The operators who can show their work

Performance marketing and cross-border operations have spent two years saturated with a single claim, that artificial intelligence was used and that it worked. The claim is cheap to write and, for most of the people who read applications and pitch decks, expensive to check. The interesting practitioners are the ones who arrive with the opposite problem. Their work can be traced, line by line, to what it changed, and the trail sits somewhere the practitioner does not fully control. Several of them surfaced in the Best in Business Awards 2025, in categories with little in common, cross-border legal recovery, business-to-business search, and e-commerce media buying.

Abilda Zhunis took the 2025 recognition for AI integration in logistics and supply-chain operations. She is an international lawyer with fifteen years in cross-border compliance and recovery, Head of Legal at Aurum International, a member of the International Bar Association and of the Association of Business Process Management Professionals, and the author of a monograph, Anti-Sanctions Guide: International Recovery, drawn from that practice. The award was for a platform she architected herself, Compliance & Recovery Hub, which is what fifteen years of that work compresses into software.

A cross-border recovery file is slow for a specific reason. The same regulatory term carries one meaning in a statute and another in how a given court enforces it, and a recovery case accumulates correspondence in several languages before it resolves. The platform puts a natural-language layer on the regulatory and legal text, scores compliance obligations and recovery cases for risk, drafts the multilingual correspondence the file generates, and connects through an API to the ERP and logistics systems where the obligations actually sit. The risk scoring decides which of a few hundred open obligations a team looks at first, a triage that used to run on a senior lawyer’s memory. The parts of that design that are arguable Zhunis argued in print, in Business Law International and in the Journal of Robotics, Artificial Intelligence and Law.

“The problem with compliance and recovery work across multiple jurisdictions is not that the rules are complicated — you can learn complicated rules,” Zhunis has said. “The problem is that rules interact with enforcement reality differently in each context, and that interaction is not written down anywhere. What the AI system does is process the case history that makes those interactions visible. That is the input that changes the assessment.”

Mila Tkachuk, a Los Angeles digital-marketing manager and the author of scholarly articles in the field, won the performance-marketing category on a single business-to-business case. For an outsourced live-chat support service she reports, within six months, organic traffic up between 120 and 180 percent, top-ten Google positions on high-intent commercial terms, conversion up 25 to 40 percent, lead generation up 60 to 90 percent, and a falling cost of customer acquisition through the same window, against year-on-year growth of about 150 percent. The case is built and measured as one piece of work, with on-page search architecture, commercial-intent content, authority-building and the marketing-technology layer treated as a single system.

The easy mistake in business-to-business services is traffic that converts on nothing. What makes the Tkachuk file legible is that conversion and acquisition cost moved with the traffic in the same window, because the keyword set was chosen for commercial intent and the pages were built to close on it. A traffic chart on its own is the part of the story a client should trust least, and the file does not lean on one.

The panel reading those files was made of operators with the same habit of recording what their work did. Kateryna Roh came up through programmatic ad-tech, at Cadent, formerly AdTheorent, and through agency and e-commerce roles at Cuneo, Turboweb and 1818shop. She has directed more than 500,000 dollars in supervised media across a portfolio of over 400 e-commerce clients, holds a 66 percent average portfolio margin with individual accounts reaching the high nineties, and carries a Ukrainian Business Award and a 2025 ECDMA Gold. Alisa Aidarova builds AI marketing systems and is the author of AI-MOS Copilot, built after marketing roles at Blank8 from 2020 to 2024; her Joies campaign returned a 320 percent ROI, and her case record runs from a jewellery-brand launch in a crowded niche through a repositioning to a 2025 Chicago market launch, with the method now being written up for VAK-listed papers. Kirill Timofeev and Olena Iashchenko sat on the same 2025 panel.

A margin figure on its own tells a reader nothing until it is broken into the channel mix underneath it, and running media for four hundred e-commerce accounts is the kind of work that builds that reflex. Most growth a file claims is growth the platform would have produced anyway; the part worth scoring is the part the platform would not have produced on its own, and that is the line a portfolio operator looks for first.

Aidarova reads an AI claim the way she has to defend her own. The Joies campaign returned 320 percent, and that figure only carries weight because the decisions the model tested can be named, decisions a person would never have had time to reach. AI-MOS Copilot widens how many campaign decisions get tested in a cycle; the judgement on each one still belongs to the marketer, which is the part of the claim that makes the number defensible at all.

Timofeev judges from the acquisition side. He runs Mad Lead Printer out of Belgrade and is a managing partner and fractional marketing chief at GFS Group Tradelines, a Los Angeles credit marketplace he has worked since 2019, where revenue grew roughly elevenfold while the cost of acquiring a customer fell from about 1,700 dollars to about 850. His agency portfolios are run to the same kind of figure, around a fifth of client revenue driven by email lifecycle work, with cold-outreach response rates he can quote from memory.

A growth multiple with no acquisition cost beside it is half a sentence. Eleven times revenue means one thing if the cost of a customer held and a very different thing if it doubled to get there, and the files that hold up are the ones that put both numbers on the same line and let a reader divide one by the other. The GFS figures are quotable because they were instrumented that way from the start.

These files share a property more telling than the size of any single result. Each one is fastened to something a third party could in principle pull on. Zhunis’s platform method is documented in journals anyone can read and disagree with. Tkachuk’s result is a ranked keyword and a tracked conversion line on a named account. Aidarova’s is a campaign with a measured before and after on the same book of business. The figures here are as the submissions to BIBA 2025 reported them; the programme published its judging criteria but not the individual scores, and none of the systems was opened for inspection for this piece. Said plainly, what remains is still uncommon in this market. The people behind these files brought work whose effect was recorded somewhere the entrant did not fully control.

Most awards in marketing and operations never reach that question. The claim is cheaper to accept than to test, and the people accepting it are seldom the people who would know where to push. A practitioner who says a model drove a result is rarely asked for the line the result would have followed without it, because asking is slow and the answer is often dull. The cases that survive contact with a working panel are the ones where the instrumentation was built before anyone asked for it. Tkachuk’s attribution ran while the campaign ran. Zhunis’s interpretation layer existed as a defended argument in print before it was a trophy. Roh reads margins broken out by channel because that is how she runs her own accounts. The habit shows in a file long before any judging starts.

It also carries a cost the practitioner absorbs and the claim-writer skips. Documenting a method precisely enough to publish it invites someone to disagree with it. Tracking acquisition cost honestly through a growth campaign means owning the month it moved the wrong way. Breaking a portfolio margin into its channels exposes the channels that lost money. The operators in this cohort pay that cost as routine, which is why their numbers carry information a cleaner-looking file does not. A result presented with its own failure modes attached is a different object from a result presented alone, and an evaluator who has run the same kind of work can tell them apart on the first page.

The 2025 list in these categories also named Satenik Ghazaryan among its winners. A year from now the citations will have faded and the checkable things will not have. Zhunis’s papers will still be on the shelf, Tkachuk’s account will still carry its numbers, Roh can still break a margin down to the channel that earned it. The operators worth remembering from BIBA 2025 are the ones who built that trail before anyone told them it would be read.