Feedback Is a Direction, Not an Instruction: A Conversation with Julia Salanko

Julia Salanko has run hundreds of A/B tests on a product with more than 15 million users. Her most useful lesson from all of them is that you should never take your users entirely at their word.

She's a growth and experimentation specialist at Nord Security, optimizing the funnel behind NordVPN. Before that she was one of the early team members at Glovo in Poland, where she helped launch the market from scratch and scaled partner acquisition from zero to 60,000 leads. She's fascinated by psychology and decision-making, mentors the next generation of marketers as a team lead at Turing College, and holds a clear philosophy: sustainable growth doesn't come from hacks or playbooks.

Feedback is a direction, not an instruction

Julia's first and sharpest point is to take user feedback with a grain of salt. What people tell you in an interview is a hypothetical, and real decisions are primarily emotional. She's watched it play out directly: a few years ago her team explored an idea, ran extensive qualitative research, user testing, interviews, and every signal said the idea had huge potential and users loved it. Then they built it, ran the A/B test, and the results were the complete opposite.

So she treats conversations with users as a direction to explore what problems they have, not a recommendation for what to build. It's the Henry Ford "faster horse" problem: people describe what they know, not what's possible, so you use their input to understand the destination and your own expertise to design the vehicle.

The illusions founders hold about A/B testing

The main illusion, and it's not just founders but managers too, is that A/B testing is a magic tool that delivers immediate answers to every question. It isn't. It's a tool that brings some color and some answers, but it can't replace strategy, vision, or the knowledge different teams hold. The second illusion is that tests return clean yes/no verdicts. In reality most experiments give you some certainty and plenty of uncertainty at once: some segments react better, some worse, and false positives and false negatives exist no matter how good your systems are. What testing really does is eliminate some risk, not hand you the truth.

What changes when you optimize at scale

Optimizing a product with millions of users is a different craft than a young startup. Early-stage products often lack product-market fit, which makes optimization essentially impossible, so your effort should go into finding the right angle or rebuilding, with A/B testing used as research rather than a way to nudge metrics. On an established product, small changes here and there produce huge impact because of the sheer volume of users and transactions. Small products, by contrast, need bolder, bigger bets, because traffic and revenue are lower and you're chasing higher returns at higher risk. At scale, growth is a hunt for incremental gains everywhere in the funnel, not one magic lever.

The lever almost nobody watches

Asked what's underused, Julia points past acquisition. Teams, especially at bigger companies, over-focus on the first stage, getting users in, and under-think what happens next: the product experience, activation, retention. Her take is to hold growth and product holistically rather than treating it as only acquisition or only retention, because what happens after the click is where the durable value lives.

How to stop testing random things

Running experiments just to feel productive is a real failure mode, and it happens on any team, especially when someone is attached to a favorite idea they're sure will work. Julia's guard against it is evidence-based hypotheses. Before a test, answer the "why," then go find data that supports the idea, internal analytics, competitor research, studies from other companies, and prioritize clearly. If you pour effort into small tweaks with no proof they'll work, you're burning time, money, and resources that a higher-impact idea in the backlog deserved. Testing five shades of red is pointless; reworking a whole flow, which can both help and reduce risk, is worth it.

What breaks when growth starts working

Growth breaking things is predictable, she says, the way you know a certain car will fail at a certain mileage. When volumes rise, internal systems and tools that weren't built for that load break first. Messy data becomes chaotic data at higher volume. Then the product itself can start limiting growth, because a new segment or partner demand needs development, and development takes time an early-stage company can't always afford to wait for. Her Glovo lesson was to accept it'll be messy, sometimes forever, optimize what can be optimized, and keep focus on the big picture instead of fixing minor things. Done is better than perfect, because you're never 100% ready anyway. You launch, learn, and fix on the go.

Launch from zero, or optimize at scale?

Both are hard in different ways. Launching means acting under huge uncertainty, aligning a team where everyone has a different opinion, and proving a strategy you can't yet be sure of. Optimizing a mature product means watching your gains shrink over time until they're nearly invisible, at which point you face a choice: enjoy the stability, or effectively return to step zero with drastic structural changes. That second path has a high chance of failure, but where continuous iteration buys you 1% gains, a bold structural bet might, one time in ten, buy you 20%.

AI lowered the barrier, so what's the moat?

AI has made parts of the job easier, data analysis, competitor and market research, even surfacing experiment hypotheses she hadn't considered, but it can't replace whole functions. And it doesn't change the fundamental question. If everyone can build the same thing, the moat is still whether people will use your product and whether they'll pay for it. Most products fail not for lack of tooling but because they target non-existential needs, or problems that exist but rank low in people's priorities. As she learned from her own psychologist, people adapt to many of their problems and simply never fix them. AI won't fix that for you.

What young marketers get wrong entering tech

Teaching at Turing College, Julia sees marketers from traditional backgrounds over-index on promotion, ads, messaging, discounts, brand communication, without deeply understanding the product, the customers, and the friction and anxiety they feel using it. Marketing in tech is always tied to the product. You don't just count conversions, you estimate how each action affects lifetime value across the whole customer journey. Her other observation is the lingering belief that you need a formal degree, which she's living proof against, because the field changes so fast that short, current, two-to-four-month programs beat long ones.

Key takeaways

Feedback is a direction, not an instruction. What users say is hypothetical; what they do is emotional. Use interviews to find problems, not to spec the product.

A/B testing eliminates risk, it doesn't hand you truth. It's a tool, not a magic oracle, and it can't replace strategy or vision.

Small changes scale, bold changes launch. At millions of users, tiny tweaks compound; early products need bigger, riskier bets.

Watch what happens after acquisition. The underused lever is product experience and retention. Hold growth holistically.

The moat isn't the build, it's the need. AI lets anyone build. Products still fail when they target non-existential or low-priority problems.

Frameworks worth stealing

Evidence before experiment

Before running a test, answer why you're running it, then gather evidence that the idea might work, internal analytics, competitor research, outside studies, and prioritize against your backlog. If you can't support the hypothesis with anything but conviction, you're spending resources a higher-impact idea deserved.

Done is better than perfect, at the big-picture level

Accept that fast-scaling systems are messy, often permanently. Optimize what can be optimized, ship before you feel 100% ready, and keep your attention on the big problems rather than perfecting minor details. You'll learn more from launching and iterating than from waiting to be ready.

Quotes worth keeping

The lines I wrote down.

Take user feedback with a grain of salt. What they suggest is hypothetical. The real decision is primarily emotional.

A/B testing isn't a magic tool with immediate answers. It eliminates some risk. It doesn't replace strategy.

You can build whatever you want with AI. The question stays the same: will people use it, and will they pay for it?

Rapid fire round

Same questions every guest. Quick questions, quick answers.

Best advice you've ever received? From a former manager: do things even if you don't feel fully ready.

Advice you ignored and wish you'd listened to? Nothing comes to mind, which she figures means it wasn't relevant enough to remember.

What would you tell your younger self? Believe in yourself, everything is possible, and you can change your life and your career as many times as you want.

Ongoing challenge that keeps you up at night? A personal career transition, moving from marketing into product management and growth product.

Favorite spot? The parks near her district in Warsaw, where she runs, reads, and hopes to teach yoga classes outdoors. For food, Israeli and Middle Eastern cuisine.

Tool you can't live without? Both ChatGPT and Claude, switching depending on the question, though she leans on Claude for being faster and more universal.

---

Julia Salanko is a growth and experimentation specialist at Nord Security, working on the funnel behind NordVPN and its 15M+ users. She was an early team member at Glovo in Poland, scaling partner acquisition from zero to 60,000 leads, and mentors new marketers as a team lead at Turing College. Find her on LinkedIn.