X-Ray of Liège E-Commerce 2026

Two hosting philosophies, not two quality tiers

The two leading platforms, WooCommerce (43% of the base, 251 sites) and PrestaShop (34%, 198 sites), are open-source solutions that the retailer hosts and maintains itself: together, they cover 78% of the shops. The retailer stays responsible for its hosting, its security patches and the technical health of its shop, directly or through its provider. Shopify and Wix sit on the other side: hosted services (SaaS) where the vendor takes on the infrastructure, the server performance and the updates, in exchange for less control over the foundation, but less maintenance falling on the retailer. Combined, they account for just 10% of the base here.

This open-source autonomy handicaps nothing on the substance. On the angles that matter for acquisition, WooCommerce and PrestaShop handle natively, or through mature extensions, schema.org product markup (`Product`, `Offer`, `BreadcrumbList`), feed generation for Google Merchant Center, canonical URLs and hreflang tags, just as a Shopify does. Open-source does not mean poorly maintained: a properly kept site ticks the same technical boxes as a proprietary solution, on readability by search engines as by comparison shopping engines. The gap, when it exists, lies in the hosting and the configuration, never in the engine itself.

PrestaShop's dominance in this Liège sample is no local quirk. A French-born solution widely served by the regional agency network, PrestaShop structures the entire French-speaking e-commerce market, in France as in Belgium, where it stands as the standard. The contrast with Shopify's global weight therefore betrays no Liège lag, just the French-speaking anchoring you would expect from a base of local retailers. The panel does not reveal a gap against the norm, it makes the regional norm visible at the scale of a city.

What the platform really decides: who carries the debt

The dominant CMS is not a plumbing detail: it decides who carries the maintenance and optimization debt. Choosing a platform means choosing a mode of technical governance for the next three to five years, well beyond launch day. As long as everything runs smoothly, the distinction stays theoretical. The day you have to patch a flaw, migrate a major version, rebuild a payment funnel or clean up a pile-up of extensions, the question of who actually holds the wheel becomes concrete.

An open-source platform hands that responsibility to whoever operates it. When that operator is an SME with no dedicated technical team and no clear maintenance contract, the debt does not vanish: it builds up silently, in unpatched versions, in abandoned extensions, in performance that degrades with each new addition. That is the real subject, not the name of the CMS. The local base has positioned itself massively on the model that shifts the technical work onto SMEs that, lacking a team, do not absorb it. That is the configuration the Liège retail base makes most likely.

Speed, the only criterion where the CMS weighs measurably

On raw performance, the gap is clear on desktop, and it goes in only one direction. Hosted solutions pull ahead: Wix median at 89, Shopify at 88, ahead of WooCommerce at 82 and PrestaShop at 80, the slowest of the panel's major CMSes. On mobile, the picture tightens and Odoo brings up the rear at 53. But the mobile ranking order plays out within one to three points on small per-platform samples: at that scale, it flips from one measurement campaign to the next. No platform systematically leads on this front, and presenting mobile as a verdict would amount to over-reading statistical noise. The hierarchy can only be stated honestly on desktop.

The mechanism behind the desktop gap is structural. SaaS optimizes the infrastructure by default, with shared servers tuned to the millimeter, an application cache and a CDN shipped out of the box. Self-hosting shifts that responsibility onto the shop: it demands serious hosting (Redis or Varnish, OPcache), a configured cache and a connected CDN, three building blocks that many Liège retailers never switch on, for lack of an in-house technical profile or a provider to put them in place. A properly served PrestaShop or WooCommerce closes most of the gap: what slows things down is not the tool, it is the install left raw. The platform's brand indicates the slope, not the finish line.

This speed weighs on two fronts at once, and that combination makes it decisive. On the SEO side, Google uses Core Web Vitals (LCP, INP, CLS) as a ranking signal: a slow shop starts with a structural handicap on the same queries. On the conversion side, every hundred milliseconds of latency erodes the conversion rate, and that cost is paid on every session, across all traffic, paid as well as organic.

The hidden cost, quantified, and its flip side

The all-in-one SaaS narrative obscures the real cost of well-kept self-hosting, which the study does not measure. As an estimated order of magnitude: performant hosting, modules often on one-shot licenses of 50 to 100 euros apiece, sometimes on subscription, which quickly add up to 2 or 3 k euros at install (excluding recurring subscriptions), and maintenance dev time. On the other side, a Shopify Basic starts around 32 euros a month, but the apps you actually need often add 50 to 150 euros monthly, and transaction fees apply as long as you do not go through Shop Payments.

The flip side, to state honestly: open-source keeps a real structural advantage.

• Data ownership, no vendor lock-in.
• No Shopify Plus bill at roughly 2000 euros a month once at scale.

The right choice depends on access to a developer and on ambition, never on a logo.

The real business divide is not played out on two Lighthouse points anyway, but on the checkout funnel. Shopify ships a native Checkout (Shop Pay, one-click payment, minimal friction) optimized and maintained by the vendor, whereas WooCommerce and PrestaShop leave the retailer a hand-built funnel, often weakened after every update. That is where conversion is won or lost, not on display speed. The detail reads in the checkout funnel layer.

The practical consequence for a Liège retailer is direct: the question is not "did I choose the right platform?", but "am I exploiting the speed and the funnel my platform lets me reach?". The most profitable room for improvement almost always sits in optimizing what already exists (hosting, theme, images, third-party scripts), before considering a migration. The scores rely on Google's PageSpeed API (opens in a new window); our reading of the technical work is detailed on the expertises side.

For a decision-maker, the bottom line: your pages drag exactly where it matters, on mobile, and every marketing tool you add deepens the hole instead of filling it. Slowness is paid for twice, in search rankings and in sales that never happen. The technical detail that follows is for your provider.

The mobile divide

The median mobile score sits at 60 out of 100, against 83 on desktop. Twenty-three points of gap, and they land right where most retail visits happen, on the smartphone. That 60 is a lab Lighthouse score, a simulation run on a deliberately throttled network and processor to reproduce a mid-range mobile. It says how the page behaves in a standardized test bench without reflecting what each visitor actually experiences, and that is precisely what makes it comparable from one site to another across the base. Not to be confused with field Core Web Vitals, collected from real users via the CrUX base, which here covers only 186 sites in the sample, those receiving enough traffic to feed a reliable measurement. The two families tell the same story from two angles, the test bench and real life, and it is their convergence that grounds the diagnosis, more than any single figure taken alone. One nuance matters for search rankings: the lab Lighthouse score diagnoses but does not weigh directly into Google ranking; only CrUX field data feeds the Core Web Vitals used as an SEO signal.

The mobile distribution leans low. Just 3.7% of e-shops are in the green, against 32.9% on desktop, and 23.1% squarely in the red, that is 132 sites under 50. A site judged decent on a computer therefore collapses the moment you take it onto a phone. The base optimized for the big screen and let the small one slip.

The breakdown of the indicators hardens the finding without making it hopeless. LCP, the time it takes the largest visible element (often the product photo) to display, exceeds 4 seconds on 64% of sites, when Google recommends staying under 2.5 seconds, a threshold met by only 21% of the base. Between the two stretches a grey zone where the most accessible gains sit. Visual stability, on the other hand, is healthy: CLS, which measures layout shifts during loading (a button that moves just as you tap it, an image that pushes text down), stays good on 81% of sites. The design holds up, it is the technical weight that drags things down, and that weight is fixed exactly where a layout redesign would cost far more.

The more you equip, the more you slow down

Ranking e-shops by the number of third-party tools detected (analytics tags, advertising pixels, chat widgets, consent banners, A/B testing scripts, remote fonts), the trend contradicts the intuition that "better equipped" rhymes with "better built": the sites most loaded with scripts are among the slowest. The mobile median declines monotonically along the gradient, with no reversal from one tier to the next, from 63 for sites with no tools at all to 48 for those carrying a full stack. Fifteen points. The slope also holds on desktop (from 87 to 70), at constant platform and across all districts, from Liège to Waremme. Whatever angle you slice the sample from, the direction never reverses, and that is what distinguishes a signal from a sampling artifact.

The link between third-party overload and slowness is no mystery. Each script adds, in varying proportions:

• network: a DNS request, a TLS handshake and a round trip to a domain that is not yours, therefore outside your cache and your CDN. Multiplied by twenty third-party origins, the connection cost alone weighs down the start.
• a blocked main thread: most of these scripts run in JavaScript on the main thread, the same thread that has to paint the page and respond to clicks. It is the classic profile of a Total Blocking Time that swells as the tags pile up, often without any single one being individually to blame. This lab-measured Total Blocking Time foreshadows the INP that real users will experience, the interactivity signal Google now uses in the Core Web Vitals.
• the unpredictable: a third party, by definition, is beyond your control. Its latency, its size, its outage of the day become yours.

Correlation, not causation

The direction of cause and effect calls for caution. Claiming "third-party scripts destroy performance, plain and simple" would be dishonest: the most loaded tier rests on n=8 sites only, too thin a count to make it a law. The correlation can run both ways. Heavy scripts slow the page, but it is also the most commercially mature sites, therefore the most instrumented, that pile on the most tools without necessarily having invested in parallel in performance discipline. Tooling here is a marker as much as a cause. We keep the correlation, we do not oversell the causation on so thin a count.

One last safeguard. The study measured no conversion, no cart, no revenue. When you read that "an extra second of loading erodes the conversion rate," that is an external benchmark (the Google and Deloitte orders of magnitude on mobile retail), not a result from this Liège corpus: we connect an observed slowness to a risk documented elsewhere, plausible and to be verified site by site. The nuance separates what the data proves (a base that is often slow and heavily instrumented) from what it suggests by analogy (a commercial shortfall yet to be confirmed).

The real fix: the right place, not fewer tools

The right reflex, then, is not to remove the tools, but to load them cleanly (see tracking). The tracking-versus-performance dilemma is largely solved in 2026 by server-side tagging. With an sGTM, a tag server that routes the measurement through your own server rather than through the browser (via a service like Stape or Addingwell), you pull GA4, the Meta Pixel and Google Ads out of the visitor's browser: the client gets lighter, and first-party cookies gain lifespan. Two settings remain on the loading side, deferred loading of non-critical scripts and Consent Mode v2, which conditions firing on consent instead of sending everything off at the first byte. A badly sequenced tag blocks the main thread at the worst moment, right when the browser needs to paint the page.

The order of magnitude is documented in an external benchmark, not measured here: taking a mobile LCP from 4 seconds to 2.5 seconds translates, according to Google benchmarks, into a double-digit conversion gain. More reliable measurement, faster page, compliance maintained: the three goals stop pulling against each other.

The measurements rely on the Google PageSpeed API (opens in a new window), which exposes in a single call both the lab side (Lighthouse) and the field side (CrUX). Our reading of the technical work, from tag sequencing to lightening the critical rendering path, is detailed on the expertise side.

A flattering score by absence of subject

Measured on the homepage, consent compliance splits into four grades. 41% of sites earn a grade A (235 sites), while 44% fall into the weak zone (grade C or D). The raw number is misleading. Of those 235 grade A scores, 215 show no banner because they load no tracker to consent to: a good grade by absence of subject, not by control. Only twenty sites truly earn their A by managing a proper banner over active trackers.

The nuance matters, because the risk reads in the crossover between active trackers and the absence of a consent mechanism, not in the displayed grade. A site with no tags whatsoever has, by construction, neither GDPR debt nor measurement debt: it collects nothing, so it drops nothing without a legal basis. The point of exposure is the page that loads its trackers (Google Analytics, Meta Pixel, third-party tools) before any choice by the user, and without a CMP to capture it. That precise profile is the one to isolate before talking about bringing anything into compliance, rather than the overall volume of weak grades.

Consent isn't fixed inside the CMS

Consent is not a WooCommerce or PrestaShop feature. It's a dedicated tool, a CMP (consent management platform, such as Cookiebot, Axeptio, Complianz), wired into the tag manager (GTM) and paired with Google's Consent Mode v2. You add it whatever the platform, and the real work happens in the wiring between the CMP, the tag manager and each tracker, far more than in dropping the widget.

Installing a banner via a CMS module or an all-in-one plugin handles the display, not the compliance. A banner that drops trackers before the click, or that doesn't make refusal as easy as acceptance, stays non-compliant with the GDPR and the ePrivacy directive, no matter how serious the provider who set it up. A CMP only becomes an asset if it genuinely blocks tags from firing until consent is given. Layered over a GTM that fires unconditionally, it decorates the page without governing anything. The proof is in the sample: 97 sites equipped with a CMP stay at grade C or D, for lack of real blocking before choice.

No SaaS makes a site compliant by its mere presence. Dropping a OneTrust, Axeptio or Cookiebot banner puts no one in order on its own: the tool is only worth the wiring behind it, namely honest categorization of trackers, effective blocking before consent, and clean transmission of the signal to the tags. Compliance is verified in the network, request by request, never by the presence of a vendor logo in the footer. The legal framework is detailed by the CNIL (opens in a new window).

Consent Mode v2: only advanced mode unlocks modeling

Google's Consent Mode v2 comes in two forms that need distinguishing. In basic mode, Google tags stay blocked as long as refusal holds: no signal goes out, and refused conversions are lost. In advanced mode, tags load from the start but stay throttled without consent, and send anonymous pings (`consent denied`) that let Google model the missing conversions, thereby reconstructing part of what refusal cut off.

It's advanced mode, and only it, that unlocks this modeling. It's also subject to volume thresholds (traffic and conversions per country) below which Google reconstructs nothing: on a small Liège e-commerce account, those thresholds are often out of reach, which makes every conversion actually observed all the more valuable. Across the sample, only 35 sites run a Consent Mode v2 in advanced mode: the only ones technically equipped to recover part of the conversions lost to refusal.

The figure of 83% of sites without Consent Mode calls for an honest reading. It includes 221 sites with no Google tag to steer (no Google Ads, no GA4 connected), for which the absence of Consent Mode is moot. The truly actionable risk, active trackers with no consent-capture mechanism, forms a far narrower subset that needs isolating before drawing a compliance volume from it. The nuance changes prioritization: a site with no tags at all doesn't have a tracking problem but a presence problem, whereas a site spending on Ads without a consent framework burns budget with every passing day.

Beyond Google: Meta and the server layer

Consent Mode v2 only steers Google tags. The Meta Pixel, present on 101 sites and often dropped before even a solid GA4 base, keeps its own blind spot. Without a consent signal passed along and without a Conversions API (CAPI) server-side with an `event_id` to deduplicate, the match quality of events drops (EMQ, the score that measures how well Meta recognizes the user behind a conversion), and Meta doesn't model conversions lost to refusal either. Social budget then optimizes on holed data, exactly like Google budget without Consent Mode. Two ad platforms, one same reflex: make consent speak to each platform.

The CMP plus client-side Consent Mode stays the legal minimum: it signals the choice to the tags without deciding what goes out. The mature answer, the one that truly links compliance and measurement quality, is server-side tagging. A consent-aware sGTM only forwards the event to Google or Meta if consent allows, and the CAPI sent in parallel is deduplicated by `event_id` so the same conversion isn't counted twice. With 83% of sites without Consent Mode, the step is steep, but it's the server that holds the data at scale, where the browser gives out. This work follows directly from the measurement layer: a single tag flow decides both compliance and conversion reliability.

For a decision-maker, the essentials first. Half of Liège's stores have no idea where their sales come from, and some of those running ads pay against false numbers. Measuring accurately costs little and changes every decision that follows. The detail below names the tools, to read with your provider.

Google Analytics 4, the reference audience-measurement tool, shows up on only 293 sites, or 50.6% of the base. Nearly one retailer in two therefore sells without knowing where its visitors come from, on which pages they drop off, or at which step they abandon their cart. Half the base ignores its own conversion funnel. At the other end of the spectrum, 196 e-shops (33.9%) carry no measurement tool at all: no analytics, no advertising pixel, no conversion tag. They run on guesswork, on the sole basis of total revenue, never knowing which channel produces it.

Alongside, 15.2% already place a Google Ads tag to steer their spend on real acquisition cost. But it pays to qualify what « placing a tag » actually guarantees, because its presence says nothing about the quality of the measurement. A Google Ads tag loaded without conversion properly wired, or a GA4 placed without an e-commerce event configured, counts visits but stays blind to the transaction, hence useless for steering a media budget.

Instrumentation that is rare and rickety

The first surprise of the crawl lies less in the volume of tools than in their nature. The most frequent marker remains the Meta Pixel, placed in pure client-side, with no server declaration to match. Yet a pixel alone, in 2026, is no longer a reliable measurement: between Apple's tracking prevention (ITP, which caps at seven days the lifespan of first-party cookies set client-side in JavaScript, while those set server-side escape it), ad blockers and the tightening of iOS, it lets slip, depending on the share of Safari/iOS traffic and of blockers, 20 to 40% of events before they even reach the ad network's servers. The order of magnitude is heavy: one to two purchases in five vanish from the measurement.

The issue, then, is not the load order, since the pixel often fires before the analytics; it lies in what makes advertising instrumentation credible and what is missing here: the Conversions API (CAPI) to mirror collection server-side, and above all an `event_id` shared between browser and server to deduplicate. The CAPI sends the event back from the server, out of reach of the blocker and of ITP; the shared `event_id` lets the platform recognise that a purchase seen twice, browser-side and server-side, counts only once. Without this pair, you choose between two evils: under-counting (pixel alone) or double-counting (both without deduplication). The server layer forms the foundation, catching what the browser loses, provided it is wired cleanly.

The hierarchy of the stack speaks louder than any figure on its own

• Universal Analytics still present: 101 sites (17.4%) load this version, shut down by Google in July 2023 (opens in a new window). It has collected nothing for nearly three years, but no one has removed it, because no one looks at the data. It is the marker of tracking installed once and never audited: a monitored dead tag would have reported zero sessions, and someone would have cleaned it up. On these sites, the GA4 supposed to have taken over is almost never checked either.
• The tag manager in the minority: 121 sites (20.9%), two and a half times rarer than the analytics it is meant to orchestrate. Cross-referencing these 121 containers with the GA4-equipped sites, we estimate, as a common-sense approximation, that roughly four GA4 installs in ten lean on an orchestration layer; the other six hard-code their tags into the theme. The gap matters: without a tag manager, every tag addition or fix goes back through a developer, whereas a container handles events, triggers and consent without touching the code, and opens a clean path to the server. It is the layer that makes tracking maintainable, and it is missing on eight sites in ten.
• Paid acquisition in the minority: Meta Pixel on 101 sites, Google Ads tag on 88 (15.2%). The same total of 101 for two distinct tools, the dead Universal Analytics and the live Meta Pixel, is a coincidence, not a double count. That the advertising pixel is on a par with a dead Universal Analytics shows that the tag is sometimes installed before there is even a measurement foundation that holds. The tag is there, the brand pays for it, but without CAPI or `event_id`, it already sees only part of what it buys.

What emerges from the whole is not an absence of tools, it is an absence of architecture. The markers are there, sometimes in number, but assembled without the backbone that turns a collection of signals into a measurement system: browser-server deduplication, centralised tag management, governed consent. Between a structured fringe and a base that piles up scripts at the root of the theme, the gap comes down to method more than to technology.

GA4 present does not mean GA4 that measures

Detecting a GA4 script says nothing about what it instruments. A loaded tag does not make a measurement plan: you still need to track business events (`purchase`, `add_to_cart`, `begin_checkout`), feed cart and revenue through enhanced e-commerce, filter internal traffic and hold attribution. The 293 GA4 in the base are overwhelmingly default GA4, which counts page views without ever reading the funnel. The real work for a retailer flying blind goes beyond a simple « put GA4 in »: define a measurement plan aligned with its business events, then instrument it.

The cost of this leaky measurement is not only down to the ghost GA4 that load with no one to read the data. The 88 sites equipped with Google Ads and the 101 carrying a Meta Pixel optimise their media budget on false numbers: lacking Consent Mode v2, the conversions lost when cookies are refused are no longer modelled; without Enhanced Conversions on the Google side or a CAPI paired with an `event_id` on the Meta side, the ROAS displayed understates real performance and degrades. These stores over- or under-invest and steer their campaigns on a truncated measurement. To this is added a performance cost, which ties back to the performance finding: every tag weighs on load and on execution on the main thread, and the median mobile speed slides from 63 with no tool at all to 48 with a full stack of four tags. The correlation is strong, probably directional rather than strictly causal (the high tiers have small numbers, 8 sites with 4 tools), but the direction stays clear: you degrade the speed that conditions conversion while harvesting fragile data.

The 2026 answer, sequenced

Server-side tracking answers both ills in a single move. On performance, it pulls scripts out of the visitor's browser. On reliability, it extends first-party cookies, partly bypasses blockers and deduplicates conversions via the CAPI, hence more complete data when cookies are refused. It is the structural answer, to put in place after the foundations, never as a starting point: attempted too early on a rickety base, it adds complexity without fixing anything. The sequence reads as effort against impact.

• Remove the 101 dead Universal Analytics: five minutes per site, immediate weight savings, zero risk since the tag sends nothing back any more.
• GTM plus Consent Mode v2 on the 88 Google Ads and 101 Meta Pixel: this is where the leaky data costs cash, in budget steered blind. The step that secures compliance before investing more.
• A clean GA4 backed by a measurement plan: measure what you decide on, not everything.
• Server-side: for the volumes that justify it, last.

Our reading of this work is detailed on the expertises side.

The reading scale shifts here

This layer changes its focus. We leave the 579 e-shops measured one by one and move down into the real buying journey, payment and delivery, read page by page on a sub-sample of 59 stores crawled in depth, the ones whose payment and delivery pages were actually readable. What follows therefore describes a frequency of presence, a ranking, a direction, not market shares: a method that ranks first was simply detected across more stores, with no way of knowing the weight it carries across the base. Knowing how many stores a payment method shows up on says nothing about the euros that flow through it. We read the stores' technical shop window, their till stays shut, and that is exactly the reading a conversion audit lays down before touching anything else.

This shift of scale comes with a trade-off to own plainly: what we gain in granularity we lose in representativeness, we see better but across fewer stores. The rarer a method, the more its rank depends on the draw. The solid signals are the broad contrasts; the tight calls, we set aside. Three reading reflexes for what follows: a frequently detected method reflects a retailer's intent to offer, never proof of use; an absence weighs as heavily as a presence, most often a deliberate local choice rather than a technical oversight; the coherence of a stack matters as much as any single method taken in isolation, a broad panel behind a single aggregator not telling the same story as a panel cobbled together brick by brick.

A very local portrait, from payment to delivery

At the till, the landscape leaves no ambiguity. Bancontact and bank transfer dominate, two methods anchored deep in Belgian habits, followed by invoice payment, PayPal, then the aggregator Mollie. The signature of a business that serves a local customer base first, and that takes payment the way it always has.

This leading duo says a lot about the buyer being targeted. Bank transfer means opening your banking app, copying over an amount and a reference, then waiting for the funds to land before the parcel ships: a journey that assumes trust already in place, not an impulse buy. Bancontact is the quintessential Belgian domestic reflex, smooth for anyone who holds an account in the country, but off the table the moment you cross the border. These two methods quietly filter for the local buyer without slamming the door on anyone: they outline a commercial perimeter that is often chosen. A retailer rooted in its catchment area is not necessarily meant to take payment from the other end of the continent, and this choice of methods probably reflects that perimeter more than it suffers it. Invoice payment confirms the reading: paying after receipt assumes a level of trust that few pure players extend to an anonymous first-time buyer.

Against this picture, one misreading to dismiss right away. Mollie leading the field betrays neither immaturity nor local retreat: it is the de facto standard of Benelux e-commerce, the aggregator stores plug in to offer Bancontact, cards, wallets and local methods behind a single integration. Its presence marks technical maturity, not a lag to fix. PayPal, alongside it, adds international reassurance and card payment without having to enter your details on the merchant's site.

On the delivery side, the same local tone, the same caution. Shipping handled directly by the retailer comes ahead of Bpost, ahead of in-store pickup, with the DPD and Mondial Relay pickup-point networks bringing up the rear. This in-house logistics tells of a direct, almost artisanal relationship between seller and customer. The study crawls delivery-option pages, without ever touching volumes or shipping history: it tells what stores offer, not how much they ship. To draw a scalability diagnosis from it would be an excessive inference. A retailer may well handle its own shipping because it fits its area and its rhythm, with no aim to grow, and the absence of industrialised pickup-point networks says nothing about a growth ceiling that the buyer, for their part, probably never even perceives.

The cheapest brake to release, and the real decider

The payment method goes beyond a technical step at the bottom of the funnel: it is a filter. Two families of stores stand apart on friction. On one side, those plugged into an aggregator like Mollie or Stripe: the buyer lands on a standardised line-up, Bancontact, card, sometimes PayPal or instant transfer, and pays in a few seconds within a journey already lived a hundred times elsewhere. Near-zero cognitive cost. On the other, those accepting only the classic bank transfer or Bancontact alone, with no aggregator behind: a quiet filter cut out for a buyer already in confidence, most often local and loyal to the brand.

The nuance sits right there: this filter costs little to whoever targets only their regulars, and dearly to whoever would like to broaden out. For a lukewarm visitor, caught via a search or an ad, still with no tie to the brand, the absence of an immediate payment adds a step right when intent is at its most fragile. Without measuring sessions, we cannot claim they close the tab, but it is the most likely scenario. Still, of all the brakes observed, it is the cheapest to release: plugging in an aggregator touches neither the catalogue, nor the positioning, nor the logistics, and bank transfer stays an option for the regulars who prefer it. One of the rare points where the gap between implementation effort and potential gain tips so clearly to one side.

Still, payment is only one step, and it is the whole funnel that decides the sale. Four points weigh as much as the payment method. Forcing account creation instead of a guest checkout pushes back the rushed buyer who just wanted to pay once. Every superfluous field, against a checkout condensed into a single page, offers one more chance to close the tab. Shipping costs dropped at the final screen are the most documented cause of cart abandonment, around 48% according to the Baymard benchmark, far more brutal than the absence of delivery info. Bancontact on mobile, finally, where more than 60% of traffic flows, opens a redirect to the app or a QR code: real friction the moment the integration is botched.

Where the store delivers, and for how much

• Nearly 60% of the 537 commerce profiles, almost three stores in five, state nowhere where or how the store delivers: no covered zone, no lead time, no price grid before adding to cart.
• When the info exists, it most often limits itself to Belgium, sometimes extended to neighbouring countries, rarely formalised into a standalone Delivery page that would carry weight.
• The visitor left in the dark leaves rather than writes in: a cart that never started, a sale that never fires. The study crawls pages, not sessions: this upstream abandonment remains an inference, but it has the signature of silent forfeit, invisible in the statistics since nothing happened there that analytics could count.

Three clear lines on the delivery page demand neither a rebuild nor a budget: the kind of job an expertises team settles in a few hours. The CRO rule fits in one sentence: everything the buyer has to guess, they count as a risk. A stated lead time, a named carrier, a free-shipping threshold on display, a readable returns policy, and the page moves from a question mark to a verifiable promise. This signal ties back to the click-and-mortar finding, where two thirds of e-shops have a store behind them: part of the traffic probably comes from people ready to collect in store, provided the page says so.

For a decision-maker, the essential: three sales models coexist, to consumers, between professionals, or both, and none has pulled ahead. The choice that matters is not technical. It comes down to deciding who you are speaking to before optimising anything at all. The detail below puts numbers on that uniformity.

Across the 522 e-shops segmented by their customer model (a subset of the 537 profiles in the commerce layer, 15 profiles too ambiguous set aside for lack of clear signals, and a denominator still distinct from the study's global 579), three families coexist without any one crushing the others: 286 sell to consumers (B2C, 55%), 126 between professionals (B2B, 24%) and 110 mix both audiences (hybrid, 21%). B2C dominates in number, without a landslide all the same. First signal to retain: none of the three models is marginal. Each weighs at least a fifth of the panel, which forbids telling the story of Liège e-commerce as a purely consumer affair.

Nobody is mature, and that is the lesson

The three models share the same local fabric and equipment of the same order. This uniformity of under-equipment, more than the split of audiences, draws the real playing field. Median maturity index of 1.08 in B2C, 0.97 in B2B, 0.93 in hybrid: a single tool detected per segment. The gap between the best and the worst rated caps at 0.15 point on the index scale, a ranking to read in relative terms, not as a hierarchy. Performance and compliance, too, stay indifferent to the commercial model: median mobile perf of 60 to 61 everywhere, weak GDPR grades between 39% and 44%. Selling to consumers, to professionals or to both changes neither the site's speed nor its level of compliance, which points to a common cause upstream of the model, on the tooling and stack-choice side.

No profile comes out of the pack mature. The best students tick two or three technical boxes, rarely more; the mass ticks zero or one. No leading pack that would have industrialised its measurement while the others lag, but an entire fabric starting from roughly the same place. Bad news for anyone looking for a local benchmark to imitate, excellent news for anyone wanting to get ahead: the ground is flat, and the first real measurement infrastructure laid on this market will have no one to overtake, just a standard to set.

Three reading caveats

The maturity index aggregates four signals of different natures: analytics tool, advertising pixels, marketing tags and consent banner. These bricks are not equal. The banner, most often, is placed by default by the CMS or the theme, with no active step from the retailer: it mechanically inflates the score while saying nothing about real measurement. When a profile's median comes out at a single tool, that lone point is therefore frequently automatic consent, not deliberate tracking. It is a ceiling of presence, never a floor of use: the ability to measure and activate the data (GA4 configured, a pixel reporting conversions) is in practice rarer still than the aggregate would lead you to believe.

Second reservation, the bridge between the profiles layer and the commerce layer. The 18% of sites exposing in-store pickup (n=522) reads against the 66% of establishments holding a physical store (n=537, commerce layer), but these two percentages do not rest on the same subset: the 18% includes B2C pure players devoid, by construction, of any point of sale where an order could be collected. Figuring the "gap" as if it were the same scope would overstate the lag to close. The real dormant reserve sits in the intersection of players who have both a store and a site, never in the raw subtraction of the two rates. The trend, for its part, holds: a substantial share of retailers with a high-street presence does not expose pickup on their site.

Third reservation, the status of the signals. Everything the index captures is a detected presence, not a proven adoption nor a generated revenue. The "crumb of revenue" formula is an in-house thesis, a reading the study proposes without measuring it: we observe that a tag is loaded, we observe neither whether it feeds a decision, nor the revenue it serves. To hold as a bounded working hypothesis, not as an established result.

Choosing your audience is not enough, you have to choose your channel

Segmenting B2C, B2B or hybrid only has operational value once translated into an acquisition channel. Without that bridge, segmentation stays a diagnosis with no grip on action.

• Local B2C: non-brand Search on purchase intent, Meta for latent demand, and a web-to-store logic to bring people back in store, consistent with the two thirds of physical stores.
• B2B: tight intent-driven Search, LinkedIn to target the job function, and lead generation rather than a cart. In-store pickup stays an outlier here between professionals (only 2%); the return comes from lead and quote generation: quote form, lead tracking, scoring, integration with the CRM and with invoicing.
• Hybrid: at 21% of the base, the trap is serving both audiences the same way. The consumer wants a simple cart, a displayed price and a payment in three clicks; the professional expects a quote on account, negotiated terms, deferred invoicing. A single generic site calibrated for neither imposes a permanent splits stance where neither one nor the other is served. A site that ignores who it serves also ignores what to optimise: every trade-off (a button's wording, the order of the blocks, the placement of the price) assumes a clear-cut target to make sense.

The strategic framing goes all the way to the choice of channel, not just of audience. That is the work our expertises steer.

Where to start, segment by segment

Prioritising means attacking where the effort-to-impact ratio is highest.

• B2C (286 sites, 55%): the volume quick win. In-store pickup is exposed by only 18% of them while the majority has a high-street presence. Plugging in Click & Collect connects two assets already paid for, with no rebuild and no media budget. The return remains to be measured on the retailer's side rather than presumed from the storefront.
• Hybrid (110 sites, 21%): a heavy strategic project. Serving the consumer and the professional on a generic site imposes a repositioning ahead of any technical optimisation.
• B2B (126 sites, a quarter of the base): the most profitable job stays lead and quote generation, well ahead of in-store pickup.

In order, B2C first: maximum volume, minimum effort. Clarifying the target before optimising execution reads on the expertise side.

The physical anchoring of Liège e-commerce

Across the 537 complete profiles analyzed, 354 stores, or 66%, lean first on a very real point of sale. These are click-and-mortar businesses: the click of the web grafted onto the brick of the high-street store. Online, the channel extends the storefront more than it replaces it. Two thirds of local e-commerce is therefore first local retail that has fitted itself with a website.

This physical root sheds light on the rest of the study. If two thirds of the base are first stores, the gaps measured elsewhere, mobile slowness, compliance endured, the absence of measurement, reread as traditional retail at the start of its digitalization, not as negligence. A physical business learning a second trade, the web, without the budget or the culture that go with it. The nuance changes the diagnosis. What's missing isn't seriousness, it's time: running a store and a website at once, often alone, doesn't leave the room a pure player devotes to its sole technique.

Click & Collect, the asset asleep in the back room

This physical root sheds light on the most telling missing piece, Click & Collect, the in-store pickup of an order placed online. The most natural bridge between the store and the site, the one that demands the least reinvention since it builds on a store already there, equips only 13% of Liège e-shops, all profiles combined (18% for B2C alone, detailed below). A retailer who already runs a counter, stock and opening hours owns the essentials of the infrastructure. What they mostly lack is the online reservation and the promise kept at pickup.

The contrast with HTTPS, the basic encryption of a connection, near-universal at 577 sites out of 579, says a lot. When the foundation activates by default, asking nothing of the retailer, the host laying the certificate on its own at account opening, the base follows en masse. When you have to decide, prioritize and wire it yourself, like Click & Collect, it drops off. HTTPS gets ticked without a thought; Click & Collect stays a project to carry out.

The realistic potential doesn't climb to 100% for all that. A poorly run Click & Collect, stock out of sync between the site and the shelf, fuzzy availability times, disappoints more than it serves. The storefront opens the door to pickup, it doesn't guarantee it's done well. The gap to close therefore doesn't read as a mechanical jump from 13% to a theoretical ceiling, but as a margin open to brands that have a point of sale, reliable stock and the rigor to keep the promise.

A reading by segment, with clean boundaries

Crossed by customer typology, the contrast holds. Of the 537 complete profiles, 522 split into three models, the remaining 15 escaping the breakdown for lack of a usable signal. We flag that gap rather than smoothing it over before any reading in percentages.

• B2C: Click & Collect at 18% (286 players). The most natural ground for quick pickup. A local B2C basket bears shipping fees and delays poorly, which in-store pickup wipes out at a stroke.
• Hybrid: Click & Collect at 14% (110 players). Torn between two audiences with opposite logistics expectations: the consumer wants to pick up fast, the professional wants an invoiced, tracked delivery. This profile has to make the two promises coexist.
• B2B: Click & Collect at 2% (126 players). The invoice and the delivery remain the norm. But stopping at the norm would be reductive: for a local B2B reseller, professional pickup answers a real need, the spare part a tradesperson comes to collect urgently in the morning rather than waiting for a delivery round. Where that use exists, the low rate owes less to sector fate than to a missed opportunity.

A hypothesis to measure, a link upstream

A methodological caution remains, because it separates the finding from the promise. That Click & Collect is the most natural bridge for a physically rooted base is a structurally solid reading. That its installation mechanically generates growth remains, for its part, a hypothesis. The study observes equipment rates and typologies, not a single euro of store-by-store revenue. No incremental basket, no margin, no transaction tied to pickup was measured here. The link between the option and the revenue is plausible and documented elsewhere in retail, but it's verified case by case, on each brand's real figures.

A second safeguard is in order before wiring anything at all. Installing a Click & Collect without measuring the store visits it generates shifts the measurement problem instead of solving it. A web reservation not honored can't be told apart from an actual pickup. A customer who spots a product online then buys it on the shelf stays a black hole. That points straight back to the measurement layer: one e-shop in two is already flying blind, with no GA4. Without tracking of store visits or reconciliation from online to offline, the impact of pickup stays invisible, and therefore can't be arbitrated.

Wiring pickup, moreover, answers a demand already captured, the bottom of the funnel. For a click-and-mortar, the most profitable work often happens earlier, in local discovery: Google Business Profile listing, local SEO, store locator, store pages that rank on geo-targeted queries. That's where it's decided who walks through the door. Reducing web-to-store to Click & Collect alone means missing the top of the funnel, where digital truly brings the customer into the store. The two thirds of e-shops backed by a store have as much to gain from being found as from organizing pickup. That's exactly the ground of our missions in web-to-store, where digital feeds the store instead of replacing it.