Sezonski in neredni nakupni cikli

Kako predvideti, kdaj je sezonski kupec pripravljen na nov nakup

Posted on July 18, 2026 by

Kdaj je sezonski kupec pripravljen na nov nakup, predvidite tako, da združite dva signala: njegov lastni pretekli čas (mesec ali okno, v katerem je kupoval prej) in svežo vedênje prav zdaj (odprtja, kliki, obiski strani, preverjanje razpoložljivosti na zalogi). Pretekli čas vam pove približno kdaj začeti biti pozoren. Trenutno vedenje vam pove, da se prav ta oseba ogreva pred množico. Preprosto povedano: z datumom lanskega nakupa nastavite okno, nato znotraj tega okna prežite na aktivnost, da sporočilo sprožite pri posamezniku in ne pri koledarskem povprečju. Ravno pravilo “90 dni od zadnjega naročila” tega ne zmore — ignorira oba signala. Sledi, kako zgraditi napoved, ki je dovolj konkretna za ukrepanje, z uporabo le podatkov, ki jih že imate.

Dva signala, ki dejansko napovesta sezonski ponovni nakup

Večina nasvetov o “napovedovanju ponovnega nakupa” poseže po strojnem učenju ali po nepregledni oceni nagnjenosti. Za sezonsko trgovino tega redko potrebujete. Dva običajna signala, združena, vas pripeljeta večino poti.

Signal ena: njihovo pretekli nakupno okno. Če je nekdo lani decembra kupil smučarske rokavice, osnovna stopnja pravi, da bo najverjetneje znova kupil okoli naslednjega novembra ali decembra. To je sidro vaše napovedi — okno, ne dan. Kdorkoli je kupil v danem mesecu, je del vaše skupine “verjetno se bo takrat vrnil”.

Signal dva: trenutna aktivnost znotraj tega okna. Ko se okno približuje, se nekateri začnejo pojavljati zgodaj — odprejo predsezonsko e-pošto, kliknejo skoznjo, obiščejo stran, preverijo, ali je lanski izdelek spet na zalogi. To vedenje je znamenje, ki loči “verjetno bo kupil čez šest tednov” od “zdaj gleda”.

Noben signal sam ni dovolj. Pretekli čas brez trenutnega vedenja se sproži pri povprečju in zgreši zgodnje ptice in zamudnike. Trenutno vedenje brez sezonskega okna vas utopi v šumu — poletni obisk strani zimske trgovine morda ne pomeni nič. Skupaj sta resnično napovedna. Pravi pretekli del je odvisen od čistih podatkov o mesecu nakupa: kako segmentirati stranke po mesecu, v katerem običajno kupujejo.

Problem: vaše orodje ugiba po napačni uri

Tale je položaj. Vaša e-poštna platforma privzeto razmišlja v času od zadnjega nakupa. Ve, koliko dni je od zadnjega naročila, in to uporablja, da odloči, kdo je “aktiven”, kdo “odhaja” in kdaj spodbuditi. Za trgovino, ki prodaja nekaj, kar ljudje kupujejo vsakih nekaj tednov, je ta ura v redu.

Za sezonsko trgovino je dejavno zavajajoča. Dnevi od nakupa za decembrskega kupca vztrajno naraščajo vse leto in dosežejo vrhunec novembra — ravno v trenutku, ko je najverjetneje, da bo znova kupil. Ura časa od zadnjega nakupa pravi “najbolj izgubljen” natanko takrat, ko je resnica “najbolj pripravljen”. Torej če pustite privzeti logiki, da poganja vašo časovno razporeditev, boste ljudi najbolj preganjali v izvensezoni, ko so hladni, in jih obravnavali kot izgubljene tik preden bi pretvorili.

Podatkov vam ne primanjkuje. Imate datume naročil, zgodovino izdelkov, aktivnost v e-pošti in vedenje na strani. Težava je v tem, da standardni stroj te podatke bere skozi napačno lečo. To je isti razlog, zakaj se splošna pravila o nedejavnosti razsujejo pri sezonskih izdelkih: zakaj pravilo o 90-dnevni nedejavnosti odpove pri sezonskih izdelkih.

Zakaj “počakaj, da se vrnejo” in “kar pošlji vsem” odpovesta

Sezonske lastnike mikata dve pasivni strategiji in obe puščata denar na mizi.

Počakaj, da se stranka vrne sama. Nekatere se bodo. A časovno razporeditev boste prepustili naključju, tekmovali med vrhuncem sezone poleg vseh drugih v svoji kategoriji in povsem zgrešili zgodnje kupce — tiste, ki bi ob dobro časovanem opomniku pretvorili tedne prej, preden bi jih dosegli konkurenti.

Pošlji celemu seznamu, takoj ko se sezona približa. To povsem ignorira napoved in le razpršuje. Deluje malo, ker so bili nekateri prejemniki pripravljeni. A za večino ga časovno zgrešite, razdražite tiste, ki kupujejo kasneje v sezoni, in zapravite aktivnost (in dostavljivost) z razpošiljanjem ljudem, ki še niso topli. Napoved obstaja prav zato, da lahko pošljete pravo sporočilo pravemu delu ob pravem trenutku, namesto enega sporočila vsem hkrati.

Kje vas nenatančnost stane

Vrzel med “približno ko se sezona začne” in “kdaj je ta oseba dejansko pripravljena” je mesto, kjer izteka prihodek.

Zgodnji kupci, doseženi pozno. Stranke, ki bi kupile šest tednov pred vrhuncem, če bi jih spodbudili, namesto tega kupijo med vrhuncem — če jih konkurent ni dosegel prvi. Naročilo ste ujeli, a v natrpanem, dragem trenutku, ali pa ste ga izgubili.
Pozni kupci, razdraženi zgodaj. Ljudje, katerih resnično okno je pozno v sezoni, prejmejo vaš velik naval prezgodaj, se ohladijo in so hladnejši, ko njihov trenutek dejansko pride.
Resnični odhodi, napačno prebrani. Stranke, ki se resnično ne bodo vrnile, so videti enake kot pravočasne stranke, če gledate le čas od zadnjega nakupa — torej bodisi preveč vlagate v reaktivacijo ljudi, ki bi se tako ali tako vrnili, bodisi odpišete ljudi, ki bi jih lahko rešili.

Groba ponazoritev. Recimo, da je lansko sezono kupilo 1.000 strank in 300 od njih bi se odzvalo na zgoden, dobro časovan sunek dva do štiri tedne pred svojim osebnim oknom. Dosezite teh 300 ob pravem trenutku in približno 70 € × 300 = 21.000 € povpraševanja potegnete naprej, pred gnečo vrhunca sezone, pri polni marži. Zgrešite čas in se za ista naročila borite kasneje, po višjem strošku, z nižjo gotovostjo. (Ponazoritveno — vaša velikost seznama, vrednost naročila in stopnja odziva bodo drugačne.)

Praktična metoda: najprej okno, nato vedenje

Podatkovne znanosti ne potrebujete. Potrebujete dvoslojno pravilo, ki ga lahko zgradite v običajni e-poštni platformi.

Sloj 1 — Nastavite napovedno okno iz zgodovine

Za vsako stranko vzemite njen mesec zadnjega nakupa (in izdelek) ter odprite okno “verjetno se bo vrnil”, ki se začne šest do osem tednov pred tem mesecem in teče skozenj. Okno decembrskega kupca se odpre pozno oktobra. Tu živi večina vaše napovedi in izhaja naravnost iz oznake datuma nakupa, ki jo (bi morali) že imeti. Uporaba preteklega časa nakupa za načrtovanje avtomatizacije za celotno sezono je globlja tema tukaj: uporaba preteklega časa nakupa za načrtovanje avtomatizacije naslednje sezone.

Sloj 2 — Prežite na zgodnje vedenje znotraj okna

Ko se okno stranke odpre, začnite točkovati trenutne signale:

Odprla ali kliknila predsezonsko e-pošto
Obiskala stran ali ustrezno kategorijsko podstran
Preverila ali se naročila na obvestilo o razpoložljivosti lanskega izdelka
Kliknila predogled “kaj je novega to sezono”

Katerikoli od teh, znotraj okna, povzdigne stranko iz “pričakovano se vrne” v “zdaj se ogreva” — in to je vaš sprožilni signal za bolj neposredno, osebno sporočilo.

Sloj 3 — S skrajnimi primeri ravnajte pošteno

Nobenega vedenja do sredine okna? Vseeno pošljite standardno predsezonsko zaporedje; okno samo je še vedno spodobna napoved.
Vedenje, preden se okno odpre? Obravnavajte ga kot dodaten zgodnji signal, a ne pretiravajte zaradi enega samega poletnega ogleda strani.
Okno je minilo brez nakupa in brez aktivnosti? Zdaj je tišina pomenljiva — to je resničen odhod in trenutek, ko reaktivacijska logika končno dobi smisel.

Kaj natančno avtomatizirati

Sprožilec (odprtje okna): na datum vezan, izračunan kot mesec zadnjega nakupa minus približno šest do osem tednov, na segment.
Segment: kupci, označeni po mesecu nakupa in izdelku.
Vedenjska stopnjevanja: znotraj odprtega okna aktivnostni dogodek (klik e-pošte, obisk strani, preverjanje razpoložljivosti) premakne stranko v vejo z višjim namenom.
Časovna razporeditev: sporočilo ob odprtju okna je mehko (“sezona prihaja”); sporočilo, sproženo z vedenjem, je neposredno in osebno (“pogledali ste — tukaj je, kar je novega v tem, kar ste kupili lani”), poslano v dnevu ali dveh po signalu, dokler je namen svež.
Kanal: e-pošta za sunek ob odprtju okna; SMS-opomnik se splača za stranke, ki so privolile in pokazale trenutni namen blizu vrhunca.
Izhod: ob nakupu zapustijo napovedno pot in preidejo na po-nakupno.
Cilj: pretvoriti zgodaj in po polni ceni, pred natrpanim vrhuncem, ob tem pa kupce s poznim oknom pustiti pri miru do njihovega trenutka.

Primer iz trgovine

Trgovina, ki prodaja smučarske rokavice, ima stranke razpršene čez zimo — nekateri kupijo ob prvem mrazu novembra, nekateri počakajo na praznike, nekateri zgrabijo par januarja pred potovanjem.

Pod pravilom časa od zadnjega nakupa so bili vsi videti “najbolj izgubljeni” jeseni in dobili enak naval. Predelano z napovedjo: okno vsakega kupca se odpre šest tednov pred njegovim lanskim mesecem nakupa. Stranka, ki je kupila novembra, dobi mehko e-pošto “zima prihaja” pozno septembra; ko jo klikne in v začetku oktobra obišče stran, to vedenje v dnevu sproži neposredno, personalizirano sporočilo — “lani ste vzeli X; tukaj je letošnja različica in kar se z njo poda” — in oktobra kupi, precej pred vrhuncem. Januarski kupec pa ni potisnjen do decembra, tako da ni razdražen do tega, da bi se zgodaj ohladil. Isti seznam, časovno razporejen na osebo, več polnocenovskih naročil, smiselno razporejenih čez sezono. (Ponazoritveni primer — vaša sezonska razpršitev in izdelek bosta drugačna.)

Kaj meriti

Časovni odmik naročil — kako daleč pred tednom vrhunca kupijo vaše napovedane in spodbujene stranke. Naraščajoč odmik pomeni, da napoved deluje.
Stopnja pretvorbe sporočil, sproženih z vedenjem v primerjavi z navadnim sporočilom ob odprtju okna — vrzel pokaže, koliko doda sloj trenutnega vedenja.
Delež polnocenovskih naročil v sezoni — napoved naj vam omogoči, da pretvorite zgodaj brez opiranja na popuste.
Stopnja lažnih odhodov — koliko strank bi odpisali po času od zadnjega nakupa, a so dejansko kupile, ko je prišlo njihovo pravo okno. Opazujte, kako pada.
Prihodek na prejemnika skozi napovedno okno — povzemna številka.

Kako pomaga Omnisend

Napoved je tu pravzaprav dvoje mehanike: računanje datumov na zgodovini nakupov, da se okno odpre, in vedenjski sprožilci, da se ujame trenutni namen znotraj njega. V svojih trgovinah uporabljam Omnisend — prešel sem nanj po preizkusu Klaviya, predvsem zaradi bolj samopostrežnega graditelja avtomatizacij in e-pošte, SMS in potisnih sporočil skupaj po boljši ceni. Za to nalogo so pomembni kosi na datum vezani sprožilci, izračunani iz zadnjega nakupa stranke, vedenjski sprožilci na klike e-pošte ter aktivnost na strani ali izdelku in segmenti, ki oboje združijo, tako da “v oknu in aktiven” postane resnično občinstvo, ki mu lahko pišete. Brezplačna različica zadošča, da dvoslojno logiko zgradite in preizkusite na lanskih kupcih, preden jo razširite. Omnisend je partner Shopimationa v pridruženem programu in ga priporočam iz dnevne uporabe — s poštenim pridržkom, da nobeno orodje ne napove namena iz redkih podatkov. Če nimate čistih datumov nakupov, to najprej popravite; napoved je le tako dobra kot zgodovina za njo.

Vaš naslednji korak

Preverite, ali ima vsaka pretekla stranka na svojem zapisu uporaben mesec zadnjega nakupa in izdelek — če ne, je to prva stvar, ki jo je treba popraviti, preden karkoli od tega deluje. Nato nastavite svoja napovedna okna tako, da se odprejo šest do osem tednov pred mesecem nakupa vsakega segmenta, in nanje naložite vedenjski sprožilec za zgodnje ptice. Ko lahko prepoznate, kdo se ogreva, postane vprašanje, kaj povedati, medtem ko čakate, da se okno odpre: kaj poslati med dvema časovno oddaljenima nakupnima sezonama. Da razumete, zakaj si lanski kupci sploh zaslužijo to pozornost, zakaj so lanski kupci vaše najboljše občinstvo za naslednjo sezono poda argument.

Seasonal and Irregular Purchase Cycles

How to Predict When a Seasonal Customer Is Ready to Buy Again

Posted on July 18, 2026 by

You predict when a seasonal customer is ready to buy again by combining two signals: their own past timing (the month or window they’ve bought in before) and fresh behavior right now (opens, clicks, site visits, a back-in-stock check). Past timing tells you roughly when to start paying attention. Present behavior tells you this specific person is warming up ahead of the pack. Put simply: use last year’s purchase date to set the window, then watch for engagement inside that window to fire the message at the individual, not the calendar average. A flat “90 days since last order” rule can’t do this — it ignores both signals. What follows is how to build a prediction that’s specific enough to act on, using only data you already have.

The two signals that actually predict a seasonal repurchase

Most “predict repurchase” advice reaches for machine learning or a black-box propensity score. For a seasonal store, you rarely need that. Two ordinary signals, combined, get you most of the way.

Signal one: their historical buying window. If someone bought ski gloves last December, the base rate says they’re most likely to buy again around next November or December. That’s your prediction’s anchor — a window, not a day. Anyone who bought in a given month is your set of “likely to return around then.”

Signal two: present-tense engagement inside that window. As the window approaches, some people start showing up early — they open the pre-season email, click through, visit the site, check whether last year’s item is back in stock. That behavior is the tell that separates “will probably buy in six weeks” from “is looking right now.”

Neither signal alone is enough. Historical timing without live behavior fires at the average and misses the early birds and the stragglers. Live behavior without the seasonal window drowns you in noise — a summer site visit to a winter store might mean nothing. Together they’re genuinely predictive. Getting the historical half right depends on clean purchase-month data: how to segment customers by the month they usually purchase.

The problem: your tool is guessing on the wrong clock

Here’s the situation. Your email platform, out of the box, thinks in recency. It knows how many days since someone’s last order, and it uses that to decide who’s “active,” who’s “lapsing,” and when to nudge. For a store selling something people buy every few weeks, that clock is fine.

For a seasonal store it’s actively misleading. Days-since-purchase for a December buyer climbs steadily all year and hits its maximum in November — the exact moment they’re most likely to buy again. The recency clock says “most lapsed” precisely when the truth is “most ready.” So if you let the default logic drive your timing, you’ll chase people hardest in the off-season, when they’re cold, and treat them as lost right before they were about to convert.

You’re not short on data. You have order dates, product history, email engagement, and site behavior. The problem is that the standard machinery reads that data through the wrong lens. This is the same reason blanket inactivity rules fall apart for seasonal products: why a 90-day inactivity rule fails for seasonal products.

Why “wait for them to come back” and “just email everyone” both fail

Two passive strategies tempt seasonal owners, and both leave money on the table.

Wait for the customer to return on their own. Some will. But you’ll have surrendered the timing to chance, competed at peak week alongside everyone else in your category, and missed the early buyers entirely — the ones who’d have converted weeks sooner on a well-timed reminder, before your competitors reached them.

Email the whole list the moment the season nears. This ignores prediction altogether and just sprays. It works a little, because some recipients were ready. But you mistime it for most, annoy the ones who buy later in the season, and burn engagement (and deliverability) blasting people who aren’t warm yet. Prediction exists precisely so you can send the right message to the right slice at the right moment, instead of one message to everyone at once.

Where the imprecision costs you

The gap between “roughly when the season starts” and “when this person is actually ready” is where revenue leaks.

Early buyers, reached late. The customers who’d have bought six weeks before peak, had you nudged them, instead buy during peak — if a competitor didn’t reach them first. You captured the order but at the crowded, expensive moment, or you lost it.
Late buyers, annoyed early. People whose real window is late in the season get hit with your big push too soon, disengage, and are colder when their moment actually comes.
True lapses, misread. Customers who genuinely won’t return look identical to on-schedule customers if you only watch recency — so you either over-invest in reactivating people who’d have come back anyway, or write off people you could have saved.

A rough illustration. Say 1,000 customers bought last season, and 300 of them would respond to an early, well-timed nudge two to four weeks before their personal window. Reach those 300 at the right moment and you pull roughly €70 × 300 = €21,000 of demand forward, ahead of the peak-week scrum, at full margin. Miss the timing and you fight for the same orders later, at higher cost, with lower certainty. (Illustrative — your list size, order value, and response rate will differ.)

The practical method: window first, then behavior

You don’t need data science. You need a two-layer rule you can build in an ordinary email platform.

Layer 1 — Set the prediction window from history

For each customer, take their last purchase month (and product) and open a “likely to return” window that starts six to eight weeks before that month and runs through it. A December buyer’s window opens in late October. This is where most of your prediction lives, and it comes straight from the purchase-date tag you (should) already have. Using past timing to plan the whole season’s automation is a deeper topic here: using past purchase timing to plan next season’s automation.

Layer 2 — Watch for early behavior inside the window

Once a customer’s window opens, start scoring live signals:

Opened or clicked a pre-season email
Visited the site or a relevant category page
Checked or subscribed to back-in-stock on last year’s item
Clicked a “what’s new this season” preview

Any of these, inside the window, promotes the customer from “expected to return” to “warming up now” — and that’s your fire signal for a more direct, personal message.

Layer 3 — Handle the edges honestly

No behavior by mid-window? Send the standard pre-season sequence anyway; the window itself is still a decent prediction.
Behavior before the window opens? Treat it as a bonus early signal, but don’t overreact to a single summer page view.
Window passed with no purchase and no engagement? Now the silence is meaningful — this is a genuine lapse, and the moment reactivation logic finally makes sense.

What to automate, precisely

Trigger (window open): date-based, calculated as last-purchase-month minus roughly six to eight weeks, per segment.
Segment: buyers tagged by purchase month and product.
Behavioral escalation: within the open window, an engagement event (email click, site visit, back-in-stock check) moves the customer into a higher-intent branch.
Timing: window-open message is soft (“the season’s coming”); the behavior-triggered message is direct and personal (“you looked — here’s what’s new in what you bought last year”), sent within a day or two of the signal while intent is fresh.
Channel: email for the window-open nudge; an SMS reminder is worth it for opted-in customers who’ve shown live intent near peak.
Exit: on purchase, they leave the prediction track and move to post-purchase.
Goal: convert early and at full price, ahead of the crowded peak, while leaving late-window buyers undisturbed until their moment.

A store example

A store selling ski gloves has customers spread across the winter — some buy at the first cold snap in November, some wait for the holidays, some grab a pair in January before a trip.

Under a recency rule, all of them looked “most lapsed” in autumn and got the same blast. Reworked with prediction: each buyer’s window opens six weeks before their own last-year purchase month. A customer who bought in November gets a soft “winter’s coming” email in late September; when she clicks it and visits the site in early October, that behavior fires a direct, personalized message within a day — “you picked up the X last year; here’s this season’s version and what pairs with it” — and she buys in October, well before the peak. The January buyer, meanwhile, isn’t pushed until December, so he’s not annoyed into disengaging early. Same list, timed per person, more full-price orders spread sensibly across the season. (Illustrative example — your seasonal spread and product will differ.)

What to measure

Lead time on orders — how far ahead of peak week your predicted-and-nudged customers buy. Growing lead time means the prediction is working.
Conversion rate of behavior-triggered messages vs. the plain window-open message — the gap shows how much the live-behavior layer adds.
Full-price share of in-season orders — prediction should let you convert early without leaning on discounts.
False-lapse rate — how many customers you’d have written off on recency who actually bought once their real window came. Watch it fall.
Revenue per recipient across the prediction window — the summary number.

How Omnisend helps

Prediction here is really two mechanics: date math on purchase history to open the window, and behavioral triggers to catch live intent inside it. I run Omnisend in my own stores — I moved to it after testing Klaviyo, mainly for the more self-serve automation builder and email, SMS, and push together at a better price. For this job, the pieces that matter are date-based triggers computed from a customer’s last purchase, behavioral triggers on email clicks and site or product activity, and segments that combine both so “in-window and engaged” becomes a real audience you can message. The free tier is enough to build and test the two-layer logic on last season’s buyers before you scale it. Omnisend is an affiliate partner of Shopimation, and I recommend it from daily use — with the honest caveat that no tool predicts intent from thin data. If you don’t have clean purchase dates, fix that first; the prediction is only as good as the history behind it.

Your next step

Check whether every past customer has a usable last-purchase month and product on their record — if not, that’s the one thing to fix before any of this works. Then set your prediction windows to open six to eight weeks before each segment’s buying month, and layer a behavioral trigger on top for the early birds. Once you can spot who’s warming up, the question becomes what to say while you wait for the window to open: what to send between two distant buying seasons. And to understand why last season’s buyers deserve this attention in the first place, why last year’s buyers are your best audience for the next season makes the case.