What Amperity shipped
Amperity, the customer data platform used by more than 400 brands across retail, travel, financial services, and hospitality, launched an AI agent called Per on October 13. The product is built on top of Amperity's existing customer profiles and is designed to take a marketing team from a natural-language question to an approved action inside the tools they already use, without the dashboard-building, manual data pulls, and cross-team meetings that normally sit between asking a question and actually executing a campaign based on the answer.
The workflow follows a specific sequence designed to keep a person in the loop at the point that matters most. A user asks a question in plain language, the kind a marketer would ask a colleague rather than a query a database administrator would write. Per investigates what changed, identifies which customers are affected, and explains its reasoning in terms a marketer can evaluate without needing to understand the underlying data model. It then prepares an audience or an action, such as a campaign segment or a triggered message, and holds it for human review and approval before anything runs against live customer data or connected marketing systems.
The strategic move: from identity resolution to execution
Amperity's core business has been identity resolution, the unglamorous but foundational work of stitching together a single customer record across the dozens of systems a large retailer or hospitality brand runs. Per represents a deliberate move up the stack, from being the system that makes customer data trustworthy to being the system that acts on it. Kabir Shahani, Amperity's co-founder and co-CEO, framed the distinction in exactly those terms, arguing that the value of AI is not simply the answer it generates but what happens after.
That is a meaningful bet. Every customer data platform vendor is under pressure to show an AI story beyond faster dashboards, and the obvious next step is agentic execution. But execution changes the vendor relationship: a CDP that only serves up data carries limited blast radius if something goes wrong. A CDP whose agent prepares campaign actions and triggers workflows is now a system that can directly affect customer-facing marketing output, which raises the governance stakes considerably.
Why the approval gate is the whole story
Amperity built in a specific control: Per prepares an action and explains its reasoning, but a human reviews and approves before anything executes against enabled marketing and business tools. That is the right default, and it is the detail that separates a defensible agentic product from a reckless one. Geoff Clark, CRM Manager at early adopter Johnnie-O, drew the useful contrast directly, describing most AI tools as reactive and characterizing Per as proactive, surfacing what to act on before someone has to ask.
The catch with any approval gate is that it only works if the people approving are actually equipped to catch a bad proposal. Elise Keefe, Senior Manager of Business Intelligence at MacKenzie-Childs, credited the system with helping her team reach trusted, actionable findings faster by eliminating grunt work. That speed is the selling point, but speed and rigorous review pull against each other. A marketing team moving faster because an agent did the investigation needs a review process substantive enough to catch an agent's mistake, not just a rubber stamp on the way to execution.
What this means for data governance teams specifically
Customer data platforms have historically occupied a governance gray zone. The volume of personal information running through them rivals a core data warehouse, yet the marketing teams who typically own the vendor relationship rarely apply the same scrutiny a data engineering team would bring to a comparable system. An agent that can propose and execute actions against that data moves the platform from passive storage to active decision-making, which raises the stakes on that existing gray zone considerably and makes the ownership question harder to leave unanswered.
Any organization evaluating Per, or the equivalent agentic layer from a competing CDP, should ask who owns the audit trail for agent-proposed actions, whether that trail is reviewable by the same compliance function that audits other customer-data uses, and whether the approval step gets logged in enough detail to hold up if a regulator or an internal audit later asks why a specific segment received a specific message on a specific date.
The decision for CDOs and marketing technology leaders
The real decision here is not whether agentic layers on customer data platforms are coming. They are, across every vendor in this category, because the dashboard-to-answer workflow is a genuinely weak point in most marketing stacks today. The decision is whether to adopt early with retail-scale brands as the proof point, as Amperity's first two customers suggest, or wait for the governance tooling around agentic CDPs to mature alongside the agents themselves.
For PE-backed retail and commerce companies under pressure to show marketing efficiency gains quickly, the appeal of an agent that collapses weeks of manual segmentation work into a single conversation is real and immediate. The cost of getting the governance wrong is also real: a mis-targeted campaign at scale, triggered by an agent's flawed reasoning that an overworked reviewer waved through without a close enough look, is a brand risk that simply did not exist when a human built every segment by hand and understood exactly why each customer was included. That tradeoff deserves an explicit answer, with a named owner and a defined review standard, before the agent goes live, not an implicit one discovered for the first time after a bad campaign has already gone out to customers.
