An Agent Aimed At The Hard Problem
Rokt mParticle has launched a Performance Engine whose centerpiece is an Audience Agent, a system that reasons over a marketer's first-party data and proposes high-value audiences through a conversational workflow. Rather than dropping a marketer into a query builder, the agent interprets the underlying customer data, surfaces segments worth targeting, and hands them back for human approval. It is a pointed design choice. Most of the industry's AI investment has gone into the creative and execution end of the funnel, and Rokt is deliberately planting its flag upstream, at the audience and identity layer where campaigns are actually won or lost before a single ad renders.
Chief Technology Officer Sam Dozor draws the contrast sharply. "The industry is pouring its AI energy into execution, creative, copy, channel orchestration. The harder problem, the one that actually determines whether a campaign works, is who you reach, how accurately you can identify them, and how confidently you can predict what they'll do next. The agent does the heavy reasoning. The marketer keeps the judgment." We find that framing persuasive because it names the part of the stack that is genuinely defensible. Anyone can generate ad copy now. Fewer companies can resolve identity across fragmented first-party data and predict behavior with confidence, and that is where mParticle's customer data heritage gives Rokt an edge.
Four Accelerators Behind One Interface
The Performance Engine bundles four accelerators that the Audience Agent orchestrates: Audience Expansion, Household Reach, Match Boost and Predictive Audiences. Each targets a distinct failure mode in modern targeting. Audience Expansion widens a proven segment without diluting it, Household Reach extends targeting across the people who actually share purchasing decisions, Match Boost lifts the rate at which known customers are recognized on ad platforms, and Predictive Audiences forecasts who is likely to convert next. Packaged together, they turn a pile of first-party data into deployable, high-intent audiences that can be pushed to the channels a marketer already uses, without a data science team assembling each segment by hand.
What makes the bundle strategic is that the accelerators compound. Better identity resolution through Match Boost feeds more accurate expansion, which in turn sharpens predictive modeling. That is the advantage of owning the data layer rather than bolting AI onto execution: improvements accumulate across the whole pipeline instead of being trapped in a single campaign. For retail and CPG technology leaders evaluating this against point solutions, the question is whether their current stack lets these gains reinforce one another, or whether identity, matching and prediction live in separate tools that never share signal. Rokt's argument is that unifying them under one reasoning agent is where the outsized returns hide.
The Results Brands Are Reporting
The launch arrives with concrete numbers from recognizable brands. Hardee's saw a 117% increase in Google Ads match rates, while sister brand Carl's Jr. reported 62% on Google Ads and 22% on Meta. Beauty brand Tatcha drove 8.5x revenue and 5x conversion using Predictive Audiences. Match rate improvements may sound like plumbing, but they are among the most valuable gains in advertising today, because a higher match rate means more of the customers a brand already paid to acquire can actually be reached again. When the addressable base expands by double or triple digits, every downstream dollar works harder without any increase in media spend.
Cathy McGill, Director of Digital Products at CKE Restaurant Holdings, described the experience in operational terms: "We saw an immediate lift in reach after enabling Match Boost. It helped us connect with more of the audiences we'd already invested in, without any incremental effort." That phrase, without any incremental effort, is the commercial promise in miniature. The Tatcha figures point at the other end of the value chain, where predictive targeting reaches people before they have signaled intent elsewhere. We would want to see these results hold across more categories and longer windows, but the spread from quick-service restaurants to prestige beauty suggests the mechanics are not confined to a single vertical.
Human Judgment Stays In The Loop
A detail we would not gloss over is the approval step. The Audience Agent proposes audiences, but a marketer signs off before anything ships. That human-in-the-loop design is more than a safety feature in a regulated, brand-sensitive discipline like advertising. It is a governance model that lets teams adopt agentic tooling without surrendering accountability for who gets targeted and how customer data is used. Dozor's line that the marketer keeps the judgment is the operating principle, and for CPG leaders wary of black-box automation touching their audience strategy, it lowers the barrier to bringing an agent into a live workflow.
The engine is available to select clients now and expands through summer 2026. We read the phased rollout as a sign Rokt wants proof points before opening the gates, which is consistent with a company betting on trust in the data layer rather than a race to ubiquity. The larger strategic message for the market is that the audience and identity layer, long treated as unglamorous infrastructure, is becoming the battleground for AI differentiation in retail media. If Rokt is right that reach and prediction determine campaign outcomes more than creative does, then owning that layer with a reasoning agent is a genuinely strong position to build from.
Why This Matters For CPG Technology Leaders
For CPG and retail technology leaders, the Performance Engine is a prompt to reassess where first-party data actually lives in the organization and how usable it is. Many brands have spent years pouring investment into customer data platforms only to see that data sit idle, too raw for a marketer to activate without engineering help. An agent that reasons over that data and proposes audiences directly is an argument that the value of a data platform is realized only when activation is conversational and immediate. The lesson is that collecting first-party data is table stakes; the differentiation is how quickly a non-technical marketer can turn it into a targetable, high-intent segment.
There is also a competitive timing dimension worth flagging. As third-party identifiers continue to erode, the brands that build durable first-party identity and prediction capabilities will compound an advantage that rivals cannot easily buy back later. Rokt is positioning its Performance Engine as the on-ramp to that capability, and the reported match rate gains at Hardee's and Carl's Jr. show why identity resolution is not a back-office concern but a direct driver of reach and efficiency. We would advise leaders evaluating the space to press vendors on how their tools improve identity and prediction together, because that combination, not another creative generator, is what will separate the winners in the next phase of retail media.
