also use Customer Data Agent for strategic planning (turning a high-level objective into a concrete execution plan), data exploration (asking questions about your customer data to inform targeting decisions), and iterative refinement (adjusting segments or journeys through continued conversation).
How is Customer Data Agent different from a copilot-style AI assistant?
Most AI assistants in marketing platforms are advisory. They answer questions, suggest next steps, or help you write queries, but you still have to take action on their recommendations manually.
Customer Data Agent is agentic, meaning it can plan and execute complex, multi-step workflows on your behalf. When you approve a plan, Customer Data Agent creates the segments and journeys directly in your tenant. This eliminates the gap between insight and execution.
The key distinction is autonomy with oversight. Customer Data Agent handles the execution, but you remain in control through plan review before anything runs, full visibility into what was created, the ability to edit segments or journeys within the conversation, and undo capabilities to reverse changes if needed.
How does Customer Data Agent handle user permissions?
Customer Data Agent enforces the same permission boundaries that apply when you work manually. Every action the AI takes on your behalf is authorized against your user role before execution begins.
If your role doesn't allow you to create journeys, Customer Data Agent won't create them for you. If you can view segments but not edit them, the AI operates under that same constraint. Permissions are enforced at the request validation layer, meaning capability and access stay aligned regardless of how a request is phrased.
Administrators retain full control over who can access AmpAI and can enable or disable the feature for specific users or the entire tenant.
Can I undo changes made by Customer Data Agent?
Yes. Customer Data Agent includes full undo capabilities so you can reverse course if the results don't match your expectations.
You have two options: undo the last change the AI made (rolling back a single action) or delete all segments and journeys created during a conversation in one step (clearing the slate entirely). This means experimentation carries minimal risk. You can explore what Customer Data Agent builds, evaluate whether it fits your needs, and remove everything cleanly if you'd rather take a different approach.
What should I look for in an AI-powered CDP?
When evaluating AI capabilities in a customer data platform, focus on four dimensions:
Permission alignment. Does the AI respect your existing access controls? An AI agent should never have more access than the person using it. Look for platforms where permissions are enforced at the execution layer, not just the interface.
Transparency. Can you see what the AI plans to do before it executes? The best implementations generate a reviewable plan and show you exactly what was created, including segment logic and customer counts.
Reversibility. Can you undo AI-generated changes? Enterprise workflows require the ability to roll back mistakes without involving support tickets or manual cleanup.
Behavioral consistency. Will the AI behave predictably as underlying models update? Look for platforms that run rigorous evaluation frameworks before shipping model changes to production.
Customer Data Agent is designed around all four principles, with plan review before execution, canvas visibility into created assets, complete undo capabilities, and tested model transitions.