Principal Basil C. Puglisi sees this week’s customer data news as a question every marketing leader will have to answer in writing: when an AI agent proposes a move on a customer segment, who approves it, and where does that approval live? On October 8, Amperity introduced Pér, an agent that takes a team from a question about customers and campaigns to an answer, a recommendation, and, with approval, action inside the tools the brand already uses. Two days earlier, Treasure AI launched Personalization Studio and said it would tie email pricing to customer actions such as clicks rather than send volume.
Both announcements move AI closer to the point where a marketing decision touches a real customer. That makes the approval step part of the marketing plan, and Puglisi wants it designed with the same care as the audience and the offer.
What Amperity and Treasure AI announced
Amperity describes Pér as grounded in a brand’s own identity-resolved customer profiles. The agent can investigate what changed, identify affected customers, explain the evidence behind a recommendation, and prepare an audience or action for approval. In a launch post, co-CEOs Kabir Shahani and Derek Slager wrote that the user can review the reasoning, revise the recommendation, reject it or approve it, and that the person remains accountable for the decision. Pér works inside the permissions and enabled workflows the enterprise already manages.
Treasure AI’s Personalization Studio lets marketers pick a page, choose an audience, and let AI propose the creative, after which the team reviews, tweaks, and launches. The company framed its new email pricing as a bet on its own intelligence: if the platform does not help drive engagement and conversions, the customer should not pay for send volume alone. Treasure AI did not publish rates or a start date in the release.
How Puglisi reads the approval step
Amperity’s review, revise, reject, or approve sequence lines up closely with the accept, modify, or reject decision at the center of Puglisi’s Checkpoint-Based Governance. In his vocabulary, that human decision at a binding point is AI Governance. Responsible AI is the machine-side checking that runs without a person, such as a rule that blocks an audience built on stale consent data before anyone sees it. Marketing teams need both, and the vendor’s approval button only becomes governance when the team decides who holds it and records what they chose.
His HAIA-RECCLIN method adds the piece most approval screens leave out. RECCLIN’s Reasoning format asks an AI to show its sources, score its own confidence, flag conflicts, and hand the final choice back to a person. An agent that says “send a win-back offer to 40,000 lapsed buyers” should also say which signals drove the pick, how sure it is, and what it chose to ignore. Pér’s promise to explain the evidence behind a recommendation is the right instinct, and the brand should check that the explanation is good enough to support a yes or a no.
The pricing news ties back to an older argument on basilpuglisi.com about how AI disrupted the traditional marketing funnel. When AI shortens the path from insight to send, volume metrics lose their meaning quickly, and a vendor that bills on clicks is moving its incentive toward the middle of the funnel where engagement happens. Clicks still sit upstream of revenue, so a click-priced contract rewards the vendor for attention while the brand still has to prove the sale.
A practical setup for marketing teams
Puglisi’s guidance for a brand testing an agent like Pér or a tool like Personalization Studio starts with an approval map written before the first recommendation arrives. The map names each action the agent can propose, such as building a segment, changing a website content spot, or launching a send, and assigns an owner who accepts, modifies, or rejects it. Low-risk actions on small audiences can carry a lighter review, while anything that changes pricing, offers, or messages to a large customer group gets a named approver and a written reason.
The approval log then becomes a marketing asset. A team that tracks how often recommendations are accepted unchanged, modified, or rejected learns where the agent is reliable and where the customer data underneath it is weak. A high modify rate on win-back audiences, for example, points to a data problem worth fixing before the team grants the agent more room.
For click-based pricing, the tactic is to model the contract against the outcome the business sells. Compare cost per click under the new model with cost per order and revenue per thousand sends under the old one, and ask the vendor how the agent’s recommendations will be logged so a click can be traced back to a decision. Puglisi expects more martech vendors to follow with outcome-linked pricing, and the brands that keep their own approval records will be the ones able to judge whether the outcome was the agent’s work or the team’s.
Sources
- Amperity. (2026, October 8). Amperity introduces Pér, an AI agent with a complete picture of every customer and the ability to act on it [Press release]. https://amperity.com/announcements/amperity-introduces-per-an-ai-agent-with-a-complete-picture-of-every-customer-and-the-ability-to-act-on-it
- Shahani, K., & Slager, D. (2026, October 8). Meet Pér. Amperity. https://per.amperity.com/blog/meet-per
- Treasure AI. (2026, October 6). Treasure AI launches Personalization Studio to turn customer context into real-time website experiences [Press release]. Business Wire via Yahoo Finance. https://finance.yahoo.com/technology/ai/articles/treasure-ai-launches-personalization-studio-143000164.html
Questions readers ask
What is Amperity’s Pér?
Pér is an AI agent Amperity introduced on October 8, 2026. It uses a brand’s identity-resolved customer profiles to answer questions about customers and campaigns, recommend a next step, and, with human approval, move an audience or action into the marketing tools the brand already uses.
Does Pér act without human approval?
Amperity says users can review the reasoning, revise the recommendation, reject it, or approve it before work moves into an enabled workflow, and that the person remains accountable for the decision.
What is Treasure AI’s engagement-based email pricing?
Announced on October 6, 2026, it ties email costs to customer actions such as clicks rather than the number of messages sent. Treasure AI did not publish rates or a start date in its release.
How does Checkpoint-Based Governance apply to marketing AI agents?
Checkpoint-Based Governance places a named human at each binding decision point to accept, modify, or reject AI output, with a record of who decided and why. For a marketing agent, that means assigning an approver to each type of action the agent can propose and logging every decision.
What should a marketing team measure when an AI agent recommends campaigns?
Track how often recommendations are accepted unchanged, modified, or rejected, and connect approved actions to orders and revenue rather than clicks alone. A rising modify or reject rate shows where the agent or its underlying customer data needs work before it gets more autonomy.
#AIg
Leave a Reply