Personalizacija in priporočila izdelkov

Kako vedenje med brskanjem uporabiti za boljša priporočila izdelkov

Posted on July 18, 2026 by

Vedenje med brskanjem prekaša vsak drug signal pri eni določeni nalogi: pove vam, kaj si stranka želi prav zdaj. Zgodovina nakupov vam pove, kaj si je želela pred meseci; seznam najbolje prodajanih vam pove, kaj si želijo vsi drugi. Toda izdelki, ki si jih je nekdo ogledoval ta teden – še posebej tisti, h katerim se je vračal – so živa izjava o nameri, in priporočila, zgrajena na njih, dosledno prekašajo splošne mreže. Metoda na kratko: sledite ogledom izdelkov, ogledom kategorij in iskanju na strani; ponovljene in nedavne oglede utežite najvišje; nato to pošljite v bloke izdelkov v svojih emailih in avtomatizacijah, tako da tisti, ki brska za pohodniškimi čevlji, dobi čevlje in pohodniške nogavice, ne pa vaše deseterice najbolje prodajanih po celotni trgovini. Ta članek pokriva, kako signal zajeti, ga razvrstiti in uporabiti – ne da bi zašli v zasledovalsko območje.

Ena meja takoj na začetku: tu gre za to, kaj priporočiti na podlagi brskanja. Namenski tok za opuščanje brskanja (časovni razpored, zaporedje, popusti) je svoja tema, bonton sklicevanja na podatke o brskanju v besedilu pa je pokrit v kako personalizirati emaile, ne da bi delovali vsiljivo.

Težava: vaša priporočila prezrejo, kaj so vam obiskovalci pravkar povedali

Takole je stanje v večini trgovin s soliden prometom. Obiskovalci brskajo – veliko. Tipična seja se dotakne treh do šestih strani izdelkov. Nato jih večina odide in naslednji email, ki ga dobijo od vas, izpostavlja … kar koli ste dali v predlogo kampanje. Najbolje prodajane, novo kolekcijo, letošnjo promocijo.

Obiskovalka torej deset minut primerja stoječe mize na vaši strani, dva dni pozneje pa ji pošljete email o dišečih svečah, ker so sveče v trendu. Ne klikne. Ne zato, ker bi bila slaba priložnost – je ena vaših najboljših –, ampak zato, ker ste odgovorili na vprašanje, ki ga nikoli ni postavila.

Podatki za rešitev tega že obstajajo v vaši analitiki in sledenju strani vaše email platforme. Večina trgovin jih zbira in nikoli ne usmeri v posamezno sporočilo.

Zakaj “kar pokažimo najbolje prodajane” pušča denar za sabo

Najbolje prodajani so udobna privzeta izbira in niso neuporabni – za prvega obiskovalca, o katerem ne veste ničesar, je priljubljenost razumno ugibanje. A mreža najbolje prodajanih je povprečen odgovor za povprečno stranko, nobena vaša stranka pa ni povprečna.

Odpoved je najostrejša pri trgovinah s širokim katalogom. Če prodajate 800 izdelkov v desetih kategorijah, se vaši najbolje prodajani verjetno kopičijo v eni ali dveh. Vsak email, ki ga pošljete, jih vodi v ospredju, stranka, ki brska po vaših drugih osmih kategorijah, pa nikoli ne vidi, da bi bili odsevani njeni interesi. Sklene, da vaši emaili niso zanjo, in jih neha odpirati. Padajoče stopnje odpiranja na seznamu, polnem dejavnih brskalcev, so pogosto točno to neujemanje.

Kje izteka prihodek: vrzel med ogledom in nakupom

Le majhen delež obiskovalcev strani izdelkov kupi v tisti seji. Ostali odidejo z izraženim interesom in brez spremljave, ki bi bila nanj vezana. To je izteka.

Postavite številko za svojo trgovino. Recimo, da 8.000 mesečnih obiskovalcev doseže stran izdelka, vaša stopnja prepoznanih stikov je 25 % in 3 % sej z ogledi konvertira. To pusti približno 1.940 prepoznavnih ljudi vsak mesec, ki so si ogledali določene izdelke in o njih niso dobili nobenega sporočila. Če email s priporočili na podlagi brskanja pripelje nazaj k nakupu vsaj 2 % od njih ob povprečni vrednosti naročila 60 €, je to približno 2.300 € na mesec, ki jih trenutno ne pobirate. (Ilustrativne številke – vstavite svoje.)

Bistvo izračuna ni točna številka. Bistvo je, da izteka narašča s prometom, ki ste ga že plačali.

Gradnja sistema: trije signali, ena razvrstitev

Korak 1 – zajemite tri signale brskanja

Potrebujete nameščen sledilni izsek vaše email platforme, da se vedenje poveže z znanimi stiki (naročniki in prijavljene stranke). Nato zberite:

Ogledi izdelkov – osrednji signal. Katere izdelke, kolikokrat, kako nedavno.
Ogledi kategorij – eno raven višje in bolj stabilni od podatkov o posameznem izdelku. Nekdo, ki poskakuje po “moškem teku”, vam je povedal svoj oddelek, tudi če ni izstopal noben posamezen čevelj. Celoten vodnik je v uporaba naklonjenosti kategorijam izdelkov.
Iskalni izrazi na strani – najbolj ekspliciten signal od vseh. Iskanje “nepremočljiva jakna” je prošnja z jasnimi besedami. Manj platform ga uporablja; če ga vaša izpostavi, ravnajte z njim kot z zlatom.

Korak 2 – razvrstite po nedavnosti, pogostosti in globini

Niso vsi ogledi enaki. Praktično uteževanje, brez diplome iz podatkovne znanosti:

Ponovljeni ogledi enega izdelka – najmočnejši posamezni pokazatelj. Vrnitev na isto stran izdelka dvakrat v enem tednu je blizu dvignjeni roki.
Nedavno premaga staro. Ogled od včeraj prekaša pet ogledov izpred dveh mesecev. Namera pri brskanju hitro razpada – za večino katalogov je večinoma zastarela v nekaj tednih, hitreje pri impulzivnih kategorijah, počasneje pri premišljenih nakupih, kot je pohištvo.
Globina premaga odboj. 90-sekundni obisk, ki je preletel galerijo, prekaša 4-sekundni odboj, kadar vaše orodje zna ločiti med njima.

Korak 3 – priporočajte okoli ogledanega izdelka, ne le samega izdelka

To je korak, ki ga večina trgovin preskoči. Ponovni prikaz točno ogledanega izdelka deluje – to je klasični blok nedavno ogledanih, ki je pravilno pokrit v uporaba nedavno ogledanih izdelkov v emailih. A brskanje vam pove več kot “želi si to svetilko”. Pove vam, da opremlja sobo. Močan blok na podlagi brskanja je torej: najprej ogledani izdelek, nato dva ali trije sosedje – ista kategorija, sosednja cena ali pogosto kupljeno skupaj. Odgovarjate na temeljno nakupovalno poslanstvo in še vedno konvertirate, kadar določen izdelek ni bil čisto pravi.

Korak 4 – uporabite izključitve

Preden se kar koli pošlje: zavrite že kupljene izdelke, izdelke, ki jih ni na zalogi, in vse, ki so v zadnjih 30 dneh kupili iz te kategorije. Nič ne spodkoplje “vemo, kaj želite” tako kot priporočanje tistega, kar je nekdo kupil prejšnji teden – ta način odpovedi dobi svoj članek v kako se izogniti priporočanju izdelkov, ki jih je stranka že kupila.

Kaj avtomatizirati: dva toka in ena nadgradnja kampanje

Tok 1 – opuščanje brskanja. Sprožilec: izdelek ogledan, brez dodajanja v košarico, seja končana. Segment: naročeni, brez nakupa v tej kategoriji v 30 dneh. Zamik: nekaj ur. Kanal: email. Vsebina: zadeva, uokvirjena s kategorijo, ogledani izdelek vodi blok štirih izdelkov s priporočili. Cilj: ponovni obisk in nakup v enem tednu.

Tok 2 – spremljava po interesu za kategorijo. Sprožilec: 3 ali več ogledov ene kategorije v 14 dneh, brez nakupa. Zamik: 1–2 dneva po prekoračitvi praga. Vsebina: kratko sporočilo o tej kategoriji (dobro deluje pristop nakupovalnega vodnika) z najboljšimi izbori iz nje. Cilj: prvi nakup v kategoriji. Ta ujame raziskovalce, ki jih tok posameznega izdelka zgreši.

Nadgradnja kampanje. V rednih e-novicah statično mrežo izdelkov zamenjajte z dinamičnim blokom, ki ga poganja zgodovina brskanja vsakega prejemnika, z najbolje prodajanimi kot rezervo za stike brez podatkov o brskanju. Ista kampanja, drugačni izdelki za vsako osebo – običajno najhitrejša zmaga od treh, ker se dotakne vsakega pošiljanja.

Primer trgovine s kolesi (ilustrativno)

Trgovina, ki prodaja kolesa in opremo, ima naročnika, ki si v desetih dneh ogleda dve gravel kolesi (eno od njiju trikrat), brska po kategoriji plaščev in išče “tubeless nastavitev”. Brez nakupa.

Logika najbolje prodajanih mu pošlje email o pocenjenem mestnem e-kolesu. Logika brskanja prebere sled: tok vodi trikrat ogledano gravel kolo, podprto s kompletom tubeless plaščev in mini tlačilko, pod zadevo o vožnji po makadamu. Kolesa za 2.400 € morda ne bo kupil iz emaila – a klikne, in tudi če se zdaj prime le za 70 € dodatkov, je trgovina potrdila njegov interes in lahko naprej sporoča pravemu oddelku, namesto da bi ugibala.

Kako to izmeriti

Stopnja klikov pri blokih na podlagi brskanja v primerjavi z vašimi statičnimi mrežami – prvi in najjasnejši signal, viden v enem ali dveh pošiljanjih.
Prihodek na prejemnika pri tokovih, sproženih z brskanjem, v primerjavi s kampanjami. Vedenjski tokovi bi morali zmagovati z veliko razliko; če ne, je vaša razvrstitev ali so izključitve narobe.
Stopnja konverzije toka opuščanja brskanja konkretno.
Premik v stopnji odpiranja po kategorijah – če odpiranja narastejo, potem ko začnete ujemati vsebino z brskanimi kategorijami, je bila teorija o neujemanju pravilna.

Kje se vključi Omnisend

Vsaka platforma s sledenjem strani lahko izvede različico tega. V svojih trgovinah ga poganjam na Omnisendu – Klaviyo in Omnisend sem preizkusil vzporedno in obdržal Omnisend zaradi lažje samopostrežne nastavitve in cene. Deli, ki tu štejejo: sledilni izsek veže oglede izdelkov in kategorij na stike, opuščanje brskanja je standarden sprožilec toka in ne izdelava po meri, blok priporočil izdelkov pa se lahko napolni iz nedavno ogledanih izdelkov z uporabljenimi izključitvami. Tok interesa za kategorijo zahteva malce več dela s segmenti, a nič, kar bi potrebovalo razvijalca.

Poštene omejitve: podatki o brskanju se prilepijo le na prepoznane obiskovalce, zato ves ta sistem teče na podmnožici prometa, ki je naročena ali prijavljena – še en razlog, zakaj rast seznama množi vse ostalo. In vrzeli pri sledenju (izbire glede soglasja s piškotki, več naprav) pomenijo, da so podatki usmerjevalni, nikoli popolni.

Vaš naslednji korak

Danes preverite eno stvar: ali je sledenje strani vaše email platforme dejansko nameščeno in se proži na straneh izdelkov. Prav ta ena integracija je vrata do vsega zgoraj. Ko je delujoča, najprej zgradite tok opuščanja brskanja – najvišja namera, najhitrejše povračilo – nato pa razširite z kako priporočiti izdelke, ki jih bo stranka najverjetneje kupila.

Personalization and Product Recommendations

How to Use Browsing Behavior for Better Product Recommendations

Posted on July 17, 2026 by

Browsing behavior beats every other signal for one specific job: telling you what a customer wants right now. Purchase history tells you what they wanted months ago; a bestseller list tells you what everyone else wants. But the products someone viewed this week — especially the ones they returned to — are a live statement of intent, and recommendations built on them consistently outperform generic grids. The method in short: track product views, category views, and on-site search; weight repeat views and recent views highest; then feed that into your email and automation product blocks so a browser of hiking boots gets boots and trail socks, not your storewide top ten. This article covers how to capture the signal, rank it, and use it — without crossing into stalker territory.

One boundary up front: this is about what to recommend based on browsing. The dedicated browse abandonment flow (timing, sequence, discounts) is its own topic, and the etiquette of referencing browse data in copy is covered in how to personalize emails without being creepy.

The problem: your recommendations ignore what visitors just told you

Here’s the situation in most stores doing solid traffic. Visitors browse — heavily. A typical session touches three to six product pages. Then most leave, and the next email they get from you features… whatever you put in the campaign template. Bestsellers, the new collection, this month’s promotion.

So a visitor spends ten minutes comparing standing desks on your site, and two days later you email her about scented candles because candles are trending. She doesn’t click. Not because she’s a bad lead — she’s one of your best — but because you answered a question she never asked.

The data to fix this already exists in your analytics and your email platform’s site tracking. Most stores collect it and never route it into a single message.

Why “just show bestsellers” leaves money behind

Bestsellers are the comfortable default, and they’re not useless — for a first-time visitor you know nothing about, popularity is a reasonable guess. But a bestseller grid is the average answer for the average customer, and none of your customers are average.

The failure is sharpest for stores with wide catalogs. If you sell 800 products across ten categories, your top sellers likely cluster in one or two of them. Every email you send leads with those, and the customer who browses your other eight categories never sees her interests reflected. She concludes your emails aren’t for her and stops opening. Falling open rates on a list full of active browsers is often exactly this mismatch.

Where the revenue leaks: the gap between viewing and buying

Only a small fraction of product-page visitors buy in that session. The rest leave with expressed interest and no follow-up tied to it. That’s the leak.

Put a number on it for your own store. Say 8,000 monthly visitors reach a product page, your identified-contact rate is 25%, and 3% of viewing sessions convert. That leaves roughly 1,940 identifiable people each month who looked at specific products and got no message about them. If a browse-based recommendation email brings even 2% of them back to buy at a €60 average order, that’s about €2,300 a month you’re currently not collecting. (Illustrative numbers — swap in your own.)

The point of the calculation isn’t the exact figure. It’s that the leak scales with traffic you already paid for.

Building the system: three signals, one ranking

Step 1 — capture the three browse signals

You need your email platform’s tracking snippet installed so behavior links to known contacts (subscribers and logged-in customers). Then collect:

Product views — the core signal. Which items, how many times, how recently.
Category views — one level up, and steadier than single-product data. Someone bouncing around “men’s running” has told you their department even if no single shoe stood out. There’s a full guide in using product category affinity.
On-site search terms — the most explicit signal of all. A search for “waterproof jacket” is a request in plain words. Fewer platforms use it; if yours exposes it, treat it as gold.

Step 2 — rank by recency, frequency, and depth

Not all views are equal. A practical weighting, no data science degree required:

Repeat views of one product — the strongest single indicator. Returning to the same product page twice in a week is close to a raised hand.
Recent beats old. A view from yesterday outranks five views from two months ago. Browse intent decays fast — for most catalogs it’s mostly stale within a couple of weeks, faster for impulse categories, slower for considered purchases like furniture.
Depth beats bounce. A 90-second visit that scrolled the gallery outranks a 4-second bounce, where your tooling can tell the difference.

Step 3 — recommend around the viewed product, not just the product itself

This is the step most stores skip. Re-showing the exact viewed item works — that’s the classic recently-viewed block, covered properly in using recently viewed products in email. But browsing tells you more than “she wants this lamp.” It tells you she’s furnishing a room. So a strong browse-based block is: the viewed item first, then two or three neighbors — same category, adjacent price, or commonly bought together. You’re answering the underlying shopping mission, and you still convert when the specific item wasn’t quite right.

Step 4 — apply the exclusions

Before anything sends: suppress products already purchased, out-of-stock items, and anyone who bought from that category in the last 30 days. Nothing undermines “we know what you want” like recommending the thing someone bought last week — that failure mode gets its own article in avoiding recommending products a customer already bought.

What to automate: two flows and one campaign upgrade

Flow 1 — browse abandonment. Trigger: product viewed, no add-to-cart, session ended. Segment: subscribed, no purchase in that category in 30 days. Delay: a few hours. Channel: email. Content: category-framed subject, viewed product leading a four-item recommendation block. Goal: return visit and purchase within a week.

Flow 2 — category-interest follow-up. Trigger: 3+ views in one category within 14 days, no purchase. Delay: 1–2 days after the threshold. Content: a short note on that category (a buying guide angle works well) with top picks from it. Goal: first purchase in the category. This one catches researchers the single-product flow misses.

Campaign upgrade. In regular newsletters, swap the static product grid for a dynamic block driven by each recipient’s browse history, with bestsellers as the fallback for contacts with no browse data. Same campaign, different products per person — usually the fastest win of the three because it touches every send.

A bike-shop example (illustrative)

A store selling bikes and gear has a subscriber who, over ten days, views two gravel bikes (one of them three times), browses the tire category, and searches “tubeless setup.” No purchase.

Bestseller logic emails him the discounted city e-bike. Browse logic reads the trail: the flow leads with the thrice-viewed gravel bike, backed by a tubeless tire set and a mini pump, under a subject about gravel riding. He may not buy the €2,400 bike from an email — but he clicks, and even if only the €70 in accessories lands now, the store has confirmed his interest and can keep messaging the right department instead of guessing.

How to measure it

Click-through rate on browse-driven blocks vs. your static grids — the first and clearest signal, visible within a send or two.
Revenue per recipient on browse-triggered flows vs. campaigns. Behavioral flows should win by a wide margin; if they don’t, your ranking or exclusions are off.
Conversion rate of the browse abandonment flow specifically.
Category open-rate drift — if opens climb after you start matching content to browsed categories, the mismatch theory was right.

Where Omnisend fits

Any platform with site tracking can do a version of this. In my stores I run it on Omnisend — I tested Klaviyo and Omnisend side by side and kept Omnisend for the easier self-service setup and the price. The pieces that matter here: the tracking snippet ties product and category views to contacts, browse abandonment is a standard flow trigger rather than a custom build, and the product recommender block can populate from recently viewed items with your exclusions applied. The category-interest flow takes a little more segment work, but nothing that needs a developer.

Honest limits: browse data only attaches to identified visitors, so this whole system runs on the subset of traffic that’s subscribed or logged in — one more reason list growth compounds everything else. And tracking gaps (cookie consent choices, multiple devices) mean the data is directional, never complete.

Your next step

Check one thing today: whether your email platform’s site tracking is actually installed and firing on product pages. That single integration is the gate to everything above. Once it’s live, build the browse abandonment flow first — highest intent, fastest payback — and then widen out with how to recommend products a customer is most likely to buy.