PropellerAds rebuilt its NIKO campaign agent into a conversational co-pilot on July 27, 2026, extending chat control across targeting, bidding, creatives and reporting for every ad format the network sells. The company reports internal test results that it has not opened to outside review.

What changed on July 27

The announcement, issued from Limassol, Cyprus, describes a change in what NIKO is rather than an addition to what it does. Until this release, according to PropellerAds, the agent functioned as a setup assistant. The rebuilt version is positioned as a full campaign co-pilot able to manage an advertiser workflow end to end, entirely through conversation.

That framing carries a specific technical consequence. Campaign creation, editing and monitoring now run across every format the network offers - Push Notifications, Onclick (Popunder), Telegram Ads, Interactive Ads and Paid Social Traffic - without the advertiser opening the platform dashboard. The interface has not been supplemented. It has been made optional.

PropellerAds describes NIKO as functioning like a 24/7 account manager, combining fifteen years of the company's advertising operations with AI to launch campaigns, optimise bids, analyse performance and answer platform questions in real time. The upgraded agent is available immediately to all PropellerAds advertisers at no additional cost, according to the company.

The targeting parameters now exposed to chat

The detail that separates a conversational wrapper from a working operator sits in the parameter list, and the announcement is unusually specific about it.

Advertisers can set full targeting through the chat interface, covering GEO, region, city, operating system, device, browser, language, internet service provider, connection type, and VPN or proxy filtering. Dayparting schedules are configurable in the same conversation. So is creative management, along with zone whitelists and blacklists - the inclusion and exclusion lists that determine which publisher placements a campaign is permitted to reach.

That last item matters more than its position in the list suggests. Zone-level control is the primary lever media buyers on performance networks use to isolate profitable traffic and cut the rest. Exposing it to natural language changes the operational profile of the tool from a launcher to something closer to a day-to-day optimisation surface.

The recommendation engine was extended in parallel. Bid and test-budget guidance now appears at any stage of a campaign rather than only during setup, and the agent can retrieve performance statistics, account balances and campaign details on request. PropellerAds states that NIKO was trained on the company's full Knowledge Base, allowing it to answer platform questions and propose strategies matched to an advertiser's vertical and goals rather than returning generic model output.

Training an agent on proprietary platform documentation is the same architectural choice made across the sector over the past year. It addresses the failure mode that has dogged general-purpose assistants applied to advertising work: fluent answers about advertising in the abstract, wrong answers about the specific platform in front of the user.

The numbers, and what stands behind them

Three figures carry the announcement. Each comes from PropellerAds internal testing.

Advertisers using NIKO reach a first campaign draft eleven times faster than through the manual setup flow, according to the company. Campaigns launched through the agent have generated two to four times more revenue than manually configured ones. And advertisers who supply more detailed campaign and funnel information see up to 144% more conversions than those providing minimal input.

The third figure explains the mechanism PropellerAds credits for the first two. The company attributes the gains to how the conversation is structured: NIKO walks advertisers through targeting, bidding and creative decisions in sequence, which the company says leads them to configure campaigns more completely and to notice optimisation opportunities they would otherwise skip. Read that way, the revenue difference is partly a measure of configuration completeness rather than of model quality - a distinction that matters when assessing whether the result would hold for an experienced buyer who already fills in every field.

No sample size, test window, vertical breakdown or control methodology accompanies any of the three numbers. The 144% conversion figure is explicitly conditional on advertiser input depth, which means it describes a correlation between how much a buyer tells the agent and how the campaign performs, not an effect the agent produces on its own.

Julia Larionova, Head of Marketing at PropellerAds, framed the release as a structural change rather than a feature addition. "We didn't just add features to NIKO - we rebuilt what it means to run a campaign on PropellerAds," according to Larionova. She added that "a large majority of advertisers who've tried the new NIKO come back to it for their next campaign." That retention claim is not quantified in the announcement.

Vendor-supplied performance figures have become a recurring measurement problem across agentic advertising. Criteoreported up to 60% improvement in recommendation relevancy from its Agentic Commerce Recommendation Service on February 5, 2026, also from internal testing. A study circulating from June 8, 2026 reported that AI agents cut task time by 87%, with the data supplied by Perplexity, whose two products were the only ones compared and three of whose employees appeared among the four authors. Speed gains have tended to hold up across independent and vendor studies alike. Claims about revenue, quality and scope have proved softer.

Marketplace-level data has been less flattering than single-vendor tests. DataBeat found agentic buyers clearing 13.4% below conventional programmatic demand on cost per thousand impressions across its network in May 2026, against earlier vendor-reported results such as the December 2025 PubMatic and Butler/Till campaign that logged 40% more impressions and 5.5 times cost efficiency for a single brand test.

Chat as the default entry point

PropellerAds is not moving alone, and the direction of travel across the industry over the past nine months has been consistent.

Amazon Ads introduced Ads Agent at its unBoxed conference on November 11, 2025, processing natural language instructions to build campaign structures across Amazon Marketing Cloud and Amazon DSP. Yahoo embedded agentic capabilities directly into its demand-side platform on January 6, 2026, in one of the first implementations where agents could autonomously execute campaign operations rather than only recommend them. PubMatic launched AgenticOS on January 5, 2026 with WPP Media, Butler/Till and MiQ as early participants. LiveRamp deployed live agents for audience building and cross-media measurement on March 3, 2026, and Skai positioned an agent-native marketing operating system in May 2026.

Google consolidated its separate in-product assistants into a single agent at Marketing Live on May 20, 2026, connecting Google Ads, Google Analytics, Google Marketing Platform and Merchant Center through one conversational interface. Displayce brought the same pattern to digital out-of-home at Cannes Lions on June 25, 2026, exposing planning data to external assistants through the Model Context Protocol.

PropellerAds sits at a different point in that market. Founded in 2011, the company serves media buyers, agencies and affiliate marketers across more than 195 countries and regions, holds ISO/IEC 27001 certification and is a member of IAB Europe, according to its announcement. Its formats - push, popunder, Telegram, interactive and paid social - are performance channels bought by operators who run many campaigns at thin margins rather than by brand teams working through holding companies. Setup time and configuration completeness translate directly into unit economics in that segment. An eleven-times reduction in time to first draft, if it holds outside internal conditions, is a different kind of proposition there than it is inside an enterprise console.

One structural difference is worth noting. Most of the announcements listed above route through shared protocols. The Ad Context Protocol launched on October 15, 2025 with six founding companies, built on the Model Context Protocol, and the IAB Tech Lab has since gathered agentic work under the AAMP umbrella. The NIKO announcement makes no reference to interoperability standards. The agent is a native interface to one network, not a component other agents can call. That keeps the deployment simpler and keeps it closed.

The forecast and its counterweight

The pitch accompanying the release opens with a Gartner projection that 60% of brands will use agentic AI to deliver personalised customer interactions by 2028. The same figure has appeared repeatedly in vendor announcements this year, including the Displayce release in June.

It travels alongside a second Gartner number that appears far less often in press materials. The research firm predicted on June 25, 2025 that more than 40% of agentic AI projects would be cancelled by the end of 2027 on escalating costs, unclear business value and inadequate risk controls, a figure PPC Land has tracked as a counterweight to adoption forecasts. The two are not in conflict. Together they describe a market where adoption is broad and survival is selective, and where agents attached to specific, measurable, time-consuming tasks are better placed than agents whose value is diffuse.

PropellerAds also cites Salesforce State of Marketing 2026 research finding that 87% of marketers now use generative AI in at least one workflow, and argues that only a fraction have embedded it into daily execution. That gap is the one the company says NIKO is built to close.

Capital has been moving in the same direction. Artificial intelligence sales and marketing startups raised $3.7 billion globally during 2026 by early May, with investor conviction centred on systems that execute marketing tasks autonomously rather than assist with them.

What is testable and what is not

For media buyers, the parts of this release that can be checked are the parts that describe scope. The format coverage, the targeting fields, the dayparting and zone-list controls, and the on-demand retrieval of stats and balances are all verifiable inside an account within an hour of use. So is the pricing, which is zero.

The performance figures are a different category. Eleven times faster to a first draft, two to four times more revenue, and up to 144% more conversions on richer input are internal measurements released without methodology, and they describe advertisers who chose to use a new tool against a baseline of manual configuration. Self-selection alone could account for a meaningful share of any gap.

What the announcement does establish, independent of the numbers, is a direction that no longer looks experimental. When a network selling push and popunder inventory to affiliate buyers makes conversation the primary control surface and charges nothing for it, chat has stopped being a premium layer sold to enterprise accounts. It has become table stakes at the performance end of the market.

Timeline

Summary

Who: PropellerAds, a global multisource advertising network founded in 2011 and headquartered in Limassol, Cyprus, serving media buyers, agencies and affiliate marketers across more than 195 countries and regions. Julia Larionova, Head of Marketing, provided the company's comment.

What: A rebuild of NIKO, the company's AI campaign agent, from a setup assistant into a conversational co-pilot that creates, edits and monitors campaigns across Push Notifications, Onclick (Popunder), Telegram Ads, Interactive Ads and Paid Social Traffic without dashboard access. Chat control extends to GEO, region, city, operating system, device, browser, language, ISP, connection type and VPN or proxy filtering, plus dayparting, creatives and zone whitelists or blacklists. The recommendation engine now supplies bid and test-budget guidance at any campaign stage and retrieves stats, balances and campaign details on request. Internal testing figures cited are eleven times faster to first draft, two to four times more revenue than manual configuration, and up to 144% more conversions where advertisers supply richer campaign and funnel detail.

When: Announced on Monday, July 27, 2026, with the upgraded agent available immediately.

Where: Issued from Limassol, Cyprus, covering PropellerAds advertisers across the network's operating footprint of more than 195 countries and regions.

Why: The company positions the upgrade against a gap between generative AI adoption and daily execution, citing Salesforce research that 87% of marketers use generative AI in at least one workflow and a Gartner projection that 60% of brands will use agentic AI by 2028. For the marketing community, the significance sits less in the individual claims than in where the pattern has now reached. Conversational campaign control moved through enterprise consoles and demand-side platforms across late 2025 and the first half of 2026, and its arrival as a free default on a performance network selling push and popunder inventory marks the point at which the interface stopped being a premium enterprise feature. The performance figures, however, remain internal, unaudited and unaccompanied by methodology, at a time when marketplace-wide measurement of agentic buying has been running below what individual vendors report.