Acquisition and Product Discovery: Expert Insights From Dexfinity

We spoke with Pavol Adamčák, Founder and CEO of performance agency Dexfinity. We asked him about connecting paid acquisition with product discovery, localizing for new markets, and why acquisition and onsite teams need to start sharing data.

Dexfinity is a performance marketing agency with offices in Slovakia and Czechia. We focus on paid acquisition: Google Ads and Meta Ads, SEO, and AI-powered creative advertising. We have dozens of e-commerce specialists and have worked on more than a hundred projects across 34+ markets, including Germany, the US, and Singapore.

For ten consecutive years, we’ve held Google Premier Partner status, which gives us deeper access to data and betas. That directly shapes how we approach both campaign strategy and onsite measurement.

Over the years, we’ve pushed hard into full-funnel thinking. Driving qualified traffic is only half the equation. If the acquisition strategy brings in the right visitors and they then fail to find the right product on site, we’ve wasted both the client’s budget and their time.

We treat it as a closed loop rather than two separate workstreams. On the acquisition side, we segment campaigns by search intent. Branded, category-level, and specific product queries each carry different user expectations. A user who searched for a specific supplement by name expects to find it immediately. A user who searched “best protein for muscle gain” needs the discovery layer to do the heavy lifting.

We cross-reference Google Ads query data with internal site search reports to identify exactly where that handoff breaks down. Fragmented campaign structures prevent machine learning from seeing enough conversion signals, and that same fragmentation masks where users fail to find the right products.

With Powerlogy, fixing the acquisition structure and onsite tracking simultaneously led to a 170% increase in shopping campaign revenue over five months and a 35% reduction in Google Ads costs. For Northfinder, we went a step further and implemented hourly product feed updates rather than the standard daily cycle so that inventory and pricing data reaching the discovery layer were always current. Combined with campaign restructuring, that contributed to 62% revenue growth year-on-year and a 30% reduction in cost-to-revenue ratio.

Three metrics tell us most of what we need to know.

First, internal search usage rate: if a large share of sessions include a site search event, navigation alone is failing to help users find products. There’s a discovery gap that needs closing.

Second, internal search exit rate. Users who search on the site and then leave are telling you explicitly that the results weren’t relevant enough to stay.

Third, the conversion rate gap between sessions with and without internal search interaction. On catalogs with more than five thousand SKUs, we almost always see a meaningful lift in sessions where users engaged with site search. This confirms the purchase intent is there. The discovery engine just needs to be good enough to capture it.

We saw this clearly with BioTechUSA. Once we restructured their campaigns to generate cleaner acquisition data per market, we could identify which category landing experiences were underperforming in Romania versus Slovakia. The gap was largely a discovery problem, not a traffic quality problem. Once both sides were addressed, Slovakia grew 40% year-on-year, and Romania reached 93% year-on-year growth.

This is one of the most common tensions we see with clients, especially during seasonal peaks or clearance cycles. Our approach is to treat relevance as the foundation and business rules as a modifier. Never as a full override.

When commercial pressure completely replaces relevance logic, users start distrusting the results. They notice they keep seeing the same pushed products regardless of what they searched for, and they stop engaging altogether. What works much better is a soft boost model: if a high-margin or overstocked product is already reasonably relevant for a user’s intent, a small boost can surface it into top results without corrupting the experience.

With Foxigy, we learned that clearance and promotional mechanics performed very differently depending on the market. Clearance sales drove strong results in Romania, while they had almost no effect in Hungary, where dynamic campaign formats worked far better. That same market-level nuance has to be built into discovery rules, not just acquisition campaigns. The right setup gives you precise controls per market and per segment, not one blunt override across the entire catalog.

For clients starting their personalization journey, we focus on three layers before anything else.

First, query understanding. The discovery engine needs to correctly interpret synonyms, common misspellings, and category intent from the raw query. If someone types “trainers” and gets no results because the catalog uses “sneakers,” no amount of machine learning will fix the experience downstream.

Second, in-session behavioral signals. What a user clicks in the first two to three interactions tells you a lot about price range preference, brand affinity, or product type. A good discovery engine uses those signals in real time, not only in long-term profile models.

Third, popularity and trending data are the default. When you don’t know enough about a specific user, surfacing trending and bestselling items is clearly better than alphabetical or margin-sorted results.

These three layers alone produce measurable conversion improvements for e-shops in the ten to one hundred thousand SKU range, and they create a clean foundation before you layer in more sophisticated personalization.

With Irisimo, a watch retailer we helped grow from €400,000 to €4 million in annual revenue while expanding to 14 European markets, clean and structured product data was the non-negotiable starting point. Slovak and Czech buyers search for different brands than those in Romania or Bulgaria. For campaigns in each new market to perform, the product catalog had to reflect what local buyers actually searched for. Without that foundation, neither acquisition campaigns nor onsite discovery can surface relevant results. You end up paying to bring the right people to a catalog that doesn’t speak to their market.

Language is the obvious answer, but most teams dramatically underestimate how deep localization needs to go. Translating product names and categories is just the surface. The real challenge is search query behavior. How Slovak users phrase a search is meaningfully different from that of Czech users, and both differ from Romanian, Hungarian, or Bulgarian shoppers.

When we helped Northfinder recover their Romanian expansion, the market moved from underperforming to delivering 160% growth and a 55% reduction in acquisition cost. A big part of that was aligning what users searched for locally with what the campaigns and onsite experience served back to them.

Before going live in any new market, we run a query audit: what are users actually typing, and is the discovery engine returning relevant results for those local phrasings? We’ve seen cases where a large share of site searches in a new market returned zero results simply because query understanding was built on the home market’s language patterns. With Foxigy, the approach for each new market had to be built from scratch.

Search campaigns worked in Bulgaria, dynamic formats in Hungary, promotional mechanics in Romania, and Slovenia required its own calibration.

The second element to localize is the popularity signals. Ranking built on global data will surface the home market’s bestsellers, which often have no overlap with local demand. Popularity scores need to be calculated per market from day one, even when the initial dataset is thin.

One pattern we keep seeing, regardless of market or client size, is the silo between acquisition and onsite teams. They operate with separate KPIs and no shared reporting, and nobody formally owns the handoff between them. The traffic team delivers volume and cost-per-click. The site team delivers a conversion rate.

Working closely with a product discovery layer like Luigi’s Box made us realize how much signal flows in both directions. Site search queries returning poor results are literally keywords worth bidding on in acquisition. Users are already showing purchase intent on your site; you just need to meet it better on both sides of the click. Eyerim covered 12 markets with six specialists by treating product feed data, acquisition structure, and onsite discovery as one connected system. When those sides share data, the whole system outperforms without simply raising bids.

Most teams run acquisition and discovery as two separate operations and never find out how much they’re losing at the handoff. The signal is already there on both sides. Teams that connect them stop paying for traffic that the site then loses.