A/B-testiranje in optimizacija

Kaj naj spletna trgovina najprej preveri z A/B-testom?

Posted on July 18, 2026 by

Testirajte ponudbo v avtomatiziranem emailu, ki vam že zdaj prinaša največ denarja — pri večini trgovin je to email za opuščeno košarico. Ne barve gumba, ne časa pošiljanja, ne pisave. Prvi test naj bo čim bližje prihodku, naj teče na emailu z največjim obsegom, ki ga imate, in naj primerja dve resnično različni različici, saj samo ta kombinacija da odgovor, ki mu majhna ali srednje velika trgovina lahko zares zaupa. Vse ostalo pri testiranju — zadeve, časovnica, oblikovanje — je pomembno, a pride na vrsto pozneje. Ta članek pojasni, zakaj je to pravi prvi test, kaj za zdaj preskočiti in kako ga izpeljati tako, da rezultat kaj pomeni.

Pravi problem: preveč idej, premalo prometa

Če ustvarite 20.000–100.000 € na mesec, ste verjetno prebrali iste nasvete o testiranju kot jaz. Testirajte vse. Vedno testirajte. Petdeset idej za vaš naslednji A/B-test. Zato odprete svoje orodje za email, zagledate bleščeč gumb »A/B-test« in v torkovi kampanji testirate dve zadevi na 4.000 ljudeh.

Ena različica dobi 21 % odpiranj, druga 22,4 %. Izberete zmagovalca, počutite se znanstveno in v svojem poslu ne spremenite ničesar.

To je past. Trgovina vaše velikosti nima Amazonovega prometa. Morda dobite peščico smiselnih rezultatov testov na četrtletje, ne na dan. Vsak testni termin je redek, zato je koristno vprašanje veliko ožje kot »kaj bi lahko testirali?« — bliže je vprašanju »kateri en sam test je vreden, da nanj porabimo cel mesec prometa?«

Zakaj testiranje tistega, kar je najlažje, ne deluje

Običajni pristop je testirati tisto, kar vam orodje ponudi najlažje: zadeve, emoji proti brez emojija, oranžen gumb proti zelenemu. Dve stvari gresta narobe.

Prvič, majhne spremembe prinesejo majhne razlike, majhne razlike pa se na kratkem seznamu utopijo v šumu. Če različico A odpre 200 ljudi, različico B pa 212, je ta razlika enaka metanju kovanca. Test bi morali ponoviti mnogokrat, preden bi vzorec kaj pomenil, in tega skoraj nihče ne počne. Bolj poglobljena razlaga, zakaj je tako, je v kako velik mora biti email seznam za A/B-testiranje.

Drugič, tudi resnična zmaga na plitki metriki je lahko vredna skoraj nič. Zadeva, ki dvigne odpiranja, ne dvigne nujno tudi naročil. Zadeve so mi zmagale bitko za odpiranja in izgubile na prihodku, ker je bila »zmagovalna« vrstica klikvaba, ki je pritegnila napačno pozornost. Odpiranja so vrh lijaka; najemnino plačate na dnu.

Kje se izguba dejansko dogaja

Tu je tiha cena tega, da najprej testirate napačno stvar: priložnost. Vsak teden, ko vaš testni termin zaseda poskus z barvo gumba, je teden, ko niste izvedeli, ali vaš email za košarico sploh potrebuje popust — vprašanje, vredno pravega denarja.

Preračunajte za svojo trgovino. Recimo, da vaš tok za opuščeno košarico doseže 600 ljudi na mesec in ob priloženem 10-odstotnem popustu obnavlja naročila po neki stopnji. Če bi različica brez popusta obnovila skoraj enako veliko naročil, bi obdržali skoraj enak prihodek in nehali podarjati maržo pri vsaki obnovljeni košarici. Pri povprečnem naročilu 70 € je 10 % kar 7 € marže na naročilo, pri vsakem naročilu, za vedno. Noben test zadeve ne premakne toliko denarja. (Številke so ponazoritvene — vstavite svoje.)

To je oblika prvega testa, ki ga je vredno izvesti: odgovor spremeni to, kar počnete, sprememba pa se z leti kopiči.

Kako izbrati prvi test: trije filtri

Vsako kandidatno idejo preženite skozi te filtre, po vrsti.

Obseg. Email mora doseči dovolj ljudi, da se test zaključi v mesecu ali dveh. Vaša največja avtomatizacija ali vaš redni seznam za kampanje ustreza. Email, ki se sproži 40-krat na mesec, ne.
Razdalja do denarja. Dajte prednost testom, ki jih sodite po naročilih ali prihodku na prejemnika, pred testi, ki jih sodite po odpiranjih. Ponudba, glavno sporočilo in poziv k dejanju so blizu denarja. Kozmetika ni.
Velikost kontrasta. Testirajte dva smiselno različna pristopa, saj veliki kontrasti prinesejo razlike, dovolj velike, da jih na skromnem seznamu vidite skozi šum. Popust proti brez popusta. En izdelek proti mreži priporočil. Kratko golo besedilo proti oblikovani predlogi.

Pri večini trgovin ti filtri kažejo na isto mesto: spodbuda v emailu za opuščeno košarico. Visoka namera, stabilen obseg, neposreden odčitek prihodka in preprost velik kontrast. Če toka za košarico še nimate, pozdravna ponudba sledi isti logiki, le en korak prej.

Če želite bolj formalen način razvrščanja testnih idej po vsej trgovini — ne le pri emailu — se pristop z ocenjevanjem v kateri poskus izvesti najprej: okvir za testiranje na podlagi prihodka poglobi bolj.

Česa izrecno še ne testirajte

Kratek seznam testov, ki bi jih prvih nekaj mesecev odložil:

Zadeve — vredno jih je narediti pravilno pozneje; glejte kako pravilno A/B-testirati zadeve emailov za to, kako doseči, da kaj pomenijo.
Časi pošiljanja — pravi učinki obstajajo, a so majhni in počasi zaznavni na srednje velikem seznamu.
Popravki oblike in postavitve — barve, pisave, velikosti slik. Za okus je v redu, redko odločilno za prihodek.
Karkoli z dvema spremembama hkrati — nova ponudba in nova postavitev vam o nobeni od njiju ne povesta ničesar. Ena sprememba na test, vedno: kako testirati eno spremenljivko naenkrat.

Postavitev prvega testa, konkretno

Za primer vzemimo test ponudbe v košarici:

Sprožilec: košarica opuščena, brez nakupa v okviru zamika vašega toka.
Delitev: 50/50, naključno razdeljeno z orodjem, uporabljeno ob vstopu v tok, tako da vsak, ki opusti košarico, dobi natanko eno vejo.
Različica A: vaš trenutni email, vključno s popustom.
Različica B: isti email, ista časovnica, popust odstranjen — zamenjan s preprostim opomnikom in morda zagotovilom glede dostave ali vračil. Nič drugega se ne spremeni.
Pripis: v vsaki veji uporabite drugačno kodo za popust (tudi če je »koda« ene veje le sledene povezave), da se naročila čisto preslikajo na različice.
Trajanje: naj teče, dokler vsaka veja ne prejme vsaj nekaj sto dostav in po možnosti nekaj deset naročil — pri mnogih trgovinah je to 4–8 tednov. Ne kukajte vsak dan in ne ustavite predčasno, ker ena stran v prvem tednu skoči naprej. Zgodnje prednosti se nenehno obračajo.

Med tekom testa se uprite skušnjavi, da bi katero od vej »samo malo popravili«. Vsaka sprememba sredi testa ponastavi uro.

Kako izgleda rezultat (ponazoritveno)

Recimo, da je po šestih tednih vsaka veja dostavila 900 emailov. Veja s popustom je prinesla 54 naročil po 63 € povprečno po popustu; veja brez popusta 45 naročil po 70 €. Prihodek: 3.402 € proti 3.150 € — popust je v prednosti. Marža ob 30-odstotni stroškovni osnovi: približno 1.512 € proti 1.890 € — veja brez popusta je v prednosti z veliko razliko. Te številke so izmišljene, da prikažejo mehaniko, a vzorec je dovolj pogost: popust »zmaga« na naročilih in izgubi na dobičku. Katerega zmagovalca okronate, je odvisno od tega, ali optimizirate za prihodek, maržo ali pridobivanje novih strank — to se odločite pred začetkom testa, ne potem, ko ste že videli številke.

Metrike, ki odločijo

Tri številke na vejo:

Naročila na prejemnika — odčitek konverzije.
Prihodek na prejemnika — naročila, obtežena s tem, koliko so ljudje dejansko porabili.
Ocenjena marža na prejemnika — prihodek minus strošek izdelka minus dani popust. Tej bi prepustil dokončno odločitev.

Odpiranja in kliki so diagnostika, ne razsodba. Če veja brez popusta dobi klike, a ne naročil, je težava v ponudbi; če ne dobi klikov, je težava v sporočilu.

Kako to izvesti v Omnisendu

To zmore vsaka resna email platforma. V Omnisendu — ki ga uporabljam v svojih trgovinah, potem ko sem preizkusil Klaviyo, večinoma zato, ker sem lahko tokove gradil in spreminjal sam, ne da bi se bal vmesnika — na vrh delovnega toka za opuščeno košarico dodate delitev, zgradite dve email različici ter berete naročila in prihodek na vejo kar v poročilu samega delovnega toka, namesto da bi številke sešivali iz izvozov. Omnisend je partner Shopimationa v okviru pridruženega programa; priporočilo izhaja iz vsakodnevne uporabe, ne iz provizije.

Ena poštena pripomba: orodje naključno razdeli in poroča, ne bo pa vas ustavilo, da bi test končali po petih dneh, ker ena veja izgleda vznemirljivo. Ta disciplina je na vas.

Vaš naslednji korak

Odprite poročila o avtomatizacijah in poiščite email, ki je v zadnjih 90 dneh ustvaril največ prihodka. To je vaše testno prizorišče. Zapišite eno največjo predpostavko, vgrajeno vanj — običajno »ta popust mora biti tu« — in postavite delitev 50/50, ki jo izzove. Začnite ta teden; ura teče le, dokler teče test.

A/B Testing and Optimization

What Should an Ecommerce Store A/B Test First

Posted on July 17, 2026 by

Test the offer in the automated email that already makes you the most money — for most stores, that’s the abandoned cart email. Not the button color, not the send time, not the font. The first test should sit as close to revenue as possible, run on the highest-volume email you have, and compare two genuinely different versions, because that combination is the only one that produces an answer a small or mid-sized store can actually trust. Everything else about testing — subject lines, timing, design — matters, but it comes after. This article explains why that’s the right first test, what to skip for now, and how to run it so the result means something.

The real problem: too many ideas, not enough traffic

If you’re doing €20,000–100,000 a month, you’ve probably read the same testing advice I did. Test everything. Always be testing. Fifty ideas for your next A/B test. So you open your email tool, see a shiny “A/B test” toggle, and test two subject lines on a Tuesday campaign to 4,000 people.

One version gets 21% opens, the other 22.4%. You pick the winner, feel scientific, and change nothing about your business.

That’s the trap. A store your size doesn’t have Amazon’s traffic. You might get a handful of meaningful test results per quarter, not per day. Each test slot is scarce, so the useful question is much narrower than “what could we test?” — it’s closer to “which single test is worth burning a month of traffic on?”

Why testing whatever’s easiest doesn’t work

The usual approach is to test whatever the tool makes convenient: subject lines, emoji vs. no emoji, orange button vs. green button. Two things go wrong.

First, small changes produce small differences, and small differences drown in noise on a small list. If 200 people open version A and 212 open version B, that gap is coin-flip territory. You’d need to repeat the test many times over before the pattern means anything, and almost nobody does. There’s a fuller explanation of why in how large does an email list need to be for A/B testing.

Second, even a real win on a shallow metric can be worth almost nothing. A subject line that lifts opens doesn’t necessarily lift orders. I’ve had subject lines win the open battle and lose on revenue because the “winning” line was clickbait that attracted the wrong attention. Opens are the top of the funnel; your rent gets paid at the bottom.

Where the loss actually occurs

Here’s the quiet cost of testing the wrong thing first: opportunity. Every week your test slot is occupied by a button-color experiment is a week you didn’t learn whether your cart email needs a discount at all — a question worth real money.

Run the numbers on your own store. Say your abandoned cart flow reaches 600 people a month and recovers orders at some rate with a 10% discount attached. If a no-discount version recovered nearly as many orders, you’d keep almost the same revenue and stop giving away margin on every recovered cart. On a €70 average order, 10% is €7 of margin per order, every order, forever. No subject line test moves that kind of money. (Figures illustrative — plug in your own.)

That’s the shape of a first test worth running: the answer changes what you do, and the change compounds.

How to pick your first test: three filters

Run every candidate idea through these, in order.

Volume. The email must reach enough people to finish a test within a month or two. Your biggest automation or your regular campaign list qualifies. An email that fires 40 times a month doesn’t.
Distance to money. Prefer tests judged in orders or revenue per recipient over tests judged in opens. Offer, main message, and call to action sit close to money. Cosmetics don’t.
Size of the contrast. Test two meaningfully different approaches, because big contrasts produce differences big enough to see through noise on a modest list. Discount vs. no discount. One product vs. a recommendation grid. Short plain-text note vs. designed template.

For most stores those filters point at the same place: the incentive in the abandoned cart email. High intent, steady volume, direct revenue readout, and an easy big contrast. If you don’t have a cart flow yet, the welcome offer is the same logic applied one step earlier.

If you want a more formal way to rank test ideas across your whole store — not just email — the scoring approach in which experiment should you run first: a revenue-based testing framework goes deeper.

What to explicitly not test yet

A short list of tests I’d park for the first few months:

Subject lines — worth doing properly later; see A/B testing email subject lines properly for how to make them mean something.
Send times — real effects exist but they’re small and slow to detect on a mid-sized list.
Design and layout tweaks — colors, fonts, image sizes. Fine for taste, rarely decisive for revenue.
Anything with two changes at once — a new offer and a new layout tells you nothing about either. One change per test, always: how to test one variable at a time.

Setting up the first test, concretely

Using the cart-offer test as the example:

Trigger: cart abandoned, no purchase within your flow’s delay.
Split: 50/50, randomized by the tool, applied at the flow’s entry so each abandoner gets exactly one branch.
Variant A: your current email, discount included.
Variant B: same email, same timing, discount removed — replaced with a plain reminder and maybe a shipping or returns reassurance. Nothing else changes.
Attribution: use a different discount code in each branch (even if one branch’s “code” is just tracked links), so orders map cleanly to variants.
Duration: run until each branch has received at least a few hundred deliveries and, ideally, a few dozen orders — for many stores that’s 4–8 weeks. Don’t peek daily and don’t stop early because one side jumps ahead in week one. Early leads reverse constantly.

While the test runs, resist the urge to “just tweak” either branch. Any mid-test edit resets the clock.

What a result looks like (illustrative)

Say after six weeks each branch delivered 900 emails. The discount branch produced 54 orders at €63 average after discount; the no-discount branch produced 45 orders at €70. Revenue: €3,402 vs. €3,150 — discount ahead. Margin at 30% cost basis: roughly €1,512 vs. €1,890 — no-discount ahead by a wide gap. These numbers are invented to show the mechanics, but the pattern is common enough: the discount “wins” on orders and loses on profit. Which winner you crown depends on whether you’re optimizing for revenue, margin, or new-customer acquisition — decide that before the test starts, not after you’ve seen the numbers.

The metrics that decide it

Three numbers, per branch:

Orders per recipient — the conversion readout.
Revenue per recipient — orders weighted by what people actually spent.
Estimated margin per recipient — revenue minus product cost minus discount given. This is the one I’d let make the final call.

Opens and clicks are diagnostics, not verdicts. If the no-discount branch gets clicks but no orders, the problem is the offer; if it gets no clicks, the problem is the message.

Running it in Omnisend

Any serious email platform can do this. In Omnisend — which I use in my own stores after trying Klaviyo, mostly because I could build and change flows myself without dreading the interface — you add a split at the top of the abandoned cart workflow, build the two email variants, and read orders and revenue per branch in the workflow’s own report rather than stitching numbers together from exports. Omnisend is an affiliate partner of Shopimation; the recommendation comes from daily use, not the commission.

One honest caveat: the tool randomizes and reports, but it won’t stop you from ending a test after five days because one branch looks exciting. That discipline is on you.

Your next step

Open your automation reports and find the email that generated the most revenue in the last 90 days. That’s your test venue. Write down the one biggest assumption baked into it — usually “this discount needs to be here” — and set up a 50/50 split that challenges it. Start it this week; the clock only runs while the test does.