I build operating models that make online growth repeatable
Building a team is the first step. What lasts is the operating model behind it: who owns which number, how work moves between commercial, supply chain, marketing and IT, which parts run automatically, and how the rest of the company learns to use the data. Over sixteen years I have built that model in telecom, retail banking, consumer electronics and professional tools, and each time it showed in the numbers.
At APEX Tool Group the Amazon Business channel grew from about EUR 3.6M to over EUR 5M in one year. At iSTYLE online sales grew 45% in a year across nine CEE markets. At OTP Bank Croatia digital cash-loan disbursements went from EUR 2M to EUR 3.8M a month within a year. In my current role conversion is up 40%, and AI-assisted workflows added capacity without adding headcount.
This role has the two halves I have always run together. One is the storefront: catalog quality, promotions, launches in several markets at once. The other is measurement: GA4, BigQuery, consent, and revenue in the report that matches the revenue in the ERP. On the data side I set the standard and teach the organisation to use it; I am not the person writing production code.
Today I lead the 20-person growth, content and data unit I built in 14 months, with five direct reports and a HUF 500M annual media budget. I have had direct reports without a break since 2021. I work in English every day; my German is intermediate.
Selected results
Baseline, result and time frame
€3.6M → €5M+ in 2023
Amazon Business channel revenue, +40%, then +20% in 2024
APEX Tool Group · professional hand tools · B2B buyers in Europe and Australia · 5 direct reports, full P&L
+45% in 12 months (2022)
Online sales across 9 CEE markets
iSTYLE · Apple Premium Reseller · consumers · 5 direct reports, EUR 1M performance budget
€2M → €3.8M a month, in 12 months
Digital cash-loan disbursements, OTP Bank Croatia, +90%
Retail banking · cash loans · consumers · first agile squad in the group
+40% since 06/2025
Conversion after a behavioural-science funnel rebuild
Pos Partner Payment / OFSZ · payment terminals and banking · 20 people, HUF 500M media budget
How I work
From a team to an operating model the organisation can run
1 · Build the team
At Pos Partner I built a 20-person unit across growth, content and data in 14 months. Before that the work sat with agencies and nobody owned the outcome.
2 · Design the operating model
One owner and one target per number, agreed before launch, and clean hand-overs. At APEX fixing the hand-overs between commercial, supply chain and marketing drove most of the +40%.
3 · Automate and optimise
AI-assisted content and campaign workflows at Pos Partner; triggered messaging on Salesforce Marketing Cloud in nine markets at iSTYLE; funnels rebuilt on behavioural science.
4 · Teach the organisation
OTP’s first agile squad served as the pilot for the group’s transformation. At Magyar Telekom I ran design-thinking projects and the Deutsche Telekom eTransformation KPI set.
Payment terminals for merchants (Pos Partner)Banking (OFSZ)20 people: growth, content, data5 direct reportsHUF 500M media budget
Vision
One in-house growth team that owns the result from brief to revenue, instead of work spread across agencies with no owner.
Organisation
Built the 20-person unit across growth, content and data in 14 months, with five team leads reporting to me.
Operating model
One delivery model for growth, content and data. Every campaign gets a target metric before launch and reports into one dashboard; every new brand gets a go-to-market plan.
Automation
Content production and campaign management moved onto AI-assisted workflows; product and funnel design rebuilt on behavioural science.
Impact
Conversion up 40%; more content and campaign capacity without adding headcount.
Professional hand toolsCrescent, SATA, GearWrench, WellerB2B on Amazon BusinessEurope and AustraliaFull P&L5 direct reports
Vision
Amazon Business as a managed channel for the group’s tool brands, not a side listing.
Operating model
Content, pricing, availability and forecasting run as one storefront, with one KPI set across Europe and Australia.
Organisation
Commercial, supply chain and marketing working to the same numbers, with fixed hand-overs. Crescent and Weller moved from patchy listings to a managed catalog. Most of the growth came from here.
Impact
From about EUR 3.6M to over EUR 5M in 2023 (+40%), then +20% in 2024.
Three-month project: a three-year eCommerce roadmap, the operating KPI set and the digital sales structure behind it.
International Online Sales Manager · OTP Bank, International Retail Directorate09/2018 – 09/2021
Cash loans, mortgages and credit cards for consumers in 9 countries; online sales grew 20% a year group-wide.
Opened online sales from scratch in Albania, Montenegro and Moldova.
Agreed one measurement standard with the subsidiaries, from first visit to disbursement.
Also Product Owner of the Digital Marketing Squad at OTP Bank Croatia (09/2020 – 09/2021): set up the group’s first agile squad, the pilot for its agile transformation, and took digital cash-loan disbursements from EUR 2M to EUR 3.8M a month (+90%). 2 direct reports.
Magyar Telekom (Deutsche Telekom group) · four roles09/2010 – 09/2018
Team Lead, Onsite Campaign Management (2017–2018): team of 6 on telekom.hu campaigns, landing pages and order flows; online conversion +15% year on year; owned the CMS and its vendors.
Omnichannel Expert (2016–2017): led the Magyar Telekom side of the Deutsche Telekom eTransformation programme and its innovation KPI set; introduced personalisation in the web and mobile shop.
eCommerce Senior Manager (2013–2015): wrote the eCommerce strategy and omnichannel vision; rebuilt the sales and CRM data architecture so customer data worked the same online and in store.
Marketing Project Manager (2010–2012): marketing roadmap planning and decision packs for senior management.
Skills, recent hands-on work, education
Shop management: multi-market storefronts, catalog and feed quality, promotions and bundles, launches across markets.
Measurement: GA4, Google Tag Manager, Consent Mode, Google Ads and Merchant Center, profit-based (POAS) reporting.
Data: CRM and sales data architecture, Salesforce Marketing Cloud, BigQuery, KPI and OKR frameworks.
Conversion: funnel and cohort analysis, A/B testing, behavioural science in funnel design, personalisation.
Recent hands-on work: measurement audits for D2C shops; POAS reporting that joins Google Ads spend with product cost; data pipelines in TypeScript and Postgres; a B2B product portal on a headless CMS (Payload, Next.js) in eight languages.
Languages: Hungarian (native), English (business fluent), German (intermediate).
Interests: judo, chess, squash, sailing, reading.
Part two · woom.com seen from outside
Five things I would look at first
To
Hiring team, eCommerce, woom GmbH
Date
17 November 2026 (observations 16–17 November)
Scope
Measurement, catalog and speed, public pages only
I looked at woom.com the way a new team lead might in the week before starting: nothing bought, no consent given, no internal data. Everything below is an observation with a date, so I may be missing context your team already has. I wrote it to show how I would approach the job, not to grade the work already done.
6 of 24storefronts in hreflang return a 503 maintenance page, on both days I checked
8ad-tech and marketing vendors load on en_US before any consent choice (16 and 17 Nov, Hungarian IP); none on the EU and UK stores I tested
6.7 MBof video on the de_AT homepage in a mobile lab run, out of 10.4 MB in total
#
Observation
Weight
Likely owner
1
Six storefronts answer 503 but stay in the hreflang set of the live pages
medium
Shop management, SEO
2
US store: marketing and ad tags load before a choice while Consent Mode reports “denied”
check first
Data & analytics
3
An A/B-testing campaign sent GA4 events past the tagging endpoint (seen on one of two days)
to verify
Data & analytics
4
Catalog and translation drift between markets
medium
Shop management
5
Mobile speed may be held back by page weight and main-thread work
to verify
Product experience
Already in good shape
Consent on the three EU and UK stores I tested (de_AT, de_DE, en_GB). Consent Mode v2 defaults to denied, and “Nur erforderliche Cookies erlauben” is on the first layer in the same button style as “Alle Cookies erlauben”. Before a choice no ad or marketing vendor loads. Besides woom’s own hosts, requests go to Shopify, Google Tag Manager, the consent tool, jsDelivr (Algolia’s insights library) and, on en_GB, Algolia search; a Reviews.co.uk cookie is also set.
A first-party tagging endpoint. GTM and GA4 run through ts2strack.woom.com (consistent with server-side GTM or Google tag gateway), and the events I inspected on the own stream to prod.track.woom.com carry an event_id, a useful precondition for conversion-API deduplication.
The main image is already prioritised. On the homepage and the GO 3 page Lighthouse finds the LCP image in the initial HTML, not lazy-loaded, with fetchpriority=high and a responsive srcset.
Structured product data. All nine live GO 3 pages I checked serve Product schema with SKU, GTIN-13 and ratings; product images come as AVIF through Cloudflare image resizing.
1 Six storefronts return a 503 maintenance page but are still in hreflang medium
Homepage HTTP status of each storefront in the GO 3 page’s hreflang set · 16 Nov 19:57–20:03 UTC (3 checks) and 17 Nov 10:06 UTC
en_GB200
sv_SE200
da_DK200
en_INT200
en_BE200
en_CZ503
en_ES503
en_HU503
en_PL200
en_RO503
en_SK503
nl_NL200
nl_BE200
de_AT200
en_US200
fr_CH200
de_CH200
de_DE200
de_BE200
fr_FR200
fr_BE200
fi_FI200
pl_PL200
es_ES503
The nine live GO 3 pages I checked still list all six in their hreflang set. Search engines drop hreflang pairs they cannot fetch, so these markets currently get no language targeting. The maintenance text is German on English and Spanish URLs. From outside I cannot tell whether the markets are paused or down.
Next step: if a market is closed for good, remove it from hreflang and redirect to en_INT. If it is temporary, keep the 503 with a Retry-After header, use a local-language maintenance text, and add a per-locale uptime check.
2 On the US store, marketing and ad tags load before a choice while Consent Mode reports “denied” check first
Fresh browser profile, homepage, no click on the banner · 16 and 17 Nov 2026, same result · Hungarian IP (Cloudflare set cf_country=HU)
0ad-tech or marketing vendors on de_AT, de_DE, en_GB before a choice12–13 hosts: woom, Shopify, GTM, the consent tool, jsDelivr, Algolia
8ad-tech or marketing vendors on en_US before a choiceAd platforms: Meta Pixel, StackAdapt, Simpli.fi, MNTN. Also Klaviyo, AB Tasty, Friendbuy (fbot) and cdn.avmws.com · 32–33 hosts · 18–19 cookies, the AB Tasty one for 395 days
At the same moment GA4 is sent with gcs=G100, so analytics storage is denied. An opt-out model may well be the intended choice for US residents; I cannot see that decision from outside. Three things seem worth checking:
Attribution. GA4 sees the US visitor without cookies (and possibly modelled), while the ad platforms see the same visitor without that restriction. Comparing platform numbers against GA4 can then overstate paid channels.
Visitors from outside the US. My session came from a Hungarian IP and got the same behaviour on en_US. Whether that needs a region rule is a question for Legal; I am not drawing a conclusion here.
A first-party identifier cookie.sgtm_fpid, set by the tagging endpoint, was present before a choice on en_GB (16 Nov) and en_US (17 Nov) while gcs=G100. It is worth checking whether that identifier follows the consent state.
Next step: one written consent matrix by storefront and visitor region, agreed with Legal and applied in GTM.
3 An A/B-testing campaign sent GA4 events past the tagging endpoint to verify
en_US · seen on the first load and the reload on 16 Nov; not seen on 17 Nov, when that campaign was no longer loaded
On en_US, G-PR2NZB22GE got events through ts2strack, while G-PQR6NFGWNG (the property the EU and UK stores use, so probably the global one) also got a hit straight from the browser. In the same session the dataLayer held an AB Tasty event for campaign “Pre-BFCM GWP Banner – Mini Cart”, sent once with send_to each property. My reading is that the testing tool’s GA4 integration, not a GTM tag, opened the direct path, and it went away with the campaign. Both hits had gcs=G100. A direct path is more exposed to ad blockers and skips anything the endpoint adds.
Next step: check the GA4 targets in the AB Tasty integration, so test events follow the same route as everything else.
4 Catalog and translation drift between markets medium
woom GO 3 product page in eleven markets (nine live): visible page and JSON-LD · 16 Nov 2026
Market
What the shopper or Google sees
Probably meant
fr_FR
A price line reads “éteinte 450,90 € / au lieu du prix 498,90 €”. Éteinte means “switched off”.
“à partir de 450,90 €”
sv_SE
The same price line reads “av 5 918,00 kr / istället pris”. Av means “of” or “off”.
“från 5 918,00 kr”
en_GB
Schema name “woom GO 3 EU (Gen. H)”, which looks like an internal model code
Customer-facing name
en_GB
Schema says LimitedAvailability, the page says “Delivery 5–7 working days”; may be intended
A deliberate stock signal
pl_PL, nl_NL
“woom GO 3 - Kids’ bike - 16 cali” and “Kinderfiets - 16 inches”, partly English
“Rower dziecięcy”, “16 inch”
Next step: a nightly catalog check comparing name, price, availability and schema per SKU across locales. The same job can watch the storefront status from finding 1.
5 Mobile speed: page weight and main-thread work, not the image setup to verify
Lighthouse 12.8 mobile lab, simulated slow 4G and 4× CPU slowdown, one run per page, so noisy · 16 Nov 2026 · I did not pull CrUX field data
video 6.7 MBother 1.3 MBJavaScript 1.0 MBimages 0.9 MBHTML 0.3 MBCSS and fonts 0.2 MB
Homepage. Two hero videos download, 3.6 MB and 3.1 MB, one portrait and one landscape, both with “EN” in the file name on the German page. They are not the LCP element, but they compete for the same connection. Of the simulated 18.9 s LCP, 59% is render delay after the image has arrived: the main thread is busy.
Product page (score 30, simulated LCP 20.0 s). 776 KB of HTML (Nuxt payload) and 5,472 DOM elements. The document took 1.8 s to first byte in the actual run, which may be a CDN cache miss. In the simulation the LCP image then waits 15.5 s for the network, behind the HTML and scripts.
Tags. Because GTM is served from ts2strack.woom.com, Lighthouse counts it as first party, so “third-party” blocking looks small (the consent tool, about 250 ms, is the largest). The GTM container’s own long tasks add up to about 1.25 s on the homepage and 0.9 s on the product page.
Next step: load only the video that fits the viewport, with a poster frame; check CDN caching for product HTML; review which GTM tags must run at load; and set a payload budget per page type.
Proposal
First 90 days, if this were my desk
Vision
Every woom market shows the right product, price and stock, and every euro of revenue is counted once, in the same way, in every report.
Weeks 1–4: close the gaps
A locale health check for status, hreflang, schema and price drift (findings 1 and 4)
A consent matrix agreed with Legal and applied in GTM (finding 2)
One agreed list of live, paused and closed markets, reflected in navigation, hreflang and feeds
Weeks 5–12: one measurement truth
All GA4 traffic through the tagging endpoint; Meta CAPI deduplicated on an event_id shared by pixel and server, Google on transaction_id
A weekly GA4 versus Shopify order reconciliation by market, with a tolerance band
Performance budgets for hero media and product pages, owned with Product Experience
A short measurement playbook and a monthly numbers review with the market managers, so the standard does not depend on one person
Questions I would bring to week one
What share of Shopify orders does GA4 capture in each market, and how is it trending?
Is Meta CAPI live, and what is the deduplication rate against the browser pixel?
How is the reference price behind the “statt” prices defined for Black Friday in the EU stores?
Which of the six 503 markets are paused on purpose?
On the four stores I tested, prod.track.woom.com received events (gtm.js, cart-clear, cart_update) before a consent choice. How is that stream classified?