Umetna inteligenca v marketingu spletne trgovine

Kako z umetno inteligenco izboljšati emaile za zapuščene košarice

Posted on July 18, 2026 by

Če vaši emaili za zapuščene košarice že tečejo, a je povračanje zastalo, je AI najbolj koristna kot partner pri prepisovanju in testiranju, ne kot čarobni stroj za povračanje. Podajte ji svoje trenutne emaile za košarico ter nekaj konteksta — kaj prodajate, vaša povprečna vrednost naročila, kateri ugovor običajno zaustavi nakup — in jo prosite, naj prepiše zadeve, zbije besedilo in predlaga različice, ki jih lahko testirate. Hitra je pri delu, ki vas dolgočasi: da izdela pet poštenih pristopov namesto tistega enega, ki ga pošiljate že leto dni. Kar ne zmore, je popraviti tok košarice, ki je strukturno pokvarjen — napačna časovnica, brez opomnika na dejanski izdelek, prezgodaj podeljen popust. Ta članek govori posebej o izboljšanju samih emailov za zapuščene košarice. Če je vaša težava, da se tok sploh ne sproži dobro, je to težava nastavitve in vas bom napotil tja.

Težava: vaš tok košarice deluje, a emaili so postali zatohli

Skoraj vsaka trgovina prek 20.000 € ima email za zapuščeno košarico. Nekdo ga je nastavil, povrnil je nekaj košaric in se nato nikoli več ni spremenil. Dve leti pozneje še vedno pošilja »Nekaj ste pozabili!« s splošno zadevo in steno besedila, ki ga na telefonu nihče ne bere.

Rezultat je tok, ki na papirju deluje, v praksi pa zaostaja. Stopnja povračanja polagoma pada. Odprtja so v redu, ker je sprožilec močan — ljudje se košarice res spomnijo — a kliki in dejanska naročila zaostajajo, ker je sporočilo nehalo delovati relevantno. Prav ta razkorak med odprtji in naročili je znak, da je treba delati na vsebini, ne na sprožilcu.

Zakaj običajna rešitev ni dovolj

Refleksni popravek je dodati večji popust. Email za košarico dvignete z 10 % na 15 %, nato na 20 %, ko to ugasne. To povrne nekaj več košaric in tiho nauči vaše kupce, da košarico namerno zapuščajo, ker so se naučili, da kuponu vedno sledi. Maržo plačujete za povračanje ljudi, ki bi se vrnili že zaradi samega opomnika.

Drugi refleks je dodati več emailov v zaporedje. Trije postanejo pet. To večinoma dvigne odjave. Težava običajno ni količina — težava je, da vsak email pove isto plehko stvar. Petkratno pošiljanje istega šibkega sporočila ga ne naredi močnejšega.

Kar povračanje dejansko premakne, je boljše sporočilo: zadeva, ki si prisluži odprtje, prva vrstica, ki poimenuje resnično obotavljanje, in en jasen poziv k dejanju. Izdelovanje takšnih različic na roko je počasno, zato jih večina trgovin nikoli ne testira. To je konkretna naloga za AI.

Kje pušča denar

Puščanje v zatohlem toku košarice je prostor med »odprto« in »naročeno«. Kratek ponazoritveni sprehod: recimo, da v enem mesecu 1.000 ljudi zapusti košarico, vaš email za košarico doseže 40-odstotno stopnjo odprtja (400 odprtij), a le 8 % zapuščevalcev se povrne (80 naročil). Pri povprečnem naročilu 70 € je to 5.600 € povrnjenega. Dvignite povračanje z 8 % na 11 % — realen premik z boljšim besedilom in testiranjem zadev — in ste pri 110 naročilih, približno 7.700 €. To je okoli 2.100 € na mesec iz prepisovanja emailov, ki jih že pošiljate, brez dodatnih stroškov za oglase. (Ponazoritvene številke — uporabite svoje.)

Denar ne pušča zato, ker ljudje ne bi videli emaila. Pušča, ker email, ki ga vidijo, ne odgovori na majhen dvom, zaradi katerega so se ustavili.

Praktična rešitev, korak za korakom

AI dajte resničen kontekst, ne le »prepiši to«. Prilepite svoje trenutne emaile za košarico in dodajte: kaj prodajate, povprečno vrednost naročila ter dva ali tri razloge, zaradi katerih ljudje običajno obotavljajo (strošek dostave, dvom o velikosti, primerjava cen, »bom kasneje«). Besedilo, napisano proti resničnemu ugovoru, prekaša besedilo, napisano proti predlogi.
Najprej prepišite zadeve — ena vrstica opravi največ dela. Prosite za osem do deset zadev v različnih registrih: preprost opomnik, blago vprašanje, ena, ki poimenuje izdelek, ena o zalogi ali shranjeni košarici. Nato sami izrežite vsiljive različice z VELIKIMI TISKANIMI. Poglobljena tehnika za prav ta del živi v uporaba AI za pisanje boljših zadev email sporočil spletne trgovine.
Telo zbijte na eno nalogo. Email za košarico naj jih spomni, kaj so pustili, to jasno prikaže, odstrani en ugovor in ponudi en gumb. AI prosite, naj vaše besedilo skrči na to in napiše različico, ki vodi s sliko in imenom izdelka namesto s pozdravom.
Napišite različice, prilagojene ugovorom. Tu AI zablesti. Prosite za eno različico, ki naslovi strošek dostave, eno, ki pomirja glede vračil, eno, ki doda blago nujnost okoli zaloge — da lahko sčasoma sporočilo ujameš z verjetnim razlogom. Spreminjajte sporočilo, ne le besed.
Popust držite zunaj prvega emaila. Naj prvi email povrne z relevantnostjo. Kakršno koli spodbudo prihranite za poznejši korak in le, če to opravičujejo številke. Tako ima vaše zaporedje kam stopnjevati in preneha plačevati kupcem, da storijo, kar bi storili zastonj.
Pregled pred odpremo. AI bo občasno izmislila podrobnost — obljubo o dostavi, rok za vračilo, trditev o zalogi — ki za vašo trgovino ni resnična. Vsako vrstico preberite ob svoji resnični politiki. Kontrolni seznam za to si zasluži svoje branje: kako pregledati email besedilo, ki ga ustvari AI.

Kaj avtomatizirati — in kaj testirati

Izboljšani emaili še vedno sedijo znotraj vedenjskega toka, zato je struktura enako pomembna kot besede:

Sprožilec: košarica zapuščena, prvi email ven v približno eni uri, dokler je namera topla.
Segment: izključite vsakogar, ki je zaključil nakup; znane ponovne kupce obravnavajte drugače kot prve, če lahko.
Časovnica / kanal: email ob ~1 uri, drugi ob ~24 urah, izbirni SMS pošuk, če imate privolitev.
Vsebina: dejanski zapuščeni izdelki, dinamično — nikoli splošna mreža uspešnic.
Cilj: povračanje, merjeno kot naročila iz toka, ne odprtja.

Prepisovanje in logika toka sta dve polovici istega stroja, njuno usklajeno delovanje pa je svoja tema — glejte kako združiti ustvarjanje vsebine z AI z vedenjsko avtomatizacijo. In ker imate zdaj več različic besedila, jih testirajte pravilno, namesto da ugibate: kako z umetno inteligenco ustvariti različice emailov za A/B testiranje pokrije, kako to storiti brez samozavajanja s premajhnim vzorcem.

Primer iz trgovine

Trgovina z dodatki je osemnajst mesecev vodila isti dvoemailni tok za košarico: povračanje se je ustalilo okoli 7 %. Emaili so se odpirali v redu; skoraj nihče ni kliknil. Dovajanje emailov in ene vrstice konteksta — »kupci se obotavljajo pri 6 € dostave pod 50 €« — AI orodju je izdelalo prepis, ki je vodil s shranjenim izdelkom in z zadevo, ki je kot pošuk poimenovala brezplačno dostavo nad 50 €. Nekaj tednov so novo zadevo testirali proti stari. Kliki so si opomogli, ker je email končno nagovoril tisto, kar je ljudi dejansko zaustavilo. (Sestavljen primer — ne konkreten poročan rezultat.) Sprememba ni bil popust. Bilo je poimenovanje resničnega ugovora v prvi vrstici.

Kako izmeriti, ali je prepis deloval

Stopnja povračanja — naročila iz toka, deljena z zapuščenimi košaricami. Edina številka, ki šteje.
Stopnja klikov — ocenjuje vaše novo besedilo in zadevo; če so bila odprtja že v redu, se mora premakniti prav to.
Prihodek na prejemnika na toku, prej in potem.
Stopnja odjav, da vas ostrejše zaporedje ne stane seznama.

Opazujte razkorak med odprtji in kliki. Če odprtja ostanejo stabilna, kliki pa narastejo, je prepis opravil svoje. Če poskočijo sama odprtja, se je delo na zadevi obrestovalo.

Kje se vklopi Omnisend

Prepisovanje je le polovica zmage; emaili morajo potegniti resnične zapuščene izdelke, se sprožiti ob pravem zamiku, izstopiti ob nakupu in omogočiti deljeno testiranje zadeve brez razvijalca. V svojih trgovinah uporabljam Omnisend — izbral sem ga pred Klaviyem, potem ko sem testiral oba — ker se tok košarice odpremi kot vnaprej zgrajena avtomatizacija z dinamično vsebino košarice, postavitev A/B testa na zadevi pa je nekaj klikov in ne projekt. AI napiše različice; platforma izvede test in vam pokaže, katera je zmagala.

Pošteni omejitvi: AI lahko dopolni besedilo, a ne more povrniti košarice, ki jo je kupec zapustil z dobrim razlogom — resnično visok strošek dostave ali razprodana velikost sta težava trgovine, ne besedila. In orodje, ki tok sproži brezhibno, še vedno ne more rešiti emaila, ki ne pove nič vrednega klika. Omnisend je partner Shopimationa v pridruženem programu; priporočam ga iz vsakodnevne rabe, brezplačna raven pa zadošča, da svoj tok košarice znova zgradite in preizkusite na živem prometu.

Vaš naslednji korak

Danes izvlecite svoje trenutne emaile za zapuščeno košarico, AI orodju dajte eno vrstico o tem, zakaj ljudje obotavljajo, in prosite za deset zadev ter eno bolj zbito telo, ki vodi z izdelkom. Najboljšo novo zadevo testirajte proti trenutni. Če celoten tok košarice potrebuje več kot le osvežitev besedila — ali če je to ena postavka na daljšem seznamu — začnite z kako z umetno inteligenco poiskati priložnosti za prihodek v email podatkih, da potrdite, da je tok košarice res vaše največje puščanje, in uporabite kako z umetno inteligenco prepisati email sporočila spletne trgovine, ki slabo delujejo za isto tehniko čez druge tokove.

AI for Ecommerce Marketing

How to Use AI to Improve Abandoned Cart Emails

Posted on July 18, 2026 by

If your abandoned cart emails are already running but recovery has gone flat, AI is most useful as a rewriting and testing partner, not a magic recovery machine. Feed it your current cart emails plus a little context — what you sell, your average order value, what objection usually stops a purchase — and ask it to rewrite the subject lines, tighten the copy, and propose variations you can test. It’s fast at the part that bores you: producing five honest angles instead of the one you’ve been sending for a year. What it can’t do is fix a cart flow that’s broken structurally — wrong timing, no reminder of the actual product, a discount handed out too early. This article is specifically about improving the abandoned cart emails themselves. If your problem is that the flow doesn’t fire well at all, that’s a setup issue, and I’ll point you there.

The problem: your cart flow works, but the emails have gone stale

Almost every store past €20k has an abandoned cart email. Someone set it up, it recovered some carts, and then it never changed. Two years later it’s still sending “You left something behind!” with a generic subject line and a wall of copy nobody reads on a phone.

The result is a flow that works on paper and underperforms in practice. Recovery rate drifts down. Opens are okay because the trigger is strong — people do remember the cart — but clicks and actual orders lag, because the message stopped feeling relevant. That gap between opens and orders is the tell that the content, not the trigger, is what needs work.

Why the usual fix isn’t enough

The reflex fix is to add a bigger discount. Bump the cart email from 10% to 15%, then to 20% when that fades. This recovers a few more carts and quietly trains your customers to abandon on purpose, because they’ve learned a coupon always follows. You’re paying margin to recover people who’d have come back for a reminder alone.

The other reflex is to add more emails to the sequence. Three becomes five. That mostly raises unsubscribes. The problem usually isn’t quantity — it’s that each email says the same flat thing. Sending the same weak message five times doesn’t make it stronger.

What actually moves recovery is a better message: a subject line that earns the open, a first line that names the real hesitation, and a clear single call to action. Producing those variations by hand is slow, which is why most stores never test them. That’s the specific job to hand to AI.

Where the money leaks

The leak in a stale cart flow is the space between “opened” and “ordered.” A quick illustrative walk-through: say 1,000 people abandon a cart in a month, your cart email gets a 40% open rate (400 opens), but only 8% of abandoners recover (80 orders). At a €70 average order that’s €5,600 recovered. Lift recovery from 8% to 11% — a realistic swing from better copy and subject-line testing — and you’re at 110 orders, roughly €7,700. That’s about €2,100 a month from rewriting emails you already send, no extra ad spend. (Illustrative numbers — use your own.)

The money doesn’t leak because people don’t see the email. It leaks because the email they see doesn’t answer the small doubt that made them stop.

The practical solution, step by step

Give the AI real context, not just “rewrite this.” Paste your current cart emails and add: what you sell, average order value, and the two or three reasons people usually hesitate (shipping cost, sizing doubt, price comparison, “I’ll do it later”). Copy written against a real objection beats copy written against a template.
Rewrite subject lines first — one line does the most work. Ask for eight to ten subject lines in different registers: a plain reminder, a mild question, one that names the product, one about stock or a saved cart. Then cut the pushy ALL-CAPS ones yourself. Deeper technique for this specific piece lives in using AI to write better ecommerce email subject lines.
Tighten the body to one job. A cart email should remind them what they left, show it clearly, remove one objection, and give one button. Ask the AI to cut your copy to that and to write a version that leads with the product image and name rather than a greeting.
Write objection-specific variants. This is where AI shines. Ask for one version that addresses shipping cost, one that reassures on returns, one that adds light urgency around stock — so you can eventually match the message to the likely reason. Vary the message, not just the wording.
Keep the discount out of email one. Let the first email recover on relevance. Hold any incentive for a later step, and only if the numbers justify spending the margin. This keeps your sequence somewhere to escalate and stops you paying customers to do what they’d do free.
Review before it ships. AI will occasionally invent a detail — a shipping promise, a return window, a stock claim — that isn’t true for your store. Read every line against your real policy. The checklist for this is worth its own read: how to review AI-generated ecommerce email copy.

What to automate — and what to test

The improved emails still sit inside a behavioral flow, so the structure matters as much as the words:

Trigger: cart abandoned, first email out within an hour or so while intent is warm.
Segment: exclude anyone who completed checkout; treat known repeat customers differently from first-timers if you can.
Timing / channel: email at ~1 hour, a second at ~24 hours, an optional SMS nudge if you have consent.
Content: the actual abandoned items, dynamically — never a generic bestseller grid.
Goal: recovery, measured as orders from the flow, not opens.

The rewriting and the flow logic are two halves of the same machine, and getting them to work together is its own topic — see how to combine AI content creation with behavioral automation. And because you now have several copy variants, test them properly rather than guessing: how to use AI to create email variations for A/B testing covers doing that without fooling yourself with a too-small sample.

A store example

An accessories store had run the same two-email cart flow for eighteen months: recovery had settled around 7%. The emails opened fine; almost nobody clicked. Feeding the emails and one line of context — “customers hesitate on €6 shipping under €50” — to an AI tool produced a rewrite that led with the saved item and a subject naming free shipping over €50 as the nudge. They tested the new subject against the old one for a few weeks. Clicks recovered because the email finally spoke to the thing that had actually stopped people. (Composite example — not a specific reported result.) The change wasn’t a discount. It was naming the real objection in the first line.

How to measure whether the rewrite worked

Recovery rate — orders from the flow divided by carts abandoned. The one number that matters.
Click rate — judges your new copy and subject; if opens were already fine, this is what should move.
Revenue per recipient on the flow, before and after.
Unsubscribe rate, so a punchier sequence doesn’t cost you the list.

Watch the opens-versus-clicks split. If opens stay steady but clicks rise, the rewrite did its job. If opens themselves jump, the subject-line work paid off.

Where Omnisend fits

Rewriting is only half the win; the emails need to pull the real abandoned products, fire on the right delay, exit on purchase, and let you split-test a subject line without a developer. I use Omnisend in my own stores — I picked it over Klaviyo after testing both — because the cart flow ships as a prebuilt automation with dynamic cart contents, and setting up an A/B test on the subject line is a couple of clicks rather than a project. The AI writes the variants; the platform runs the test and shows you which won.

Honest limits: AI can polish the wording, but it can’t recover a cart the customer abandoned for a good reason — a genuinely high shipping cost or an out-of-stock size is a store problem, not a copy problem. And a tool that fires the flow flawlessly still can’t save an email that says nothing worth clicking. Omnisend is an affiliate partner of Shopimation; I recommend it from daily use, and the free tier is enough to rebuild and test your cart flow on live traffic.

Your next step

Pull your current abandoned cart emails today, give an AI tool one line about why people hesitate, and ask for ten subject lines and one tighter body that leads with the product. Test the best new subject against your current one. If the whole cart flow needs more than a copy refresh — or if this is one item on a longer list — start from using AI to find revenue opportunities in email data to confirm the cart flow is really where your biggest leak is, and use how to use AI to rewrite underperforming ecommerce emails for the same technique across your other flows.