Alisa Aidarova designs and operates performance marketing systems for companies across the United States and Europe, working across time zones and markets rather than within a single geographic location.
Yet her Python scripts manage millions in ad spend for clients in the US and Europe. Her academic papers on machine learning get cited by researchers who couldn’t find her hometown on a map. She sat on the jury of Armenia Digital Awards 2024, judging campaigns from agencies with ten times her headcount. The International E-Commerce and Digital Marketing Association lists her among its Senior Members, a title usually reserved for corporate executives at major firms.
We talked for three hours about automation, obsolescence, and why an industry built on creativity is being eaten by people who think in code.
You say your automation systems save clients 30% on ad spend. That’s a big number. Break it down for me.
Advertising waste happens when nobody’s watching. Tuesday afternoon, everything looks fine. Cost per acquisition at $15, return on ad spend at 4x. You go home. Wednesday morning, you check the dashboard and see that overnight some competitor launched a flash sale. They flooded the auction. Your cost per acquisition jumped to $45 while you were asleep. Eight thousand dollars gone, acquiring customers at a loss.
This isn’t rare. It’s constant. Auctions run 24 hours. Competitors in different time zones launch at 2 AM your time. Platform algorithms shift based on user behavior that changes by the hour. Some video goes viral and suddenly your target audience is watching memes instead of shopping.
People can’t monitor this around the clock. We need sleep. We get distracted. After six hours of staring at dashboards, attention falls apart. The auction doesn’t care.
So in 2021 I started writing what I call intervention scripts. Programs that plug into advertising platforms through APIs and make decisions based on rules. Every fifteen minutes, the script pulls fresh data. Cost per lead too high? Pause the ad. ROI dropped? Cut the bid. Performance spiked? Pump budget before the window closes. Over time, these scripts evolved into a repeatable system rather than ad hoc tools
If it’s that straightforward, why isn’t everyone doing it?
Because it requires skills marketers don’t have and don’t want to learn. Writing these scripts means understanding programming, statistics, API architecture, and ad platform mechanics all at once. Marketing education covers none of this. The industry trained people to be storytellers and brand strategists. Asking them to build automated bidding logic is like asking a novelist to design a bridge.
There’s ego involved too. Marketers built careers on intuition and creative instinct. Hard to admit a machine makes better decisions than your gut. People would rather believe they have special insight algorithms can’t touch. They don’t, but the belief feels good.
And agencies have weird incentives. Many charge by hours worked or percentage of spend. Automation that cuts hours and optimal spend means less revenue. Why would they build something that shrinks their own invoices?
I work on performance with most clients. If I save them money, I make more. If I waste budget, I get nothing. Changes how you think about everything.
You published academic research on CatBoost for predicting customer lifetime value. Why does someone running campaigns bother with peer-reviewed papers?
Because marketing has an embarrassing relationship with evidence. Go to any marketing conference. Listen to the talks. Endless claims about what works. Storytelling increases engagement. Personalization drives conversion. Video beats static images. Authenticity builds trust. Everyone nods along. Then you ask for evidence, real evidence, not cherry-picked case studies, and the room goes quiet.
The whole industry runs on folk wisdom, vendor research designed to sell products, and anecdotes that support whatever someone already believed. Actual scientific methodology barely exists outside universities. Practitioners dismiss rigor as impractical. Academics dismiss practice as unsophisticated. The gap between them wastes enormous money.
My CatBoost paper tries to bridge that. The CatBoost paper addresses this gap using a concrete commercial use case.
Concrete business problem: predicting which customers will generate profit over their lifetime versus which will cost more to acquire than they ever return. Customer acquisition costs went up maybe 60% over five years. Treating everyone the same when half will never be profitable is just negligence.
CatBoost handles categorical variables well, which matters because customer data is full of categories. Region, acquisition channel, first product purchased. The algorithm finds patterns in historical data that predict future behavior. In the paper, I show implementation, validate against holdout data, document performance precisely. Anyone with similar data can replicate it. The methodology is described in sufficient detail to be replicated on comparable datasets
That’s what separates science from marketing folklore.
The interviewer notes that Aidarova’s peer-reviewed publications represent outstanding achievements establishing theoretical foundations for practical work, rare in an industry that usually keeps thinkers and doers in separate rooms.
Your other paper covers personalization. That concept has been floating around for twenty years. What’s actually new about your approach?
The execution, really. Everyone agrees personalization works better than mass blasts. Obviously. If I send you an offer for something you actually want when you’re ready to buy, you respond better than if I spam discounts to my whole database. Nobody disputes this.
Implementation is where it falls apart. Real personalization needs infrastructure most companies don’t have. Unified customer data across touchpoints. Website behavior, email engagement, purchase history, support tickets, app usage, all connected. Real-time processing to catch signals and respond before the moment passes. Models sophisticated enough to predict individual preferences instead of segment averages. Coordination between marketing, tech, and operations teams that normally don’t talk to each other.
What most companies call personalization is a joke. They segment email lists by three variables instead of two and call it a breakthrough. They stick first names in subject lines. They retarget people with ads for products they already bought. This isn’t personalization. It’s automated irrelevance.
The paper describes the technical architecture required for large-scale personalization, data pipelines, model training, trigger logic, measurement. Basically a blueprint. Few companies will actually build it. But those that do get advantages competitors can’t copy quickly. Infrastructure takes years. That’s a moat.
You judged Armenia Digital Awards 2024. What did you learn from reviewing international campaigns?
That geography tells you about resources, not capability. I looked at submissions from agencies in major capitals. Big staffs, expensive tools, fat budgets. Some of it was fine. Technically solid, strategically obvious, creatively safe. Nothing that made me sit up.
Then I saw work from smaller teams with real constraints. Tight money, limited platform access, tiny home markets forcing them international early. Some of that work was sharper than anything the big shops produced.
The pattern kept repeating. Constraints build capability. When you can’t solve problems by throwing money, you learn how things actually work. When premium tools aren’t available, you build your own. When comfortable domestic markets don’t exist, you compete globally from day one and develop resilience that sheltered competitors never get.
There’s a hiring implication here. I’d rather work with someone who built functional systems under pressure than someone who operated expensive platforms without understanding what’s underneath. First person adapts anywhere. Second person is lost when their familiar tools disappear.
iOS 14 privacy changes in 2021 wrecked a lot of advertising operations. Some agencies lost half their clients. You seem fine. What happened?
We saw it coming and prepared. When Apple announced App Tracking Transparency, anyone paying attention understood the implications. Facebook Pixel relied on browser cookies Apple was about to block. Without that feedback, the ad algorithm loses visibility. Did someone buy after clicking? Abandon cart? Become valuable or request refund? Algorithm wouldn’t know.
Most agencies waited. Maybe Apple would soften things. Maybe Facebook would find a workaround. Maybe it would blow over.
We started implementing server-side tracking immediately. CAPI sends purchase signals from client servers directly to ad platforms. Data never touches the browser. Apple’s restrictions don’t apply to server-to-server communication.
This is engineering work. You need backend access, data pipeline configuration, deduplication logic. Marketing teams can’t do it. So when iOS 14 actually dropped, the industry panicked. Dashboards showed zero conversions while actual sales kept flowing. Agencies built on browser tracking couldn’t prove results anymore. Client relationships collapsed.
Our clients kept visibility. Pipelines stayed intact. We could still optimize based on real performance. Some agencies lost major accounts that month. Those accounts called us.
Google keeps threatening to kill cookies. Deadline moves every year. How do you plan for something that might never happen?
You assume it eventually happens and build for that. Google said 2022, then 2023, then 2024, now maybe 2025, maybe never. Delays made people relax. Agencies that panicked initially calmed down when nothing changed. They’re still dependent on infrastructure that could vanish with one Chrome update.
We treated it as eventual certainty regardless of timing. First-party data collection became top priority. Every interaction that could capture an email, phone number, account creation should capture it. That data belongs to the business. No platform can take it away.
We also built identity resolution. Someone visits your site five times anonymously, then buys and creates an account. Now you can connect all that previous behavior to a known person. Complete journey data instead of fragments.
Clients who invested during the cookie scare own assets that get more valuable as third-party data disappears. Competitors who waited own nothing but platform dependency.
You were using GPT-3 before ChatGPT launched. What pushed you there so early?
Scale problems, mainly. In 2021 we were producing creative at volume. Ten product categories, five audience segments, three platform formats, four languages. Combinatorics explode. Hundreds of variants to test. Writers, even mediocre ones, cost enough that producing hundreds of text variants gets expensive fast.
GPT-3 offered a way out. PT-3 became a practical tool for handling creative scale
Feed it product info, set tone and format, get fifty headlines in minutes. Most needed editing. Some were garbage. But baseline quality beat what we could afford manually, and editing time was less than writing time.
By the time ChatGPT went viral, we had a year of experience. We knew where language models shine and where they fall apart. We’d developed prompting techniques through trial and error. We’d built workflows that integrated AI into production, rather than treating it like a toy.
Head start compounds. While competitors ran their first prompts, we were on third-generation systems. While they debated whether AI content could work, we had data proving it could.
The interviewer observes that Aidarova’s early adoption and systematic workflow development demonstrate extraordinary abilities to spot inflection points before they become obvious.
You’ve called traditional targeting dead. Spell out what’s dying and what’s replacing it.
The whole 2015-2019 paradigm is gone. Back then, marketers assumed they knew things platforms didn’t. You understood your customer. Women 25-35, into yoga and organic food, would buy your wellness products. You translated that into targeting parameters. The platform found matching people. Value created.
Two things killed this.
First, platforms accumulated behavioral data that buries anything marketers can access. Facebook knows which posts you linger on, which ads you click, what you buy, which accounts you secretly stalk, what content makes you angry enough to engage. Billions of users, more than a decade of data. Your customer insight from surveys and focus groups can’t compete.
Second, machine learning got good enough that algorithmic audience selection beats human parameters. When you restrict targeting to yoga-interested women 25-35, you’re telling the algorithm to ignore potentially valuable customers who don’t match your assumptions. Maybe the best buyers are actually men 40-50 who never flagged yoga interest but have purchase patterns that predict conversion. Algorithm would find them. Your parameters block them.
Winning approach flips the relationship. Stop constraining the algorithm with your limited guesses. Feed it better signals instead. Proper conversion tracking so the platform knows who actually buys, not just who clicks. Customer lifetime value data so it optimizes for profit, not just transactions. Rich product feeds so the system understands what you sell. Then get out of the way.
Marketing job changes accordingly. Less time picking audiences. More time building data pipelines. Less creative intuition. More statistical analysis. Less campaign management. More system architecture.
People who built careers on targeting expertise face obsolescence. The skill lost value. Painful but true.
You hire mathematicians and programmers over marketing graduates. Doesn’t that leave gaps?
Different gaps than you’d think. Marketing fundamentals transfer in weeks. Value propositions, customer psychology, funnel mechanics, channel characteristics. None of it is complicated. Smart people with no marketing background absorb it fast.
Technical skills take much longer. API integration, database architecture, statistical methods, programming logic. Months or years of study. Doesn’t transfer quickly regardless of intelligence.
So I pick: spend weeks teaching marketing to technical people, or spend months teaching tech to marketing people. Math says hire technical.
Beyond efficiency, technical training changes how people approach problems. Engineers think in systems. Inputs, outputs, feedback loops, failure modes. They expect precise measurement. They demand logical justification. These habits translate powerfully into marketing.
Marketing-trained people often think in stories. Narratives about why campaigns worked or failed. Feels satisfying but frequently misses actual causation. The campaign worked because our message resonated with the target audience’s desire for authenticity. Did it though? How would you know? What’s the counterfactual?
My team operates more like engineers. Watch dashboards, spot deviations, trace causes through data, implement fixes, measure results. Less romance. Works better.
What happens to marketing people without technical skills? Are they all doomed?
Not all, but plenty. Same thing that happened in manufacturing. Automation didn’t eliminate every factory job. It eliminated specific categories while creating others. Manual machine operators got replaced. People who design and maintain automated systems got hired. Net employment might be stable, but individual workers got displaced unless they adapted.
Marketing automation follows the pattern. Tasks humans do slowly and inconsistently get absorbed by machines. Routine ad copy. Targeting selection. Bid adjustment. Report generation. These don’t need human judgment anymore. People who built careers on them face hard choices.
New roles emerge at the same time. Training AI to produce useful outputs. Designing data architectures. Analyzing performance at granularity impossible before. Integrating marketing tech with business systems. These need capabilities most current marketers lack.
Transition will be rough. Retraining is possible but demands effort not everyone will make. Some will insist creativity and intuition remain irreplaceable. They’ll find out that assertion doesn’t pay rent. Others adapt, learn new skills, thrive as competition thins.
I tell my team: neural networks won’t take your job. People who know how to use neural networks will take your job. Pick a side.
You mentioned agency incentive problems. How does structure shape behavior?
Most agencies charge for inputs, not outputs. Hours worked, percentage of spend, retainer fees. Compensation for effort regardless of results. Think about what that encourages. Agency billing hourly benefits when work takes longer. Agency taking percentage of spend benefits when budgets grow, even if the increase creates no value. Agency on retainer benefits when clients stay dependent forever, whether or not that serves them.
These incentives don’t encourage efficiency or automation or permanent solutions. They encourage stretched engagements, maximized billing, and avoiding any innovation that might shrink scope.
Performance-based pay flips it. I only make money when clients profit, so I want to maximize their profit with minimum resources. Automation that cuts my workload while improving their results suddenly looks attractive. Solving problems permanently frees capacity for more clients. My interests and theirs line up.
This explains why marketing automation mostly comes from outside traditional agencies. Their business model actively discourages threatening innovation. Startups and independents without those legacy structures build what incumbents won’t.
You work with clients across the US, Europe, and the CIS. In a world of remote work, does geography still influence your results?
AA: Geography is no longer a constraint; it is now an economic variable that we can optimize. I view the location not as “where I sit,” but as “how we leverage time and talent.”
Technically, the landscape is flat. The interfaces we use—Google Ads, AWS, Python libraries—are identical whether you access them from New York, Yerevan, or London. The code runs the same way. However, operating as a distributed, global entity gives us two distinct advantages that a strictly local agency often lacks.
First is Time Zone Arbitrage. We turn the clock into an asset. When our North American clients are finishing their day, our analytics team is often just starting. This means we can process the day’s data, run our predictive models, and implement optimizations while the client sleeps. They wake up to solved problems and improved metrics. We effectively create a 24-hour production cycle.
Second is Global Talent Access. By not tethering ourselves to the most expensive square mile of real estate in a single capital city, we can invest our budget in R&D rather than rent. We hire engineers and data scientists based on their mathematical capability, not their proximity to a physical office. We find that talent in emerging tech hubs often has a higher retention rate and a stronger focus on deep work compared to the churn we see in hyper-saturated markets.
Does it have challenges? Of course. You have to navigate different legal frameworks and cultural nuances. But in the world of algorithmic marketing, math is a universal language. The algorithm doesn’t care where the command comes from; it only cares about the quality of the logic.
What separates clients who succeed with advanced systems from those who fail?
Mostly it comes down to data quality and whether the organization can actually coordinate across departments. My systems run on data. Garbage in, garbage out. If conversion tracking is broken, algorithms optimize for false signals. If lifetime value calculations are wrong, predictions target wrong customers. If attribution windows don’t match actual buying behavior, budget allocation makes bad calls.
Most companies have terrible data. CRM full of duplicates and outdated records. Analytics that misattribute conversions. Finance that can’t calculate customer-level profit. Silos preventing integration. Fixing this isn’t glamorous. Leadership would rather talk AI strategy than database cleanup. But strategy fails without foundation.
Organizational alignment matters just as much. Marketing automation touches multiple departments. Engineering implements tracking. Finance provides profitability numbers. Operations shares inventory status. Sales coordinates on lead quality. If these groups don’t communicate, the system creates local optimizations that conflict with business goals.
Successful clients usually have executive backing that forces cross-department cooperation. They invest in data infrastructure before demanding fancy analytics. They accept transformation takes months. They measure rigorously and adjust based on evidence.
Failing clients want magic. They expect AI to compensate for dysfunction. They demand instant results from systems needing training data accumulated over time. They blame tools when problems come from their own processes.
I’ve gotten selective about who I work with. Technical capability means nothing if organizational context blocks it. Some prospects aren’t ready. Telling them honestly wastes less time than pretending.
Last question. If you could change one thing about how marketing works, what would it be?
I’d want the industry to adopt basic scientific standards. What we have now is collective delusion that opinion equals knowledge. People assert confidently that certain approaches work without evidence. Assertions get repeated until everyone accepts them as fact. Practices persist for years without anyone testing whether they produce results.
Scientific fields have mechanisms challenging weak claims. Peer review, replication, significance testing, negative results. Marketing has almost none of this. Case studies showcase wins and hide losses. Vendor research sells products. Conference talks optimize for inspiration over accuracy.
If the industry adopted basic scientific standards, most current best practices would fail scrutiny. What survived would have solid foundations. Resources wasted on ineffective tactics would shift to proven approaches. Charlatans prospering on telling people what they want to hear would lose credibility.
I’m not optimistic it’ll happen. Empiricism means admitting uncertainty. Marketing culture rewards confidence. People prefer gurus who speak with conviction over researchers who acknowledge limits. Incentives favor performance over accuracy.
But at margins, evidence-based practitioners beat intuition-based competitors. Can’t fix the industry. Can build businesses that deliver results while others chase trends that never materialize. That’s enough for me.
Alisa Aidarova leads marketing at Itflamingo digital agency. She specializes in performance automation for high-budget international projects, with documented results across North American and European markets. Her peer-reviewed research on machine learning in customer analytics appears in academic journals. She holds Senior Member status at the International E-Commerce and Digital Marketing Association and served on the jury for Armenia Digital Awards 2024. Her work shows that systematic methodology, applied with discipline, beats the creative intuition marketing culture celebrates but rarely bothers to validate.
